2023-06-25 14:40:30 +02:00
|
|
|
#import "ggml-metal.h"
|
|
|
|
|
2023-11-03 20:35:05 +01:00
|
|
|
#import "ggml-backend-impl.h"
|
2023-06-25 14:40:30 +02:00
|
|
|
#import "ggml.h"
|
|
|
|
|
|
|
|
#import <Foundation/Foundation.h>
|
|
|
|
|
|
|
|
#import <Metal/Metal.h>
|
2023-09-05 12:54:40 +02:00
|
|
|
|
|
|
|
#undef MIN
|
|
|
|
#undef MAX
|
|
|
|
#define MIN(a, b) ((a) < (b) ? (a) : (b))
|
|
|
|
#define MAX(a, b) ((a) > (b) ? (a) : (b))
|
2023-06-25 14:40:30 +02:00
|
|
|
|
|
|
|
#ifdef GGML_METAL_NDEBUG
|
2023-11-03 20:35:05 +01:00
|
|
|
#define GGML_METAL_LOG_INFO(...)
|
|
|
|
#define GGML_METAL_LOG_WARN(...)
|
|
|
|
#define GGML_METAL_LOG_ERROR(...)
|
2023-06-25 14:40:30 +02:00
|
|
|
#else
|
2023-11-03 20:35:05 +01:00
|
|
|
#define GGML_METAL_LOG_INFO(...) ggml_metal_log(GGML_LOG_LEVEL_INFO, __VA_ARGS__)
|
|
|
|
#define GGML_METAL_LOG_WARN(...) ggml_metal_log(GGML_LOG_LEVEL_WARN, __VA_ARGS__)
|
|
|
|
#define GGML_METAL_LOG_ERROR(...) ggml_metal_log(GGML_LOG_LEVEL_ERROR, __VA_ARGS__)
|
2023-06-25 14:40:30 +02:00
|
|
|
#endif
|
|
|
|
|
|
|
|
#define UNUSED(x) (void)(x)
|
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
struct ggml_metal_kernel {
|
|
|
|
id<MTLComputePipelineState> pipeline;
|
|
|
|
};
|
|
|
|
|
|
|
|
enum ggml_metal_kernel_type {
|
|
|
|
GGML_METAL_KERNEL_TYPE_ADD,
|
|
|
|
GGML_METAL_KERNEL_TYPE_ADD_ROW,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_ROW,
|
|
|
|
GGML_METAL_KERNEL_TYPE_DIV,
|
|
|
|
GGML_METAL_KERNEL_TYPE_DIV_ROW,
|
|
|
|
GGML_METAL_KERNEL_TYPE_SCALE,
|
|
|
|
GGML_METAL_KERNEL_TYPE_SCALE_4,
|
2024-04-14 13:14:19 +02:00
|
|
|
GGML_METAL_KERNEL_TYPE_CLAMP,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_TANH,
|
|
|
|
GGML_METAL_KERNEL_TYPE_RELU,
|
2024-05-01 23:44:26 +02:00
|
|
|
GGML_METAL_KERNEL_TYPE_SIGMOID,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_GELU,
|
2024-04-16 17:40:48 +02:00
|
|
|
GGML_METAL_KERNEL_TYPE_GELU_4,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_GELU_QUICK,
|
2024-04-16 17:40:48 +02:00
|
|
|
GGML_METAL_KERNEL_TYPE_GELU_QUICK_4,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_SILU,
|
2024-04-16 17:40:48 +02:00
|
|
|
GGML_METAL_KERNEL_TYPE_SILU_4,
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
GGML_METAL_KERNEL_TYPE_SOFT_MAX_F16,
|
|
|
|
GGML_METAL_KERNEL_TYPE_SOFT_MAX_F16_4,
|
|
|
|
GGML_METAL_KERNEL_TYPE_SOFT_MAX_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_SOFT_MAX_F32_4,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_DIAG_MASK_INF,
|
|
|
|
GGML_METAL_KERNEL_TYPE_DIAG_MASK_INF_8,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_F16,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_Q4_0,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_Q4_1,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_Q5_0,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_Q5_1,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_Q8_0,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_Q2_K,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_Q3_K,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_Q4_K,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_Q5_K,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_Q6_K,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_XXS,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_XS,
|
2024-01-30 14:14:12 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ3_XXS,
|
2024-02-24 15:23:52 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ3_S,
|
2024-02-26 17:28:38 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_S,
|
2024-02-18 17:16:55 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ1_S,
|
2024-03-27 17:55:10 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ1_M,
|
2024-02-21 15:19:39 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_NL,
|
2024-02-27 15:34:24 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_XS,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_GET_ROWS_I32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_RMS_NORM,
|
|
|
|
GGML_METAL_KERNEL_TYPE_GROUP_NORM,
|
|
|
|
GGML_METAL_KERNEL_TYPE_NORM,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_F32_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F16,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F32_1ROW,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F32_L4,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_Q4_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_Q4_1_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_Q5_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_Q5_1_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_Q8_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_Q2_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_Q3_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_Q4_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_Q5_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_Q6_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_XXS_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_XS_F32,
|
2024-01-30 14:14:12 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ3_XXS_F32,
|
2024-02-24 15:23:52 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ3_S_F32,
|
2024-02-26 17:28:38 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_S_F32,
|
2024-02-18 17:16:55 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ1_S_F32,
|
2024-03-27 17:55:10 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ1_M_F32,
|
2024-02-21 15:19:39 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_NL_F32,
|
2024-02-27 15:34:24 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_XS_F32,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F32_F32,
|
|
|
|
//GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F16,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F32,
|
|
|
|
//GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F32_1ROW,
|
|
|
|
//GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F32_L4,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q4_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q4_1_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q5_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q5_1_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q8_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q2_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q3_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q4_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q5_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q6_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_XXS_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_XS_F32,
|
2024-01-30 14:14:12 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ3_XXS_F32,
|
2024-02-24 15:23:52 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ3_S_F32,
|
2024-02-26 17:28:38 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_S_F32,
|
2024-02-18 17:16:55 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ1_S_F32,
|
2024-03-27 17:55:10 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ1_M_F32,
|
2024-02-21 15:19:39 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_NL_F32,
|
2024-02-27 15:34:24 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_XS_F32,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_F32_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_F16_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_Q4_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_Q4_1_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_Q5_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_Q5_1_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_Q8_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_Q2_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_Q3_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_Q4_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_Q5_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_Q6_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_XXS_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_XS_F32,
|
2024-01-30 14:14:12 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_XXS_F32,
|
2024-02-24 15:23:52 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_S_F32,
|
2024-02-26 17:28:38 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_S_F32,
|
2024-02-18 17:16:55 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ1_S_F32,
|
2024-03-27 17:55:10 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ1_M_F32,
|
2024-02-21 15:19:39 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_NL_F32,
|
2024-02-27 15:34:24 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_XS_F32,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_F32_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_F16_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q4_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q4_1_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q5_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q5_1_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q8_0_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q2_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q3_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q4_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q5_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q6_K_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_XXS_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_XS_F32,
|
2024-01-30 14:14:12 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ3_XXS_F32,
|
2024-02-24 15:23:52 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ3_S_F32,
|
2024-02-26 17:28:38 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_S_F32,
|
2024-02-18 17:16:55 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ1_S_F32,
|
2024-03-27 17:55:10 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ1_M_F32,
|
2024-02-21 15:19:39 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_NL_F32,
|
2024-02-27 15:34:24 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_XS_F32,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_ROPE_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_ROPE_F16,
|
|
|
|
GGML_METAL_KERNEL_TYPE_IM2COL_F16,
|
2024-01-31 14:35:41 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_IM2COL_F32,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_UPSCALE_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_PAD_F32,
|
2024-03-03 13:23:52 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_ARANGE_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_TIMESTEP_EMBEDDING_F32,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_ARGSORT_F32_I32_ASC,
|
|
|
|
GGML_METAL_KERNEL_TYPE_ARGSORT_F32_I32_DESC,
|
|
|
|
GGML_METAL_KERNEL_TYPE_LEAKY_RELU_F32,
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H64,
|
|
|
|
GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H80,
|
|
|
|
GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H96,
|
|
|
|
GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H112,
|
|
|
|
GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H128,
|
|
|
|
GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H256,
|
|
|
|
GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_VEC_F16_H128,
|
|
|
|
GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_VEC_F16_H256,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_CPY_F32_F16,
|
|
|
|
GGML_METAL_KERNEL_TYPE_CPY_F32_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_CPY_F32_Q8_0,
|
|
|
|
GGML_METAL_KERNEL_TYPE_CPY_F32_Q4_0,
|
|
|
|
GGML_METAL_KERNEL_TYPE_CPY_F32_Q4_1,
|
2024-03-27 17:55:10 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_CPY_F32_Q5_0,
|
|
|
|
GGML_METAL_KERNEL_TYPE_CPY_F32_Q5_1,
|
|
|
|
GGML_METAL_KERNEL_TYPE_CPY_F32_IQ4_NL,
|
2024-01-13 17:03:45 +01:00
|
|
|
GGML_METAL_KERNEL_TYPE_CPY_F16_F16,
|
|
|
|
GGML_METAL_KERNEL_TYPE_CPY_F16_F32,
|
|
|
|
GGML_METAL_KERNEL_TYPE_CONCAT,
|
|
|
|
GGML_METAL_KERNEL_TYPE_SQR,
|
|
|
|
GGML_METAL_KERNEL_TYPE_SUM_ROWS,
|
|
|
|
|
|
|
|
GGML_METAL_KERNEL_TYPE_COUNT
|
|
|
|
};
|
|
|
|
|
2023-06-25 14:40:30 +02:00
|
|
|
struct ggml_metal_context {
|
2023-09-05 12:54:40 +02:00
|
|
|
int n_cb;
|
2023-06-25 14:40:30 +02:00
|
|
|
|
|
|
|
id<MTLDevice> device;
|
|
|
|
id<MTLCommandQueue> queue;
|
|
|
|
|
2023-09-05 12:54:40 +02:00
|
|
|
dispatch_queue_t d_queue;
|
|
|
|
|
2024-01-28 20:50:16 +01:00
|
|
|
struct ggml_metal_kernel kernels[GGML_METAL_KERNEL_TYPE_COUNT];
|
2024-01-13 17:03:45 +01:00
|
|
|
|
|
|
|
bool support_simdgroup_reduction;
|
|
|
|
bool support_simdgroup_mm;
|
2024-01-29 10:22:23 +01:00
|
|
|
|
|
|
|
bool should_capture_next_compute;
|
2023-06-25 14:40:30 +02:00
|
|
|
};
|
|
|
|
|
|
|
|
// MSL code
|
|
|
|
// TODO: move the contents here when ready
|
|
|
|
// for now it is easier to work in a separate file
|
2024-02-17 22:03:14 +01:00
|
|
|
// static NSString * const msl_library_source = @"see metal.metal";
|
2023-06-25 14:40:30 +02:00
|
|
|
|
|
|
|
// Here to assist with NSBundle Path Hack
|
|
|
|
@interface GGMLMetalClass : NSObject
|
|
|
|
@end
|
|
|
|
@implementation GGMLMetalClass
|
|
|
|
@end
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
static void ggml_metal_default_log_callback(enum ggml_log_level level, const char * msg, void * user_data) {
|
|
|
|
fprintf(stderr, "%s", msg);
|
|
|
|
|
|
|
|
UNUSED(level);
|
|
|
|
UNUSED(user_data);
|
|
|
|
}
|
|
|
|
|
|
|
|
ggml_log_callback ggml_metal_log_callback = ggml_metal_default_log_callback;
|
2023-11-03 20:35:05 +01:00
|
|
|
void * ggml_metal_log_user_data = NULL;
|
|
|
|
|
2023-11-12 15:36:20 +01:00
|
|
|
GGML_ATTRIBUTE_FORMAT(2, 3)
|
|
|
|
static void ggml_metal_log(enum ggml_log_level level, const char * format, ...){
|
2023-11-03 20:35:05 +01:00
|
|
|
if (ggml_metal_log_callback != NULL) {
|
|
|
|
va_list args;
|
|
|
|
va_start(args, format);
|
|
|
|
char buffer[128];
|
|
|
|
int len = vsnprintf(buffer, 128, format, args);
|
|
|
|
if (len < 128) {
|
|
|
|
ggml_metal_log_callback(level, buffer, ggml_metal_log_user_data);
|
|
|
|
} else {
|
|
|
|
char* buffer2 = malloc(len+1);
|
2023-12-08 12:50:50 +01:00
|
|
|
va_end(args);
|
|
|
|
va_start(args, format);
|
2023-11-03 20:35:05 +01:00
|
|
|
vsnprintf(buffer2, len+1, format, args);
|
|
|
|
buffer2[len] = 0;
|
|
|
|
ggml_metal_log_callback(level, buffer2, ggml_metal_log_user_data);
|
|
|
|
free(buffer2);
|
|
|
|
}
|
|
|
|
va_end(args);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2024-01-13 19:45:45 +01:00
|
|
|
static void * ggml_metal_host_malloc(size_t n) {
|
|
|
|
void * data = NULL;
|
2024-05-08 21:08:10 +02:00
|
|
|
|
|
|
|
#if TARGET_OS_OSX
|
|
|
|
kern_return_t err = vm_allocate((vm_map_t) mach_task_self(), (void *) &data, n, VM_FLAGS_ANYWHERE);
|
|
|
|
if (err != KERN_SUCCESS) {
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: vm_allocate failed\n", __func__);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
#else
|
2024-01-13 19:45:45 +01:00
|
|
|
const int result = posix_memalign((void **) &data, sysconf(_SC_PAGESIZE), n);
|
|
|
|
if (result != 0) {
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: posix_memalign failed\n", __func__);
|
|
|
|
return NULL;
|
|
|
|
}
|
2024-05-08 21:08:10 +02:00
|
|
|
#endif
|
2024-01-13 19:45:45 +01:00
|
|
|
|
|
|
|
return data;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct ggml_metal_context * ggml_metal_init(int n_cb) {
|
2023-11-03 20:35:05 +01:00
|
|
|
GGML_METAL_LOG_INFO("%s: allocating\n", __func__);
|
2023-09-05 19:57:27 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
#if TARGET_OS_OSX && !GGML_METAL_NDEBUG
|
2023-09-15 11:18:18 +02:00
|
|
|
// Show all the Metal device instances in the system
|
|
|
|
NSArray * devices = MTLCopyAllDevices();
|
2024-01-18 10:03:13 +01:00
|
|
|
for (id<MTLDevice> device in devices) {
|
2024-01-26 13:16:07 +01:00
|
|
|
GGML_METAL_LOG_INFO("%s: found device: %s\n", __func__, [[device name] UTF8String]);
|
2023-09-05 19:57:27 +02:00
|
|
|
}
|
2024-01-18 10:03:13 +01:00
|
|
|
[devices release]; // since it was created by a *Copy* C method
|
2023-09-15 11:18:18 +02:00
|
|
|
#endif
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2023-09-05 19:57:27 +02:00
|
|
|
// Pick and show default Metal device
|
2024-01-18 10:03:13 +01:00
|
|
|
id<MTLDevice> device = MTLCreateSystemDefaultDevice();
|
2024-01-26 13:16:07 +01:00
|
|
|
GGML_METAL_LOG_INFO("%s: picking default device: %s\n", __func__, [[device name] UTF8String]);
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2023-09-05 19:57:27 +02:00
|
|
|
// Configure context
|
|
|
|
struct ggml_metal_context * ctx = malloc(sizeof(struct ggml_metal_context));
|
|
|
|
ctx->device = device;
|
2023-09-05 12:54:40 +02:00
|
|
|
ctx->n_cb = MIN(n_cb, GGML_METAL_MAX_BUFFERS);
|
2023-06-25 14:40:30 +02:00
|
|
|
ctx->queue = [ctx->device newCommandQueue];
|
2023-09-15 11:18:18 +02:00
|
|
|
ctx->d_queue = dispatch_queue_create("ggml-metal", DISPATCH_QUEUE_CONCURRENT);
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-28 20:50:16 +01:00
|
|
|
id<MTLLibrary> metal_library;
|
|
|
|
|
2023-11-03 20:35:05 +01:00
|
|
|
// load library
|
2024-03-14 10:55:23 +01:00
|
|
|
//
|
|
|
|
// - first check if the library is embedded
|
|
|
|
// - then check if the library is in the bundle
|
|
|
|
// - if not found, load the source and compile it
|
|
|
|
// - if that fails, return NULL
|
2023-06-25 14:40:30 +02:00
|
|
|
{
|
2023-11-03 20:35:05 +01:00
|
|
|
NSBundle * bundle = nil;
|
|
|
|
#ifdef SWIFT_PACKAGE
|
|
|
|
bundle = SWIFTPM_MODULE_BUNDLE;
|
2023-06-25 14:40:30 +02:00
|
|
|
#else
|
2023-11-03 20:35:05 +01:00
|
|
|
bundle = [NSBundle bundleForClass:[GGMLMetalClass class]];
|
|
|
|
#endif
|
2024-03-14 10:55:23 +01:00
|
|
|
|
2023-06-25 14:40:30 +02:00
|
|
|
NSError * error = nil;
|
2024-03-14 10:55:23 +01:00
|
|
|
|
|
|
|
#if GGML_METAL_EMBED_LIBRARY
|
|
|
|
const bool try_metallib = false;
|
|
|
|
#else
|
|
|
|
const bool try_metallib = true;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
NSString * path_lib = [bundle pathForResource:@"default" ofType:@"metallib"];
|
|
|
|
if (try_metallib && path_lib != nil) {
|
2024-01-02 09:57:44 +01:00
|
|
|
// pre-compiled library found
|
2024-03-14 10:55:23 +01:00
|
|
|
NSURL * libURL = [NSURL fileURLWithPath:path_lib];
|
|
|
|
GGML_METAL_LOG_INFO("%s: loading '%s'\n", __func__, [path_lib UTF8String]);
|
|
|
|
|
2024-01-28 20:50:16 +01:00
|
|
|
metal_library = [ctx->device newLibraryWithURL:libURL error:&error];
|
2024-01-25 10:26:17 +01:00
|
|
|
if (error) {
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
|
|
|
|
return NULL;
|
|
|
|
}
|
2023-11-03 20:35:05 +01:00
|
|
|
} else {
|
2024-02-11 15:41:41 +01:00
|
|
|
#if GGML_METAL_EMBED_LIBRARY
|
|
|
|
GGML_METAL_LOG_INFO("%s: using embedded metal library\n", __func__);
|
|
|
|
|
|
|
|
extern const char ggml_metallib_start[];
|
|
|
|
extern const char ggml_metallib_end[];
|
|
|
|
|
2024-03-14 10:55:23 +01:00
|
|
|
NSString * src = [[NSString alloc] initWithBytes:ggml_metallib_start length:(ggml_metallib_end-ggml_metallib_start) encoding:NSUTF8StringEncoding];
|
2024-02-11 15:41:41 +01:00
|
|
|
#else
|
2024-01-05 15:30:52 +01:00
|
|
|
GGML_METAL_LOG_INFO("%s: default.metallib not found, loading from source\n", __func__);
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-03-14 10:55:23 +01:00
|
|
|
NSString * path_source;
|
|
|
|
NSString * path_resource = [[NSProcessInfo processInfo].environment objectForKey:@"GGML_METAL_PATH_RESOURCES"];
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-03-14 10:55:23 +01:00
|
|
|
GGML_METAL_LOG_INFO("%s: GGML_METAL_PATH_RESOURCES = %s\n", __func__, path_resource ? [path_resource UTF8String] : "nil");
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-03-14 10:55:23 +01:00
|
|
|
if (path_resource) {
|
|
|
|
path_source = [path_resource stringByAppendingPathComponent:@"ggml-metal.metal"];
|
2023-11-03 20:35:05 +01:00
|
|
|
} else {
|
2024-03-14 10:55:23 +01:00
|
|
|
path_source = [bundle pathForResource:@"ggml-metal" ofType:@"metal"];
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
2024-03-14 10:55:23 +01:00
|
|
|
|
|
|
|
if (path_source == nil) {
|
2023-11-03 20:35:05 +01:00
|
|
|
GGML_METAL_LOG_WARN("%s: error: could not use bundle path to find ggml-metal.metal, falling back to trying cwd\n", __func__);
|
2024-03-14 10:55:23 +01:00
|
|
|
path_source = @"ggml-metal.metal";
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
2024-03-14 10:55:23 +01:00
|
|
|
|
|
|
|
GGML_METAL_LOG_INFO("%s: loading '%s'\n", __func__, [path_source UTF8String]);
|
|
|
|
|
|
|
|
NSString * src = [NSString stringWithContentsOfFile:path_source encoding:NSUTF8StringEncoding error:&error];
|
2023-11-03 20:35:05 +01:00
|
|
|
if (error) {
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
|
|
|
|
return NULL;
|
|
|
|
}
|
2024-03-14 10:55:23 +01:00
|
|
|
#endif // GGML_METAL_EMBED_LIBRARY
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-17 17:38:39 +01:00
|
|
|
@autoreleasepool {
|
|
|
|
// dictionary of preprocessor macros
|
|
|
|
NSMutableDictionary * prep = [NSMutableDictionary dictionary];
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2023-07-02 20:45:27 +02:00
|
|
|
#ifdef GGML_QKK_64
|
2024-03-12 13:27:20 +01:00
|
|
|
prep[@"GGML_QKK_64"] = @(1);
|
2023-07-02 20:45:27 +02:00
|
|
|
#endif
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-01-17 17:38:39 +01:00
|
|
|
MTLCompileOptions* options = [MTLCompileOptions new];
|
|
|
|
options.preprocessorMacros = prep;
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-01-17 17:38:39 +01:00
|
|
|
//[options setFastMathEnabled:false];
|
2024-01-02 09:57:44 +01:00
|
|
|
|
2024-01-28 20:50:16 +01:00
|
|
|
metal_library = [ctx->device newLibraryWithSource:src options:options error:&error];
|
2024-01-25 10:26:17 +01:00
|
|
|
if (error) {
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
|
|
|
|
return NULL;
|
|
|
|
}
|
2024-01-17 17:38:39 +01:00
|
|
|
}
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
2023-06-25 14:40:30 +02:00
|
|
|
}
|
|
|
|
|
2023-12-07 21:27:19 +01:00
|
|
|
// print MTL GPU family:
|
|
|
|
GGML_METAL_LOG_INFO("%s: GPU name: %s\n", __func__, [[ctx->device name] UTF8String]);
|
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
const NSInteger MTLGPUFamilyMetal3 = 5001;
|
|
|
|
|
2023-12-07 21:27:19 +01:00
|
|
|
// determine max supported GPU family
|
|
|
|
// https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf
|
|
|
|
// https://developer.apple.com/metal/Metal-Feature-Set-Tables.pdf
|
2024-01-13 17:03:45 +01:00
|
|
|
{
|
|
|
|
for (int i = MTLGPUFamilyApple1 + 20; i >= MTLGPUFamilyApple1; --i) {
|
|
|
|
if ([ctx->device supportsFamily:i]) {
|
|
|
|
GGML_METAL_LOG_INFO("%s: GPU family: MTLGPUFamilyApple%d (%d)\n", __func__, i - (int) MTLGPUFamilyApple1 + 1, i);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
for (int i = MTLGPUFamilyCommon1 + 5; i >= MTLGPUFamilyCommon1; --i) {
|
|
|
|
if ([ctx->device supportsFamily:i]) {
|
|
|
|
GGML_METAL_LOG_INFO("%s: GPU family: MTLGPUFamilyCommon%d (%d)\n", __func__, i - (int) MTLGPUFamilyCommon1 + 1, i);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
for (int i = MTLGPUFamilyMetal3 + 5; i >= MTLGPUFamilyMetal3; --i) {
|
|
|
|
if ([ctx->device supportsFamily:i]) {
|
|
|
|
GGML_METAL_LOG_INFO("%s: GPU family: MTLGPUFamilyMetal%d (%d)\n", __func__, i - (int) MTLGPUFamilyMetal3 + 3, i);
|
|
|
|
break;
|
|
|
|
}
|
2023-12-07 21:27:19 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
ctx->support_simdgroup_reduction = [ctx->device supportsFamily:MTLGPUFamilyApple7];
|
|
|
|
ctx->support_simdgroup_reduction |= [ctx->device supportsFamily:MTLGPUFamilyMetal3];
|
|
|
|
|
|
|
|
ctx->support_simdgroup_mm = [ctx->device supportsFamily:MTLGPUFamilyApple7];
|
|
|
|
|
|
|
|
GGML_METAL_LOG_INFO("%s: simdgroup reduction support = %s\n", __func__, ctx->support_simdgroup_reduction ? "true" : "false");
|
|
|
|
GGML_METAL_LOG_INFO("%s: simdgroup matrix mul. support = %s\n", __func__, ctx->support_simdgroup_mm ? "true" : "false");
|
2023-12-07 21:27:19 +01:00
|
|
|
GGML_METAL_LOG_INFO("%s: hasUnifiedMemory = %s\n", __func__, ctx->device.hasUnifiedMemory ? "true" : "false");
|
2024-01-16 14:33:02 +01:00
|
|
|
|
2024-01-29 12:29:46 +01:00
|
|
|
ctx->should_capture_next_compute = false;
|
|
|
|
|
2024-01-16 14:33:02 +01:00
|
|
|
#if TARGET_OS_OSX || (TARGET_OS_IOS && __clang_major__ >= 15)
|
|
|
|
if (@available(macOS 10.12, iOS 16.0, *)) {
|
|
|
|
GGML_METAL_LOG_INFO("%s: recommendedMaxWorkingSetSize = %8.2f MB\n", __func__, ctx->device.recommendedMaxWorkingSetSize / 1e6);
|
|
|
|
}
|
|
|
|
#elif TARGET_OS_OSX
|
2023-12-07 21:27:19 +01:00
|
|
|
if (ctx->device.maxTransferRate != 0) {
|
|
|
|
GGML_METAL_LOG_INFO("%s: maxTransferRate = %8.2f MB/s\n", __func__, ctx->device.maxTransferRate / 1e6);
|
|
|
|
} else {
|
|
|
|
GGML_METAL_LOG_INFO("%s: maxTransferRate = built-in GPU\n", __func__);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2023-06-25 14:40:30 +02:00
|
|
|
// load kernels
|
|
|
|
{
|
2023-09-05 12:54:40 +02:00
|
|
|
NSError * error = nil;
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-28 20:50:16 +01:00
|
|
|
for (int i = 0; i < GGML_METAL_KERNEL_TYPE_COUNT; ++i) {
|
2024-01-13 17:03:45 +01:00
|
|
|
ctx->kernels[i].pipeline = nil;
|
|
|
|
}
|
|
|
|
|
2023-11-03 20:35:05 +01:00
|
|
|
/*
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
GGML_METAL_LOG_INFO("%s: loaded %-40s %16p | th_max = %4d | th_width = %4d\n", __func__, "kernel_"#name, (void *) kernel->pipeline, \
|
2024-01-13 17:03:45 +01:00
|
|
|
(int) kernel->pipeline.maxTotalThreadsPerThreadgroup, \
|
|
|
|
(int) kernel->pipeline.threadExecutionWidth); \
|
2023-11-03 20:35:05 +01:00
|
|
|
*/
|
2024-01-13 17:03:45 +01:00
|
|
|
#define GGML_METAL_ADD_KERNEL(e, name, supported) \
|
|
|
|
if (supported) { \
|
|
|
|
struct ggml_metal_kernel * kernel = &ctx->kernels[e]; \
|
2024-01-28 20:50:16 +01:00
|
|
|
id<MTLFunction> metal_function = [metal_library newFunctionWithName:@"kernel_"#name]; \
|
|
|
|
kernel->pipeline = [ctx->device newComputePipelineStateWithFunction:metal_function error:&error]; \
|
|
|
|
[metal_function release]; \
|
2024-01-13 17:03:45 +01:00
|
|
|
if (error) { \
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: load pipeline error: %s\n", __func__, [[error description] UTF8String]); \
|
2024-01-28 20:50:16 +01:00
|
|
|
[metal_library release]; \
|
2024-01-13 17:03:45 +01:00
|
|
|
return NULL; \
|
|
|
|
} \
|
|
|
|
} else { \
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
GGML_METAL_LOG_WARN("%s: skipping %-40s (not supported)\n", __func__, "kernel_"#name); \
|
2023-09-05 12:54:40 +02:00
|
|
|
}
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
// simd_sum and simd_max requires MTLGPUFamilyApple7
|
|
|
|
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ADD, add, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ADD_ROW, add_row, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL, mul, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_ROW, mul_row, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_DIV, div, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_DIV_ROW, div_row, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SCALE, scale, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SCALE_4, scale_4, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CLAMP, clamp, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_TANH, tanh, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_RELU, relu, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SIGMOID, sigmoid, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GELU, gelu, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GELU_4, gelu_4, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GELU_QUICK, gelu_quick, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GELU_QUICK_4, gelu_quick_4, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SILU, silu, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SILU_4, silu_4, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SOFT_MAX_F16, soft_max_f16, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SOFT_MAX_F16_4, soft_max_f16_4, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SOFT_MAX_F32, soft_max_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SOFT_MAX_F32_4, soft_max_f32_4, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_DIAG_MASK_INF, diag_mask_inf, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_DIAG_MASK_INF_8, diag_mask_inf_8, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_F32, get_rows_f32, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_F16, get_rows_f16, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_Q4_0, get_rows_q4_0, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_Q4_1, get_rows_q4_1, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_Q5_0, get_rows_q5_0, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_Q5_1, get_rows_q5_1, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_Q8_0, get_rows_q8_0, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_Q2_K, get_rows_q2_K, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_Q3_K, get_rows_q3_K, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_Q4_K, get_rows_q4_K, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_Q5_K, get_rows_q5_K, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_Q6_K, get_rows_q6_K, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_XXS, get_rows_iq2_xxs, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_XS, get_rows_iq2_xs, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ3_XXS, get_rows_iq3_xxs, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ3_S, get_rows_iq3_s, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_S, get_rows_iq2_s, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ1_S, get_rows_iq1_s, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ1_M, get_rows_iq1_m, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_NL, get_rows_iq4_nl, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_XS, get_rows_iq4_xs, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_I32, get_rows_i32, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_RMS_NORM, rms_norm, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GROUP_NORM, group_norm, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_NORM, norm, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_F32_F32, mul_mv_f32_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F16, mul_mv_f16_f16, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F32, mul_mv_f16_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F32_1ROW, mul_mv_f16_f32_1row, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F32_L4, mul_mv_f16_f32_l4, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_Q4_0_F32, mul_mv_q4_0_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_Q4_1_F32, mul_mv_q4_1_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_Q5_0_F32, mul_mv_q5_0_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_Q5_1_F32, mul_mv_q5_1_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_Q8_0_F32, mul_mv_q8_0_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_Q2_K_F32, mul_mv_q2_K_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_Q3_K_F32, mul_mv_q3_K_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_Q4_K_F32, mul_mv_q4_K_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_Q5_K_F32, mul_mv_q5_K_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_Q6_K_F32, mul_mv_q6_K_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_XXS_F32, mul_mv_iq2_xxs_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_XS_F32, mul_mv_iq2_xs_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ3_XXS_F32, mul_mv_iq3_xxs_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ3_S_F32, mul_mv_iq3_s_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_S_F32, mul_mv_iq2_s_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ1_S_F32, mul_mv_iq1_s_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ1_M_F32, mul_mv_iq1_m_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_NL_F32, mul_mv_iq4_nl_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_XS_F32, mul_mv_iq4_xs_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F32_F32, mul_mv_id_f32_f32, ctx->support_simdgroup_reduction);
|
|
|
|
//GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F16, mul_mv_id_f16_f16, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F32, mul_mv_id_f16_f32, ctx->support_simdgroup_reduction);
|
|
|
|
//GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F32_1ROW, mul_mv_id_f16_f32_1row, ctx->support_simdgroup_reduction);
|
|
|
|
//GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F32_L4, mul_mv_id_f16_f32_l4, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q4_0_F32, mul_mv_id_q4_0_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q4_1_F32, mul_mv_id_q4_1_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q5_0_F32, mul_mv_id_q5_0_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q5_1_F32, mul_mv_id_q5_1_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q8_0_F32, mul_mv_id_q8_0_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q2_K_F32, mul_mv_id_q2_K_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q3_K_F32, mul_mv_id_q3_K_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q4_K_F32, mul_mv_id_q4_K_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q5_K_F32, mul_mv_id_q5_K_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q6_K_F32, mul_mv_id_q6_K_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_XXS_F32, mul_mv_id_iq2_xxs_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_XS_F32, mul_mv_id_iq2_xs_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ3_XXS_F32, mul_mv_id_iq3_xxs_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ3_S_F32, mul_mv_id_iq3_s_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_S_F32, mul_mv_id_iq2_s_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ1_S_F32, mul_mv_id_iq1_s_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ1_M_F32, mul_mv_id_iq1_m_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_NL_F32, mul_mv_id_iq4_nl_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_XS_F32, mul_mv_id_iq4_xs_f32, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_F32_F32, mul_mm_f32_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_F16_F32, mul_mm_f16_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_Q4_0_F32, mul_mm_q4_0_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_Q4_1_F32, mul_mm_q4_1_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_Q5_0_F32, mul_mm_q5_0_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_Q5_1_F32, mul_mm_q5_1_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_Q8_0_F32, mul_mm_q8_0_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_Q2_K_F32, mul_mm_q2_K_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_Q3_K_F32, mul_mm_q3_K_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_Q4_K_F32, mul_mm_q4_K_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_Q5_K_F32, mul_mm_q5_K_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_Q6_K_F32, mul_mm_q6_K_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_XXS_F32, mul_mm_iq2_xxs_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_XS_F32, mul_mm_iq2_xs_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_XXS_F32, mul_mm_iq3_xxs_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_S_F32, mul_mm_iq3_s_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_S_F32, mul_mm_iq2_s_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ1_S_F32, mul_mm_iq1_s_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ1_M_F32, mul_mm_iq1_m_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_NL_F32, mul_mm_iq4_nl_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_XS_F32, mul_mm_iq4_xs_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_F32_F32, mul_mm_id_f32_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_F16_F32, mul_mm_id_f16_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q4_0_F32, mul_mm_id_q4_0_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q4_1_F32, mul_mm_id_q4_1_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q5_0_F32, mul_mm_id_q5_0_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q5_1_F32, mul_mm_id_q5_1_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q8_0_F32, mul_mm_id_q8_0_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q2_K_F32, mul_mm_id_q2_K_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q3_K_F32, mul_mm_id_q3_K_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q4_K_F32, mul_mm_id_q4_K_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q5_K_F32, mul_mm_id_q5_K_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q6_K_F32, mul_mm_id_q6_K_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_XXS_F32, mul_mm_id_iq2_xxs_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_XS_F32, mul_mm_id_iq2_xs_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ3_XXS_F32, mul_mm_id_iq3_xxs_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ3_S_F32, mul_mm_id_iq3_s_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_S_F32, mul_mm_id_iq2_s_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ1_S_F32, mul_mm_id_iq1_s_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ1_M_F32, mul_mm_id_iq1_m_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_NL_F32, mul_mm_id_iq4_nl_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_XS_F32, mul_mm_id_iq4_xs_f32, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ROPE_F32, rope_f32, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ROPE_F16, rope_f16, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_IM2COL_F16, im2col_f16, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_IM2COL_F32, im2col_f32, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_UPSCALE_F32, upscale_f32, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_PAD_F32, pad_f32, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_TIMESTEP_EMBEDDING_F32, timestep_embedding_f32, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ARANGE_F32, arange_f32, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ARGSORT_F32_I32_ASC, argsort_f32_i32_asc, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ARGSORT_F32_I32_DESC, argsort_f32_i32_desc, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_LEAKY_RELU_F32, leaky_relu_f32, true);
|
2024-05-10 17:20:10 +02:00
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H64, flash_attn_ext_f16_h64, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H80, flash_attn_ext_f16_h80, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H96, flash_attn_ext_f16_h96, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H112, flash_attn_ext_f16_h112, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H128, flash_attn_ext_f16_h128, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H256, flash_attn_ext_f16_h256, ctx->support_simdgroup_mm);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_VEC_F16_H128, flash_attn_ext_vec_f16_h128, ctx->support_simdgroup_reduction);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_VEC_F16_H256, flash_attn_ext_vec_f16_h256, ctx->support_simdgroup_reduction);
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CPY_F32_F16, cpy_f32_f16, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CPY_F32_F32, cpy_f32_f32, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CPY_F32_Q8_0, cpy_f32_q8_0, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CPY_F32_Q4_0, cpy_f32_q4_0, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CPY_F32_Q4_1, cpy_f32_q4_1, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CPY_F32_Q5_0, cpy_f32_q5_0, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CPY_F32_Q5_1, cpy_f32_q5_1, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CPY_F32_IQ4_NL, cpy_f32_iq4_nl, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CPY_F16_F16, cpy_f16_f16, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CPY_F16_F32, cpy_f16_f32, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_CONCAT, concat, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SQR, sqr, true);
|
|
|
|
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_SUM_ROWS, sum_rows, true);
|
2023-06-25 14:40:30 +02:00
|
|
|
}
|
|
|
|
|
2024-01-28 20:50:16 +01:00
|
|
|
[metal_library release];
|
2023-06-25 14:40:30 +02:00
|
|
|
return ctx;
|
|
|
|
}
|
|
|
|
|
2024-01-13 19:45:45 +01:00
|
|
|
static void ggml_metal_free(struct ggml_metal_context * ctx) {
|
2023-11-03 20:35:05 +01:00
|
|
|
GGML_METAL_LOG_INFO("%s: deallocating\n", __func__);
|
2023-09-05 12:54:40 +02:00
|
|
|
|
2024-01-28 20:50:16 +01:00
|
|
|
for (int i = 0; i < GGML_METAL_KERNEL_TYPE_COUNT; ++i) {
|
|
|
|
[ctx->kernels[i].pipeline release];
|
2024-01-13 17:03:45 +01:00
|
|
|
}
|
|
|
|
|
2023-09-05 12:54:40 +02:00
|
|
|
[ctx->queue release];
|
|
|
|
[ctx->device release];
|
|
|
|
|
|
|
|
dispatch_release(ctx->d_queue);
|
|
|
|
|
2023-06-25 14:40:30 +02:00
|
|
|
free(ctx);
|
|
|
|
}
|
|
|
|
|
2023-12-07 21:27:19 +01:00
|
|
|
// temporarily defined here for compatibility between ggml-backend and the old API
|
2023-12-22 16:53:39 +01:00
|
|
|
|
|
|
|
struct ggml_backend_metal_buffer {
|
|
|
|
void * data;
|
|
|
|
size_t size;
|
2023-12-07 21:27:19 +01:00
|
|
|
|
|
|
|
id<MTLBuffer> metal;
|
|
|
|
};
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
struct ggml_backend_metal_buffer_context {
|
|
|
|
void * all_data;
|
|
|
|
size_t all_size;
|
|
|
|
bool owned;
|
|
|
|
|
|
|
|
// multiple buffers are used only to avoid the maximum buffer size limitation when using mmap
|
|
|
|
int n_buffers;
|
|
|
|
struct ggml_backend_metal_buffer buffers[GGML_METAL_MAX_BUFFERS];
|
|
|
|
};
|
|
|
|
|
2023-06-25 14:40:30 +02:00
|
|
|
// finds the Metal buffer that contains the tensor data on the GPU device
|
|
|
|
// the assumption is that there is 1-to-1 mapping between the host and device memory buffers, so we can find the
|
|
|
|
// Metal buffer based on the host memory pointer
|
|
|
|
//
|
2024-01-26 13:16:07 +01:00
|
|
|
static id<MTLBuffer> ggml_metal_get_buffer(struct ggml_tensor * t, size_t * offs) {
|
2023-11-03 20:35:05 +01:00
|
|
|
//GGML_METAL_LOG_INFO("%s: data tensor '%16s', offs_data = %8ld, offs_eval = %8ld, offs_cach = %8ld\n", __func__, t->name, offs_data, offs_eval, offs_cach);
|
2023-06-25 14:40:30 +02:00
|
|
|
|
|
|
|
const int64_t tsize = ggml_nbytes(t);
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
ggml_backend_buffer_t buffer = t->view_src ? t->view_src->buffer : t->buffer;
|
|
|
|
|
2024-01-26 13:16:07 +01:00
|
|
|
struct ggml_backend_metal_buffer_context * buf_ctx = (struct ggml_backend_metal_buffer_context *) buffer->context;
|
2023-11-12 14:31:08 +01:00
|
|
|
|
2023-06-25 14:40:30 +02:00
|
|
|
// find the view that contains the tensor fully
|
2024-01-26 13:16:07 +01:00
|
|
|
for (int i = 0; i < buf_ctx->n_buffers; ++i) {
|
|
|
|
const int64_t ioffs = (int64_t) t->data - (int64_t) buf_ctx->buffers[i].data;
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-26 13:16:07 +01:00
|
|
|
//GGML_METAL_LOG_INFO("ioffs = %10ld, tsize = %10ld, sum = %10ld, buf_ctx->buffers[%d].size = %10ld\n", ioffs, tsize, ioffs + tsize, i, buf_ctx->buffers[i].size);
|
|
|
|
if (ioffs >= 0 && ioffs + tsize <= (int64_t) buf_ctx->buffers[i].size) {
|
2023-06-25 14:40:30 +02:00
|
|
|
*offs = (size_t) ioffs;
|
|
|
|
|
2024-01-26 13:16:07 +01:00
|
|
|
//GGML_METAL_LOG_INFO("%s: tensor '%16s', offs = %8ld\n", __func__, t->name, *offs);
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-26 13:16:07 +01:00
|
|
|
return buf_ctx->buffers[i].metal;
|
2023-06-25 14:40:30 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2024-01-26 13:16:07 +01:00
|
|
|
GGML_METAL_LOG_ERROR("%s: error: tensor '%s' buffer is nil\n", __func__, t->name);
|
2023-06-25 14:40:30 +02:00
|
|
|
|
|
|
|
return nil;
|
|
|
|
}
|
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
static bool ggml_metal_supports_op(const struct ggml_metal_context * ctx, const struct ggml_tensor * op) {
|
2023-12-07 21:27:19 +01:00
|
|
|
switch (op->op) {
|
|
|
|
case GGML_OP_UNARY:
|
|
|
|
switch (ggml_get_unary_op(op)) {
|
2023-12-13 20:55:03 +01:00
|
|
|
case GGML_UNARY_OP_TANH:
|
2023-12-07 21:27:19 +01:00
|
|
|
case GGML_UNARY_OP_RELU:
|
2024-05-01 23:44:26 +02:00
|
|
|
case GGML_UNARY_OP_SIGMOID:
|
2023-12-07 21:27:19 +01:00
|
|
|
case GGML_UNARY_OP_GELU:
|
2023-12-13 20:55:03 +01:00
|
|
|
case GGML_UNARY_OP_GELU_QUICK:
|
|
|
|
case GGML_UNARY_OP_SILU:
|
2023-12-07 21:27:19 +01:00
|
|
|
return true;
|
|
|
|
default:
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
case GGML_OP_NONE:
|
|
|
|
case GGML_OP_RESHAPE:
|
|
|
|
case GGML_OP_VIEW:
|
|
|
|
case GGML_OP_TRANSPOSE:
|
|
|
|
case GGML_OP_PERMUTE:
|
|
|
|
case GGML_OP_CONCAT:
|
|
|
|
case GGML_OP_ADD:
|
2023-12-13 20:55:03 +01:00
|
|
|
case GGML_OP_ACC:
|
2023-12-07 21:27:19 +01:00
|
|
|
case GGML_OP_MUL:
|
|
|
|
case GGML_OP_DIV:
|
|
|
|
case GGML_OP_SCALE:
|
2024-04-14 13:14:19 +02:00
|
|
|
case GGML_OP_CLAMP:
|
2023-12-07 21:27:19 +01:00
|
|
|
case GGML_OP_SQR:
|
|
|
|
case GGML_OP_SUM_ROWS:
|
2024-01-13 17:03:45 +01:00
|
|
|
return true;
|
2023-12-07 21:27:19 +01:00
|
|
|
case GGML_OP_SOFT_MAX:
|
|
|
|
case GGML_OP_RMS_NORM:
|
2023-12-13 20:55:03 +01:00
|
|
|
case GGML_OP_GROUP_NORM:
|
2024-01-13 17:03:45 +01:00
|
|
|
return ctx->support_simdgroup_reduction;
|
2023-12-07 21:27:19 +01:00
|
|
|
case GGML_OP_NORM:
|
|
|
|
case GGML_OP_ROPE:
|
|
|
|
case GGML_OP_IM2COL:
|
2024-01-31 14:35:41 +01:00
|
|
|
return true;
|
|
|
|
case GGML_OP_POOL_1D:
|
|
|
|
case GGML_OP_POOL_2D:
|
|
|
|
return false;
|
2023-12-13 20:55:03 +01:00
|
|
|
case GGML_OP_UPSCALE:
|
|
|
|
case GGML_OP_PAD:
|
2024-03-03 13:23:52 +01:00
|
|
|
case GGML_OP_ARANGE:
|
|
|
|
case GGML_OP_TIMESTEP_EMBEDDING:
|
2023-12-07 21:27:19 +01:00
|
|
|
case GGML_OP_ARGSORT:
|
2023-12-13 20:55:03 +01:00
|
|
|
case GGML_OP_LEAKY_RELU:
|
2024-01-13 17:03:45 +01:00
|
|
|
return true;
|
2024-05-10 17:20:10 +02:00
|
|
|
case GGML_OP_FLASH_ATTN_EXT:
|
|
|
|
return ctx->support_simdgroup_mm; // TODO: over-restricted for vec-kernels
|
2023-12-07 21:27:19 +01:00
|
|
|
case GGML_OP_MUL_MAT:
|
|
|
|
case GGML_OP_MUL_MAT_ID:
|
2024-01-23 14:50:56 +01:00
|
|
|
return ctx->support_simdgroup_reduction &&
|
|
|
|
(op->src[0]->type != GGML_TYPE_F32 || op->src[1]->type == GGML_TYPE_F32);
|
2023-12-13 20:55:03 +01:00
|
|
|
case GGML_OP_CPY:
|
|
|
|
case GGML_OP_DUP:
|
|
|
|
case GGML_OP_CONT:
|
|
|
|
{
|
|
|
|
switch (op->src[0]->type) {
|
|
|
|
case GGML_TYPE_F32:
|
|
|
|
switch (op->type) {
|
|
|
|
case GGML_TYPE_F16:
|
|
|
|
case GGML_TYPE_F32:
|
|
|
|
case GGML_TYPE_Q8_0:
|
|
|
|
case GGML_TYPE_Q4_0:
|
|
|
|
case GGML_TYPE_Q4_1:
|
2024-03-27 17:55:10 +01:00
|
|
|
case GGML_TYPE_Q5_0:
|
|
|
|
case GGML_TYPE_Q5_1:
|
|
|
|
case GGML_TYPE_IQ4_NL:
|
2023-12-13 20:55:03 +01:00
|
|
|
return true;
|
|
|
|
default:
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
case GGML_TYPE_F16:
|
|
|
|
switch (op->type) {
|
|
|
|
case GGML_TYPE_F16:
|
|
|
|
case GGML_TYPE_F32:
|
|
|
|
return true;
|
|
|
|
default:
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
return false;
|
|
|
|
};
|
|
|
|
}
|
2023-12-07 21:27:19 +01:00
|
|
|
case GGML_OP_DIAG_MASK_INF:
|
|
|
|
case GGML_OP_GET_ROWS:
|
|
|
|
{
|
2024-05-08 08:30:09 +02:00
|
|
|
return op->src[0]->type != GGML_TYPE_BF16 && op->ne[3] == 1;
|
2023-12-07 21:27:19 +01:00
|
|
|
}
|
|
|
|
default:
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-03-04 10:05:42 +01:00
|
|
|
static enum ggml_status ggml_metal_graph_compute(
|
2023-06-25 14:40:30 +02:00
|
|
|
struct ggml_metal_context * ctx,
|
|
|
|
struct ggml_cgraph * gf) {
|
2023-09-05 12:54:40 +02:00
|
|
|
|
2024-02-10 11:53:28 +01:00
|
|
|
@autoreleasepool {
|
2023-09-05 12:54:40 +02:00
|
|
|
MTLComputePassDescriptor * edesc = MTLComputePassDescriptor.computePassDescriptor;
|
2024-01-13 19:45:45 +01:00
|
|
|
edesc.dispatchType = MTLDispatchTypeSerial;
|
2023-09-05 12:54:40 +02:00
|
|
|
|
2023-06-25 14:40:30 +02:00
|
|
|
// create multiple command buffers and enqueue them
|
|
|
|
// then, we encode the graph into the command buffers in parallel
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const int n_nodes = gf->n_nodes;
|
2023-09-05 12:54:40 +02:00
|
|
|
const int n_cb = ctx->n_cb;
|
2024-01-18 10:03:13 +01:00
|
|
|
const int n_nodes_per_cb = (n_nodes + n_cb - 1) / n_cb;
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-29 10:22:23 +01:00
|
|
|
const bool should_capture = ctx->should_capture_next_compute;
|
|
|
|
if (should_capture) {
|
|
|
|
ctx->should_capture_next_compute = false;
|
|
|
|
|
|
|
|
MTLCaptureDescriptor * descriptor = [MTLCaptureDescriptor new];
|
|
|
|
descriptor.captureObject = ctx->queue;
|
|
|
|
|
|
|
|
NSError * error = nil;
|
|
|
|
if (![[MTLCaptureManager sharedCaptureManager] startCaptureWithDescriptor:descriptor error:&error]) {
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: unable to start capture '%s'\n", __func__, [[error localizedDescription] UTF8String]);
|
|
|
|
GGML_ASSERT(!"capture failed");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
id<MTLCommandBuffer> command_buffer_builder[n_cb];
|
|
|
|
for (int cb_idx = 0; cb_idx < n_cb; ++cb_idx) {
|
|
|
|
id<MTLCommandBuffer> command_buffer = [ctx->queue commandBufferWithUnretainedReferences];
|
|
|
|
command_buffer_builder[cb_idx] = command_buffer;
|
2023-06-25 14:40:30 +02:00
|
|
|
|
|
|
|
// enqueue the command buffers in order to specify their execution order
|
2024-01-18 10:03:13 +01:00
|
|
|
[command_buffer enqueue];
|
2023-09-05 12:54:40 +02:00
|
|
|
}
|
2024-01-29 10:22:23 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const id<MTLCommandBuffer> *command_buffers = command_buffer_builder;
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
dispatch_apply(n_cb, ctx->d_queue, ^(size_t iter) {
|
|
|
|
const int cb_idx = iter;
|
2023-12-13 20:55:03 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
size_t offs_src0 = 0;
|
|
|
|
size_t offs_src1 = 0;
|
2024-02-19 14:18:09 +01:00
|
|
|
size_t offs_src2 = 0;
|
2024-01-18 10:03:13 +01:00
|
|
|
size_t offs_dst = 0;
|
2023-12-13 20:55:03 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
id<MTLCommandBuffer> command_buffer = command_buffers[cb_idx];
|
|
|
|
id<MTLComputeCommandEncoder> encoder = [command_buffer computeCommandEncoderWithDescriptor: edesc];
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const int node_start = (cb_idx + 0) * n_nodes_per_cb;
|
|
|
|
const int node_end = MIN((cb_idx == n_cb - 1) ? n_nodes : (cb_idx + 1) * n_nodes_per_cb, n_nodes);
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
for (int i = node_start; i < node_end; ++i) {
|
|
|
|
if (i == -1) {
|
|
|
|
[encoder memoryBarrierWithScope:MTLBarrierScopeBuffers];
|
|
|
|
continue;
|
|
|
|
}
|
2023-12-13 20:55:03 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
//GGML_METAL_LOG_INFO("%s: encoding node %3d, op = %8s\n", __func__, i, ggml_op_name(gf->nodes[i]->op));
|
|
|
|
|
|
|
|
struct ggml_tensor * src0 = gf->nodes[i]->src[0];
|
|
|
|
struct ggml_tensor * src1 = gf->nodes[i]->src[1];
|
2024-02-19 14:18:09 +01:00
|
|
|
struct ggml_tensor * src2 = gf->nodes[i]->src[2];
|
2024-01-18 10:03:13 +01:00
|
|
|
struct ggml_tensor * dst = gf->nodes[i];
|
|
|
|
|
2024-03-27 17:55:10 +01:00
|
|
|
if (ggml_is_empty(dst)) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
switch (dst->op) {
|
|
|
|
case GGML_OP_NONE:
|
|
|
|
case GGML_OP_RESHAPE:
|
|
|
|
case GGML_OP_VIEW:
|
|
|
|
case GGML_OP_TRANSPOSE:
|
|
|
|
case GGML_OP_PERMUTE:
|
|
|
|
{
|
|
|
|
// noop -> next node
|
|
|
|
} continue;
|
|
|
|
default:
|
|
|
|
{
|
|
|
|
} break;
|
|
|
|
}
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
if (!ggml_metal_supports_op(ctx, dst)) {
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: unsupported op '%s'\n", __func__, ggml_op_desc(dst));
|
|
|
|
GGML_ASSERT(!"unsupported op");
|
|
|
|
}
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-29 10:22:23 +01:00
|
|
|
if (should_capture) {
|
|
|
|
[encoder pushDebugGroup:[NSString stringWithCString:ggml_op_desc(dst) encoding:NSUTF8StringEncoding]];
|
|
|
|
}
|
2023-12-13 20:55:03 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const int64_t ne00 = src0 ? src0->ne[0] : 0;
|
|
|
|
const int64_t ne01 = src0 ? src0->ne[1] : 0;
|
|
|
|
const int64_t ne02 = src0 ? src0->ne[2] : 0;
|
|
|
|
const int64_t ne03 = src0 ? src0->ne[3] : 0;
|
|
|
|
|
|
|
|
const uint64_t nb00 = src0 ? src0->nb[0] : 0;
|
|
|
|
const uint64_t nb01 = src0 ? src0->nb[1] : 0;
|
|
|
|
const uint64_t nb02 = src0 ? src0->nb[2] : 0;
|
|
|
|
const uint64_t nb03 = src0 ? src0->nb[3] : 0;
|
|
|
|
|
|
|
|
const int64_t ne10 = src1 ? src1->ne[0] : 0;
|
|
|
|
const int64_t ne11 = src1 ? src1->ne[1] : 0;
|
|
|
|
const int64_t ne12 = src1 ? src1->ne[2] : 0;
|
|
|
|
const int64_t ne13 = src1 ? src1->ne[3] : 0; UNUSED(ne13);
|
|
|
|
|
|
|
|
const uint64_t nb10 = src1 ? src1->nb[0] : 0;
|
|
|
|
const uint64_t nb11 = src1 ? src1->nb[1] : 0;
|
|
|
|
const uint64_t nb12 = src1 ? src1->nb[2] : 0;
|
|
|
|
const uint64_t nb13 = src1 ? src1->nb[3] : 0; UNUSED(nb13);
|
|
|
|
|
|
|
|
const int64_t ne0 = dst ? dst->ne[0] : 0;
|
|
|
|
const int64_t ne1 = dst ? dst->ne[1] : 0;
|
|
|
|
const int64_t ne2 = dst ? dst->ne[2] : 0;
|
|
|
|
const int64_t ne3 = dst ? dst->ne[3] : 0;
|
|
|
|
|
|
|
|
const uint64_t nb0 = dst ? dst->nb[0] : 0;
|
|
|
|
const uint64_t nb1 = dst ? dst->nb[1] : 0;
|
|
|
|
const uint64_t nb2 = dst ? dst->nb[2] : 0;
|
|
|
|
const uint64_t nb3 = dst ? dst->nb[3] : 0;
|
|
|
|
|
|
|
|
const enum ggml_type src0t = src0 ? src0->type : GGML_TYPE_COUNT;
|
|
|
|
const enum ggml_type src1t = src1 ? src1->type : GGML_TYPE_COUNT;
|
|
|
|
const enum ggml_type dstt = dst ? dst->type : GGML_TYPE_COUNT;
|
|
|
|
|
2024-01-26 13:16:07 +01:00
|
|
|
id<MTLBuffer> id_src0 = src0 ? ggml_metal_get_buffer(src0, &offs_src0) : nil;
|
|
|
|
id<MTLBuffer> id_src1 = src1 ? ggml_metal_get_buffer(src1, &offs_src1) : nil;
|
2024-02-19 14:18:09 +01:00
|
|
|
id<MTLBuffer> id_src2 = src2 ? ggml_metal_get_buffer(src2, &offs_src2) : nil;
|
2024-01-26 13:16:07 +01:00
|
|
|
id<MTLBuffer> id_dst = dst ? ggml_metal_get_buffer(dst, &offs_dst) : nil;
|
2024-01-18 10:03:13 +01:00
|
|
|
|
|
|
|
//GGML_METAL_LOG_INFO("%s: op - %s\n", __func__, ggml_op_name(dst->op));
|
|
|
|
//if (src0) {
|
|
|
|
// GGML_METAL_LOG_INFO("%s: src0 - %4s [%5lld, %5lld, %5lld], %d, %s\n", __func__, ggml_type_name(src0t), ne00, ne01, ne02,
|
|
|
|
// ggml_is_contiguous(src0), src0->name);
|
|
|
|
//}
|
|
|
|
//if (src1) {
|
|
|
|
// GGML_METAL_LOG_INFO("%s: src1 - %4s [%5lld, %5lld, %5lld], %d, %s\n", __func__, ggml_type_name(src1t), ne10, ne11, ne12,
|
|
|
|
// ggml_is_contiguous(src1), src1->name);
|
|
|
|
//}
|
|
|
|
//if (dst) {
|
|
|
|
// GGML_METAL_LOG_INFO("%s: dst - %4s [%5lld, %5lld, %5lld], 1, %s\n", __func__, ggml_type_name(dstt), ne0, ne1, ne2,
|
|
|
|
// dst->name);
|
|
|
|
//}
|
|
|
|
|
|
|
|
switch (dst->op) {
|
|
|
|
case GGML_OP_CONCAT:
|
|
|
|
{
|
|
|
|
const int64_t nb = ne00;
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CONCAT].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:3];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:4];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:5];
|
|
|
|
[encoder setBytes:&ne03 length:sizeof(ne03) atIndex:6];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(nb00) atIndex:7];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:8];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:9];
|
|
|
|
[encoder setBytes:&nb03 length:sizeof(nb03) atIndex:10];
|
|
|
|
[encoder setBytes:&ne10 length:sizeof(ne10) atIndex:11];
|
|
|
|
[encoder setBytes:&ne11 length:sizeof(ne11) atIndex:12];
|
|
|
|
[encoder setBytes:&ne12 length:sizeof(ne12) atIndex:13];
|
|
|
|
[encoder setBytes:&ne13 length:sizeof(ne13) atIndex:14];
|
|
|
|
[encoder setBytes:&nb10 length:sizeof(nb10) atIndex:15];
|
|
|
|
[encoder setBytes:&nb11 length:sizeof(nb11) atIndex:16];
|
|
|
|
[encoder setBytes:&nb12 length:sizeof(nb12) atIndex:17];
|
|
|
|
[encoder setBytes:&nb13 length:sizeof(nb13) atIndex:18];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:19];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:20];
|
|
|
|
[encoder setBytes:&ne2 length:sizeof(ne2) atIndex:21];
|
|
|
|
[encoder setBytes:&ne3 length:sizeof(ne3) atIndex:22];
|
|
|
|
[encoder setBytes:&nb0 length:sizeof(nb0) atIndex:23];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(nb1) atIndex:24];
|
|
|
|
[encoder setBytes:&nb2 length:sizeof(nb2) atIndex:25];
|
|
|
|
[encoder setBytes:&nb3 length:sizeof(nb3) atIndex:26];
|
|
|
|
[encoder setBytes:&nb length:sizeof(nb) atIndex:27];
|
|
|
|
|
|
|
|
const int nth = MIN(1024, ne0);
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne1, ne2, ne3) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_ADD:
|
|
|
|
case GGML_OP_MUL:
|
|
|
|
case GGML_OP_DIV:
|
|
|
|
{
|
|
|
|
const size_t offs = 0;
|
|
|
|
|
|
|
|
bool bcast_row = false;
|
|
|
|
|
|
|
|
int64_t nb = ne00;
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
if (ggml_nelements(src1) == ne10 && ggml_is_contiguous(src1) && ne00 % 4 == 0 && ne10 % 4 == 0) {
|
2023-12-13 20:55:03 +01:00
|
|
|
GGML_ASSERT(ggml_is_contiguous(src0));
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
// src1 is a row
|
|
|
|
GGML_ASSERT(ne11 == 1);
|
2023-12-13 20:55:03 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
nb = ne00 / 4;
|
|
|
|
switch (dst->op) {
|
|
|
|
case GGML_OP_ADD: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_ADD_ROW].pipeline; break;
|
|
|
|
case GGML_OP_MUL: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_ROW].pipeline; break;
|
|
|
|
case GGML_OP_DIV: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_DIV_ROW].pipeline; break;
|
|
|
|
default: GGML_ASSERT(false);
|
2023-12-13 20:55:03 +01:00
|
|
|
}
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
bcast_row = true;
|
|
|
|
} else {
|
|
|
|
switch (dst->op) {
|
|
|
|
case GGML_OP_ADD: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_ADD].pipeline; break;
|
|
|
|
case GGML_OP_MUL: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL].pipeline; break;
|
|
|
|
case GGML_OP_DIV: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_DIV].pipeline; break;
|
|
|
|
default: GGML_ASSERT(false);
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
2024-01-18 10:03:13 +01:00
|
|
|
}
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:3];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:4];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:5];
|
|
|
|
[encoder setBytes:&ne03 length:sizeof(ne03) atIndex:6];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(nb00) atIndex:7];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:8];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:9];
|
|
|
|
[encoder setBytes:&nb03 length:sizeof(nb03) atIndex:10];
|
|
|
|
[encoder setBytes:&ne10 length:sizeof(ne10) atIndex:11];
|
|
|
|
[encoder setBytes:&ne11 length:sizeof(ne11) atIndex:12];
|
|
|
|
[encoder setBytes:&ne12 length:sizeof(ne12) atIndex:13];
|
|
|
|
[encoder setBytes:&ne13 length:sizeof(ne13) atIndex:14];
|
|
|
|
[encoder setBytes:&nb10 length:sizeof(nb10) atIndex:15];
|
|
|
|
[encoder setBytes:&nb11 length:sizeof(nb11) atIndex:16];
|
|
|
|
[encoder setBytes:&nb12 length:sizeof(nb12) atIndex:17];
|
|
|
|
[encoder setBytes:&nb13 length:sizeof(nb13) atIndex:18];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:19];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:20];
|
|
|
|
[encoder setBytes:&ne2 length:sizeof(ne2) atIndex:21];
|
|
|
|
[encoder setBytes:&ne3 length:sizeof(ne3) atIndex:22];
|
|
|
|
[encoder setBytes:&nb0 length:sizeof(nb0) atIndex:23];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(nb1) atIndex:24];
|
|
|
|
[encoder setBytes:&nb2 length:sizeof(nb2) atIndex:25];
|
|
|
|
[encoder setBytes:&nb3 length:sizeof(nb3) atIndex:26];
|
|
|
|
[encoder setBytes:&offs length:sizeof(offs) atIndex:27];
|
|
|
|
[encoder setBytes:&nb length:sizeof(nb) atIndex:28];
|
|
|
|
|
|
|
|
if (bcast_row) {
|
|
|
|
const int64_t n = ggml_nelements(dst)/4;
|
2023-06-25 14:40:30 +02:00
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
} else {
|
|
|
|
const int nth = MIN((int) pipeline.maxTotalThreadsPerThreadgroup, ne0);
|
2023-12-13 20:55:03 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
}
|
|
|
|
} break;
|
|
|
|
case GGML_OP_ACC:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(src0t == GGML_TYPE_F32);
|
|
|
|
GGML_ASSERT(src1t == GGML_TYPE_F32);
|
|
|
|
GGML_ASSERT(dstt == GGML_TYPE_F32);
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
GGML_ASSERT(ggml_is_contiguous(src0));
|
|
|
|
GGML_ASSERT(ggml_is_contiguous(src1));
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const size_t pnb1 = ((int32_t *) dst->op_params)[0];
|
|
|
|
const size_t pnb2 = ((int32_t *) dst->op_params)[1];
|
|
|
|
const size_t pnb3 = ((int32_t *) dst->op_params)[2];
|
|
|
|
const size_t offs = ((int32_t *) dst->op_params)[3];
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const bool inplace = (bool) ((int32_t *) dst->op_params)[4];
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
if (!inplace) {
|
|
|
|
// run a separete kernel to cpy src->dst
|
|
|
|
// not sure how to avoid this
|
|
|
|
// TODO: make a simpler cpy_bytes kernel
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F32_F32].pipeline;
|
2024-01-13 17:03:45 +01:00
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof( int64_t) atIndex:3];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof( int64_t) atIndex:4];
|
|
|
|
[encoder setBytes:&ne03 length:sizeof( int64_t) atIndex:5];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(uint64_t) atIndex:6];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:7];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:8];
|
|
|
|
[encoder setBytes:&nb03 length:sizeof(uint64_t) atIndex:9];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof( int64_t) atIndex:10];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof( int64_t) atIndex:11];
|
|
|
|
[encoder setBytes:&ne2 length:sizeof( int64_t) atIndex:12];
|
|
|
|
[encoder setBytes:&ne3 length:sizeof( int64_t) atIndex:13];
|
|
|
|
[encoder setBytes:&nb0 length:sizeof(uint64_t) atIndex:14];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(uint64_t) atIndex:15];
|
|
|
|
[encoder setBytes:&nb2 length:sizeof(uint64_t) atIndex:16];
|
|
|
|
[encoder setBytes:&nb3 length:sizeof(uint64_t) atIndex:17];
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const int nth = MIN((int) pipeline.maxTotalThreadsPerThreadgroup, ne00);
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
}
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_ADD].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:3];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:4];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:5];
|
|
|
|
[encoder setBytes:&ne03 length:sizeof(ne03) atIndex:6];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(nb00) atIndex:7];
|
|
|
|
[encoder setBytes:&pnb1 length:sizeof(pnb1) atIndex:8];
|
|
|
|
[encoder setBytes:&pnb2 length:sizeof(pnb2) atIndex:9];
|
|
|
|
[encoder setBytes:&pnb3 length:sizeof(pnb3) atIndex:10];
|
|
|
|
[encoder setBytes:&ne10 length:sizeof(ne10) atIndex:11];
|
|
|
|
[encoder setBytes:&ne11 length:sizeof(ne11) atIndex:12];
|
|
|
|
[encoder setBytes:&ne12 length:sizeof(ne12) atIndex:13];
|
|
|
|
[encoder setBytes:&ne13 length:sizeof(ne13) atIndex:14];
|
|
|
|
[encoder setBytes:&nb10 length:sizeof(nb10) atIndex:15];
|
|
|
|
[encoder setBytes:&nb11 length:sizeof(nb11) atIndex:16];
|
|
|
|
[encoder setBytes:&nb12 length:sizeof(nb12) atIndex:17];
|
|
|
|
[encoder setBytes:&nb13 length:sizeof(nb13) atIndex:18];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:19];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:20];
|
|
|
|
[encoder setBytes:&ne2 length:sizeof(ne2) atIndex:21];
|
|
|
|
[encoder setBytes:&ne3 length:sizeof(ne3) atIndex:22];
|
|
|
|
[encoder setBytes:&nb0 length:sizeof(nb0) atIndex:23];
|
|
|
|
[encoder setBytes:&pnb1 length:sizeof(pnb1) atIndex:24];
|
|
|
|
[encoder setBytes:&pnb2 length:sizeof(pnb2) atIndex:25];
|
|
|
|
[encoder setBytes:&pnb3 length:sizeof(pnb3) atIndex:26];
|
|
|
|
[encoder setBytes:&offs length:sizeof(offs) atIndex:27];
|
|
|
|
|
|
|
|
const int nth = MIN((int) pipeline.maxTotalThreadsPerThreadgroup, ne00);
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne11, ne12, ne13) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_SCALE:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(ggml_is_contiguous(src0));
|
|
|
|
|
2024-03-03 13:23:52 +01:00
|
|
|
float scale;
|
|
|
|
memcpy(&scale, dst->op_params, sizeof(scale));
|
2024-01-18 10:03:13 +01:00
|
|
|
|
|
|
|
int64_t n = ggml_nelements(dst);
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
if (n % 4 == 0) {
|
|
|
|
n /= 4;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SCALE_4].pipeline;
|
|
|
|
} else {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SCALE].pipeline;
|
|
|
|
}
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&scale length:sizeof(scale) atIndex:2];
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
|
|
|
} break;
|
2024-04-14 13:14:19 +02:00
|
|
|
case GGML_OP_CLAMP:
|
2024-05-11 15:57:53 +02:00
|
|
|
{
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CLAMP].pipeline;
|
2024-04-14 13:14:19 +02:00
|
|
|
|
2024-05-11 15:57:53 +02:00
|
|
|
float min;
|
|
|
|
float max;
|
|
|
|
memcpy(&min, ((int32_t *) dst->op_params) + 0, sizeof(float));
|
|
|
|
memcpy(&max, ((int32_t *) dst->op_params) + 1, sizeof(float));
|
2024-04-14 13:14:19 +02:00
|
|
|
|
2024-05-11 15:57:53 +02:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&min length:sizeof(min) atIndex:2];
|
|
|
|
[encoder setBytes:&max length:sizeof(max) atIndex:3];
|
2024-04-14 13:14:19 +02:00
|
|
|
|
2024-05-11 15:57:53 +02:00
|
|
|
const int64_t n = ggml_nelements(dst);
|
2024-04-14 13:14:19 +02:00
|
|
|
|
2024-05-11 15:57:53 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
|
|
|
} break;
|
2024-01-18 10:03:13 +01:00
|
|
|
case GGML_OP_UNARY:
|
|
|
|
switch (ggml_get_unary_op(gf->nodes[i])) {
|
2024-04-16 17:40:48 +02:00
|
|
|
// we are not taking into account the strides, so for now require contiguous tensors
|
|
|
|
GGML_ASSERT(ggml_is_contiguous(src0));
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
case GGML_UNARY_OP_TANH:
|
|
|
|
{
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_TANH].pipeline;
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const int64_t n = ggml_nelements(dst);
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_UNARY_OP_RELU:
|
|
|
|
{
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_RELU].pipeline;
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const int64_t n = ggml_nelements(dst);
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-05-01 23:44:26 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_UNARY_OP_SIGMOID:
|
|
|
|
{
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SIGMOID].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
|
|
|
|
const int64_t n = ggml_nelements(dst);
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_UNARY_OP_GELU:
|
|
|
|
{
|
2024-04-16 17:40:48 +02:00
|
|
|
int64_t n = ggml_nelements(dst);
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
if (n % 4 == 0) {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GELU_4].pipeline;
|
|
|
|
n /= 4;
|
|
|
|
} else {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GELU].pipeline;
|
|
|
|
}
|
2024-01-13 17:03:45 +01:00
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-04-16 17:40:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
} break;
|
|
|
|
case GGML_UNARY_OP_GELU_QUICK:
|
|
|
|
{
|
2024-04-16 17:40:48 +02:00
|
|
|
int64_t n = ggml_nelements(dst);
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
if (n % 4 == 0) {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GELU_QUICK_4].pipeline;
|
|
|
|
n /= 4;
|
|
|
|
} else {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GELU_QUICK].pipeline;
|
|
|
|
}
|
2024-01-13 17:03:45 +01:00
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
2023-12-13 20:55:03 +01:00
|
|
|
|
2024-04-16 17:40:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
} break;
|
|
|
|
case GGML_UNARY_OP_SILU:
|
|
|
|
{
|
2024-04-16 17:40:48 +02:00
|
|
|
int64_t n = ggml_nelements(dst);
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
if (n % 4 == 0) {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SILU_4].pipeline;
|
|
|
|
n /= 4;
|
|
|
|
} else {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SILU].pipeline;
|
|
|
|
}
|
2024-01-02 20:07:47 +01:00
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
2023-12-13 20:55:03 +01:00
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-04-16 17:40:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
} break;
|
|
|
|
default:
|
|
|
|
{
|
|
|
|
GGML_METAL_LOG_WARN("%s: node %3d, op = %8s not implemented\n", __func__, i, ggml_op_name(dst->op));
|
|
|
|
GGML_ASSERT(false);
|
|
|
|
}
|
|
|
|
} break;
|
|
|
|
case GGML_OP_SQR:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(ggml_is_contiguous(src0));
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SQR].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
|
|
|
|
const int64_t n = ggml_nelements(dst);
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_SUM_ROWS:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(src0->nb[0] == ggml_type_size(src0->type));
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SUM_ROWS].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
|
|
|
|
[encoder setBytes:&ne03 length:sizeof(ne03) atIndex:5];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(nb00) atIndex:6];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:7];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:8];
|
|
|
|
[encoder setBytes:&nb03 length:sizeof(nb03) atIndex:9];
|
|
|
|
[encoder setBytes:&ne10 length:sizeof(ne10) atIndex:10];
|
|
|
|
[encoder setBytes:&ne11 length:sizeof(ne11) atIndex:11];
|
|
|
|
[encoder setBytes:&ne12 length:sizeof(ne12) atIndex:12];
|
|
|
|
[encoder setBytes:&ne13 length:sizeof(ne13) atIndex:13];
|
|
|
|
[encoder setBytes:&nb10 length:sizeof(nb10) atIndex:14];
|
|
|
|
[encoder setBytes:&nb11 length:sizeof(nb11) atIndex:15];
|
|
|
|
[encoder setBytes:&nb12 length:sizeof(nb12) atIndex:16];
|
|
|
|
[encoder setBytes:&nb13 length:sizeof(nb13) atIndex:17];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:18];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:19];
|
|
|
|
[encoder setBytes:&ne2 length:sizeof(ne2) atIndex:20];
|
|
|
|
[encoder setBytes:&ne3 length:sizeof(ne3) atIndex:21];
|
|
|
|
[encoder setBytes:&nb0 length:sizeof(nb0) atIndex:22];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(nb1) atIndex:23];
|
|
|
|
[encoder setBytes:&nb2 length:sizeof(nb2) atIndex:24];
|
|
|
|
[encoder setBytes:&nb3 length:sizeof(nb3) atIndex:25];
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_SOFT_MAX:
|
|
|
|
{
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
GGML_ASSERT(!src1 || src1->type == GGML_TYPE_F16 || src1->type == GGML_TYPE_F32);
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
int nth = 32; // SIMD width
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
2024-05-11 09:32:41 +02:00
|
|
|
const bool use_f16 = (src1 && src1->type == GGML_TYPE_F16);
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
if (ne00%4 == 0) {
|
2024-05-13 10:01:07 +02:00
|
|
|
while (nth < ne00/4 && nth*ne01*ne02*ne03 < 256) {
|
2023-12-07 21:27:19 +01:00
|
|
|
nth *= 2;
|
|
|
|
}
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
if (use_f16) {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SOFT_MAX_F16_4].pipeline;
|
|
|
|
} else {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SOFT_MAX_F32_4].pipeline;
|
|
|
|
}
|
2024-01-18 10:03:13 +01:00
|
|
|
} else {
|
2024-05-13 10:01:07 +02:00
|
|
|
while (nth < ne00 && nth*ne01*ne02*ne03 < 256) {
|
2024-01-18 10:03:13 +01:00
|
|
|
nth *= 2;
|
|
|
|
}
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
if (use_f16) {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SOFT_MAX_F16].pipeline;
|
|
|
|
} else {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SOFT_MAX_F32].pipeline;
|
|
|
|
}
|
2024-01-18 10:03:13 +01:00
|
|
|
}
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-03-03 13:23:52 +01:00
|
|
|
float scale;
|
|
|
|
float max_bias;
|
|
|
|
|
|
|
|
memcpy(&scale, ((int32_t *) dst->op_params) + 0, sizeof(scale));
|
|
|
|
memcpy(&max_bias, ((int32_t *) dst->op_params) + 1, sizeof(max_bias));
|
2024-02-19 14:18:09 +01:00
|
|
|
|
|
|
|
const int64_t nrows_x = ggml_nrows(src0);
|
|
|
|
const int64_t nrows_y = src0->ne[1];
|
2024-03-03 13:23:52 +01:00
|
|
|
|
2024-05-11 09:32:41 +02:00
|
|
|
const uint32_t n_head = nrows_x/nrows_y;
|
|
|
|
const uint32_t n_head_log2 = 1u << (uint32_t) floorf(log2f((float) n_head));
|
2024-02-19 14:18:09 +01:00
|
|
|
|
|
|
|
const float m0 = powf(2.0f, -(max_bias ) / n_head_log2);
|
|
|
|
const float m1 = powf(2.0f, -(max_bias / 2.0f) / n_head_log2);
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
if (id_src1) {
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
} else {
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:1];
|
|
|
|
}
|
2024-05-11 09:32:41 +02:00
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:3];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:4];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:5];
|
|
|
|
[encoder setBytes:&scale length:sizeof(scale) atIndex:6];
|
|
|
|
[encoder setBytes:&max_bias length:sizeof(max_bias) atIndex:7];
|
|
|
|
[encoder setBytes:&m0 length:sizeof(m0) atIndex:8];
|
|
|
|
[encoder setBytes:&m1 length:sizeof(m1) atIndex:9];
|
|
|
|
[encoder setBytes:&n_head_log2 length:sizeof(n_head_log2) atIndex:10];
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setThreadgroupMemoryLength:32*sizeof(float) atIndex:0];
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne01*ne02*ne03, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_DIAG_MASK_INF:
|
|
|
|
{
|
|
|
|
const int n_past = ((int32_t *)(dst->op_params))[0];
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
if (ne00%8 == 0) {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_DIAG_MASK_INF_8].pipeline;
|
|
|
|
} else {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_DIAG_MASK_INF].pipeline;
|
|
|
|
}
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
|
|
|
|
[encoder setBytes:&n_past length:sizeof(int) atIndex:4];
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
if (ne00%8 == 0) {
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne00*ne01*ne02/8, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne00, ne01, ne02) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
|
|
|
}
|
|
|
|
} break;
|
|
|
|
case GGML_OP_MUL_MAT:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(ne00 == ne10);
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
// TODO: assert that dim2 and dim3 are contiguous
|
|
|
|
GGML_ASSERT(ne12 % ne02 == 0);
|
|
|
|
GGML_ASSERT(ne13 % ne03 == 0);
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const uint r2 = ne12/ne02;
|
|
|
|
const uint r3 = ne13/ne03;
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
// find the break-even point where the matrix-matrix kernel becomes more efficient compared
|
|
|
|
// to the matrix-vector kernel
|
|
|
|
int ne11_mm_min = 1;
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
#if 0
|
|
|
|
// the numbers below are measured on M2 Ultra for 7B and 13B models
|
|
|
|
// these numbers do not translate to other devices or model sizes
|
|
|
|
// TODO: need to find a better approach
|
|
|
|
if ([ctx->device.name isEqualToString:@"Apple M2 Ultra"]) {
|
|
|
|
switch (src0t) {
|
|
|
|
case GGML_TYPE_F16: ne11_mm_min = 2; break;
|
|
|
|
case GGML_TYPE_Q8_0: ne11_mm_min = 7; break;
|
|
|
|
case GGML_TYPE_Q2_K: ne11_mm_min = 15; break;
|
|
|
|
case GGML_TYPE_Q3_K: ne11_mm_min = 7; break;
|
|
|
|
case GGML_TYPE_Q4_0:
|
|
|
|
case GGML_TYPE_Q4_1: ne11_mm_min = 15; break;
|
|
|
|
case GGML_TYPE_Q4_K: ne11_mm_min = 11; break;
|
|
|
|
case GGML_TYPE_Q5_0: // not tested yet
|
|
|
|
case GGML_TYPE_Q5_1: ne11_mm_min = 13; break; // not tested yet
|
|
|
|
case GGML_TYPE_Q5_K: ne11_mm_min = 7; break;
|
|
|
|
case GGML_TYPE_Q6_K: ne11_mm_min = 7; break;
|
|
|
|
default: ne11_mm_min = 1; break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#endif
|
2023-09-05 12:54:40 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
// for now the matrix-matrix multiplication kernel only works on A14+/M1+ SoCs
|
|
|
|
// AMD GPU and older A-chips will reuse matrix-vector multiplication kernel
|
|
|
|
if ([ctx->device supportsFamily:MTLGPUFamilyApple7] &&
|
|
|
|
!ggml_is_transposed(src0) &&
|
|
|
|
!ggml_is_transposed(src1) &&
|
|
|
|
src1t == GGML_TYPE_F32 &&
|
|
|
|
ne00 % 32 == 0 && ne00 >= 64 &&
|
|
|
|
(ne11 > ne11_mm_min || (ggml_is_quantized(src0t) && ne12 > 1))) {
|
|
|
|
//printf("matrix: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12);
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-03-27 17:55:10 +01:00
|
|
|
// some Metal matrix data types require aligned pointers
|
|
|
|
// ref: https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf (Table 2.5)
|
|
|
|
switch (src0->type) {
|
|
|
|
case GGML_TYPE_F32: GGML_ASSERT(nb01 % 16 == 0); break;
|
|
|
|
case GGML_TYPE_F16: GGML_ASSERT(nb01 % 8 == 0); break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
2023-11-03 20:35:05 +01:00
|
|
|
switch (src0->type) {
|
2024-01-18 10:03:13 +01:00
|
|
|
case GGML_TYPE_F32: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_F32_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_F16: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_F16_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_Q4_0_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_1: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_Q4_1_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_Q5_0_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_1: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_Q5_1_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q8_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_Q8_0_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q2_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_Q2_K_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q3_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_Q3_K_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_Q4_K_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_Q5_K_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q6_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_Q6_K_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_IQ2_XXS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_XXS_F32].pipeline; break;
|
|
|
|
case GGML_TYPE_IQ2_XS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_XS_F32 ].pipeline; break;
|
2024-01-30 14:14:12 +01:00
|
|
|
case GGML_TYPE_IQ3_XXS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_XXS_F32].pipeline; break;
|
2024-02-24 15:23:52 +01:00
|
|
|
case GGML_TYPE_IQ3_S: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_S_F32 ].pipeline; break;
|
2024-02-26 17:28:38 +01:00
|
|
|
case GGML_TYPE_IQ2_S: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_S_F32 ].pipeline; break;
|
2024-02-18 17:16:55 +01:00
|
|
|
case GGML_TYPE_IQ1_S: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ1_S_F32 ].pipeline; break;
|
2024-03-27 17:55:10 +01:00
|
|
|
case GGML_TYPE_IQ1_M: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ1_M_F32 ].pipeline; break;
|
2024-02-21 15:19:39 +01:00
|
|
|
case GGML_TYPE_IQ4_NL: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_NL_F32 ].pipeline; break;
|
2024-02-27 15:34:24 +01:00
|
|
|
case GGML_TYPE_IQ4_XS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_XS_F32 ].pipeline; break;
|
2024-01-18 10:03:13 +01:00
|
|
|
default: GGML_ASSERT(false && "MUL MAT-MAT not implemented");
|
|
|
|
}
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:3];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:5];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:6];
|
|
|
|
[encoder setBytes:&ne12 length:sizeof(ne12) atIndex:7];
|
|
|
|
[encoder setBytes:&nb10 length:sizeof(nb10) atIndex:8];
|
|
|
|
[encoder setBytes:&nb11 length:sizeof(nb11) atIndex:9];
|
|
|
|
[encoder setBytes:&nb12 length:sizeof(nb12) atIndex:10];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:11];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:12];
|
|
|
|
[encoder setBytes:&r2 length:sizeof(r2) atIndex:13];
|
|
|
|
[encoder setBytes:&r3 length:sizeof(r3) atIndex:14];
|
|
|
|
[encoder setThreadgroupMemoryLength:8192 atIndex:0];
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake( (ne11 + 31)/32, (ne01 + 63)/64, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(128, 1, 1)];
|
|
|
|
} else {
|
|
|
|
int nth0 = 32;
|
|
|
|
int nth1 = 1;
|
|
|
|
int nrows = 1;
|
|
|
|
//printf("vector: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12);
|
2023-11-12 14:31:08 +01:00
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
// use custom matrix x vector kernel
|
|
|
|
switch (src0t) {
|
|
|
|
case GGML_TYPE_F32:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(src1t == GGML_TYPE_F32);
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_F32_F32].pipeline;
|
|
|
|
nrows = 4;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_F16:
|
|
|
|
{
|
|
|
|
nth0 = 32;
|
|
|
|
nth1 = 1;
|
|
|
|
if (src1t == GGML_TYPE_F32) {
|
|
|
|
if (ne11 * ne12 < 4) {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F32_1ROW].pipeline;
|
|
|
|
} else if (ne00 >= 128 && ne01 >= 8 && ne00%4 == 0) {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F32_L4].pipeline;
|
|
|
|
nrows = ne11;
|
|
|
|
} else {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F32].pipeline;
|
|
|
|
nrows = 4;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_F16_F16].pipeline;
|
|
|
|
nrows = 4;
|
|
|
|
}
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q4_0:
|
|
|
|
{
|
|
|
|
nth0 = 8;
|
|
|
|
nth1 = 8;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_Q4_0_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q4_1:
|
|
|
|
{
|
|
|
|
nth0 = 8;
|
|
|
|
nth1 = 8;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_Q4_1_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q5_0:
|
|
|
|
{
|
|
|
|
nth0 = 8;
|
|
|
|
nth1 = 8;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_Q5_0_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q5_1:
|
|
|
|
{
|
|
|
|
nth0 = 8;
|
|
|
|
nth1 = 8;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_Q5_1_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q8_0:
|
|
|
|
{
|
|
|
|
nth0 = 8;
|
|
|
|
nth1 = 8;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_Q8_0_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q2_K:
|
|
|
|
{
|
|
|
|
nth0 = 2;
|
|
|
|
nth1 = 32;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_Q2_K_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q3_K:
|
|
|
|
{
|
|
|
|
nth0 = 2;
|
|
|
|
nth1 = 32;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_Q3_K_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q4_K:
|
|
|
|
{
|
|
|
|
nth0 = 4; //1;
|
|
|
|
nth1 = 8; //32;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_Q4_K_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q5_K:
|
|
|
|
{
|
|
|
|
nth0 = 2;
|
|
|
|
nth1 = 32;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_Q5_K_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q6_K:
|
|
|
|
{
|
|
|
|
nth0 = 2;
|
|
|
|
nth1 = 32;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_Q6_K_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_IQ2_XXS:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_XXS_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_IQ2_XS:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_XS_F32].pipeline;
|
|
|
|
} break;
|
2024-01-30 14:14:12 +01:00
|
|
|
case GGML_TYPE_IQ3_XXS:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ3_XXS_F32].pipeline;
|
|
|
|
} break;
|
2024-02-24 15:23:52 +01:00
|
|
|
case GGML_TYPE_IQ3_S:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ3_S_F32].pipeline;
|
|
|
|
} break;
|
2024-02-26 17:28:38 +01:00
|
|
|
case GGML_TYPE_IQ2_S:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_S_F32].pipeline;
|
|
|
|
} break;
|
2024-02-18 17:16:55 +01:00
|
|
|
case GGML_TYPE_IQ1_S:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ1_S_F32].pipeline;
|
|
|
|
} break;
|
2024-03-27 17:55:10 +01:00
|
|
|
case GGML_TYPE_IQ1_M:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ1_M_F32].pipeline;
|
|
|
|
} break;
|
2024-02-21 15:19:39 +01:00
|
|
|
case GGML_TYPE_IQ4_NL:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_NL_F32].pipeline;
|
|
|
|
} break;
|
2024-02-27 15:34:24 +01:00
|
|
|
case GGML_TYPE_IQ4_XS:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_XS_F32].pipeline;
|
|
|
|
} break;
|
2024-01-18 10:03:13 +01:00
|
|
|
default:
|
|
|
|
{
|
|
|
|
GGML_METAL_LOG_ERROR("Asserting on type %d\n", (int)src0t);
|
|
|
|
GGML_ASSERT(false && "not implemented");
|
|
|
|
}
|
2023-11-12 14:31:08 +01:00
|
|
|
};
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
if (ggml_is_quantized(src0t)) {
|
|
|
|
GGML_ASSERT(ne00 >= nth0*nth1);
|
|
|
|
}
|
2024-01-13 17:03:45 +01:00
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
2023-12-13 20:55:03 +01:00
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:3];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:4];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:5];
|
2023-12-13 20:55:03 +01:00
|
|
|
[encoder setBytes:&nb00 length:sizeof(nb00) atIndex:6];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:7];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:8];
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setBytes:&ne10 length:sizeof(ne10) atIndex:9];
|
|
|
|
[encoder setBytes:&ne11 length:sizeof(ne11) atIndex:10];
|
|
|
|
[encoder setBytes:&ne12 length:sizeof(ne12) atIndex:11];
|
|
|
|
[encoder setBytes:&nb10 length:sizeof(nb10) atIndex:12];
|
|
|
|
[encoder setBytes:&nb11 length:sizeof(nb11) atIndex:13];
|
|
|
|
[encoder setBytes:&nb12 length:sizeof(nb12) atIndex:14];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:15];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:16];
|
|
|
|
[encoder setBytes:&r2 length:sizeof(r2) atIndex:17];
|
|
|
|
[encoder setBytes:&r3 length:sizeof(r3) atIndex:18];
|
|
|
|
|
2024-03-27 17:55:10 +01:00
|
|
|
if (src0t == GGML_TYPE_Q4_0 || src0t == GGML_TYPE_Q4_1 || src0t == GGML_TYPE_Q5_0 ||
|
|
|
|
src0t == GGML_TYPE_Q5_1 || src0t == GGML_TYPE_Q8_0 || src0t == GGML_TYPE_Q2_K ||
|
|
|
|
src0t == GGML_TYPE_IQ1_S || src0t == GGML_TYPE_IQ1_M || src0t == GGML_TYPE_IQ2_S) {
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 7)/8, ne11, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
|
|
|
}
|
|
|
|
else if (src0t == GGML_TYPE_IQ2_XXS || src0t == GGML_TYPE_IQ2_XS) {
|
|
|
|
const int mem_size = src0t == GGML_TYPE_IQ2_XXS ? 256*8+128 : 512*8+128;
|
|
|
|
[encoder setThreadgroupMemoryLength:mem_size atIndex:0];
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 7)/8, ne11, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
|
|
|
}
|
2024-02-24 15:23:52 +01:00
|
|
|
else if (src0t == GGML_TYPE_IQ3_XXS || src0t == GGML_TYPE_IQ3_S) {
|
|
|
|
const int mem_size = src0t == GGML_TYPE_IQ3_XXS ? 256*4+128 : 512*4;
|
2024-01-30 14:14:12 +01:00
|
|
|
[encoder setThreadgroupMemoryLength:mem_size atIndex:0];
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 7)/8, ne11, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
|
|
|
}
|
2024-02-27 15:34:24 +01:00
|
|
|
else if (src0t == GGML_TYPE_IQ4_NL || src0t == GGML_TYPE_IQ4_XS) {
|
2024-02-21 15:19:39 +01:00
|
|
|
const int mem_size = 32*sizeof(float);
|
|
|
|
[encoder setThreadgroupMemoryLength:mem_size atIndex:0];
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 3)/4, ne11, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
|
|
|
}
|
2024-01-18 10:03:13 +01:00
|
|
|
else if (src0t == GGML_TYPE_Q4_K) {
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 3)/4, ne11, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
|
|
|
}
|
|
|
|
else if (src0t == GGML_TYPE_Q3_K) {
|
|
|
|
#ifdef GGML_QKK_64
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 1)/2, ne11, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
|
|
|
#else
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 3)/4, ne11, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
else if (src0t == GGML_TYPE_Q5_K) {
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 3)/4, ne11, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
|
|
|
}
|
|
|
|
else if (src0t == GGML_TYPE_Q6_K) {
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 1)/2, ne11, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
|
|
|
} else {
|
|
|
|
const int64_t ny = (ne11 + nrows - 1)/nrows;
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ny, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} break;
|
|
|
|
case GGML_OP_MUL_MAT_ID:
|
|
|
|
{
|
2024-04-03 15:07:05 +02:00
|
|
|
const int n_as = src0->ne[2];
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-04-03 15:07:05 +02:00
|
|
|
// src2 = ids
|
2024-04-18 15:18:48 +02:00
|
|
|
const int64_t ne20 = src2->ne[0];
|
2024-04-03 15:07:05 +02:00
|
|
|
const int64_t ne21 = src2->ne[1];
|
|
|
|
const int64_t ne22 = src2->ne[2]; GGML_UNUSED(ne22);
|
|
|
|
const int64_t ne23 = src2->ne[3]; GGML_UNUSED(ne23);
|
|
|
|
|
|
|
|
const uint64_t nb20 = src2->nb[0]; GGML_UNUSED(nb20);
|
|
|
|
const uint64_t nb21 = src2->nb[1];
|
|
|
|
const uint64_t nb22 = src2->nb[2]; GGML_UNUSED(nb22);
|
|
|
|
const uint64_t nb23 = src2->nb[3]; GGML_UNUSED(nb23);
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-04-03 15:07:05 +02:00
|
|
|
const enum ggml_type src2t = src2->type; GGML_UNUSED(src2t);
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-04-03 15:07:05 +02:00
|
|
|
GGML_ASSERT(src2t == GGML_TYPE_I32);
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-04-03 15:07:05 +02:00
|
|
|
GGML_ASSERT(!ggml_is_transposed(src0));
|
2024-01-18 10:03:13 +01:00
|
|
|
GGML_ASSERT(!ggml_is_transposed(src1));
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
GGML_ASSERT(src1t == GGML_TYPE_F32);
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
// find the break-even point where the matrix-matrix kernel becomes more efficient compared
|
|
|
|
// to the matrix-vector kernel
|
2024-04-18 15:18:48 +02:00
|
|
|
// ne20 = n_used_experts
|
|
|
|
// ne21 = n_rows
|
|
|
|
const int dst_rows = ne20*ne21;
|
|
|
|
const int dst_rows_min = n_as;
|
2023-12-13 20:55:03 +01:00
|
|
|
|
2024-04-18 15:18:48 +02:00
|
|
|
// max size of the rowids array in the kernel shared buffer
|
|
|
|
GGML_ASSERT(dst_rows <= 2048);
|
2024-01-13 17:03:45 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
// for now the matrix-matrix multiplication kernel only works on A14+/M1+ SoCs
|
|
|
|
// AMD GPU and older A-chips will reuse matrix-vector multiplication kernel
|
|
|
|
// !!!
|
|
|
|
// TODO: for now, always use mat-vec kernels until we figure out how to improve the
|
|
|
|
// indirect matrix multiplication
|
|
|
|
// !!!
|
|
|
|
if ([ctx->device supportsFamily:MTLGPUFamilyApple7] &&
|
2024-04-03 15:07:05 +02:00
|
|
|
ne00 % 32 == 0 && ne00 >= 64 &&
|
2024-04-18 15:18:48 +02:00
|
|
|
dst_rows > dst_rows_min) {
|
2023-12-13 20:55:03 +01:00
|
|
|
|
2024-03-27 17:55:10 +01:00
|
|
|
// some Metal matrix data types require aligned pointers
|
|
|
|
// ref: https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf (Table 2.5)
|
|
|
|
switch (src0->type) {
|
|
|
|
case GGML_TYPE_F32: GGML_ASSERT(nb01 % 16 == 0); break;
|
|
|
|
case GGML_TYPE_F16: GGML_ASSERT(nb01 % 8 == 0); break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
2023-12-13 20:55:03 +01:00
|
|
|
|
2024-04-03 15:07:05 +02:00
|
|
|
switch (src0->type) {
|
2024-01-18 10:03:13 +01:00
|
|
|
case GGML_TYPE_F32: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_F32_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_F16: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_F16_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q4_0_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_1: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q4_1_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q5_0_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_1: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q5_1_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q8_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q8_0_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q2_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q2_K_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q3_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q3_K_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q4_K_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q5_K_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q6_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_Q6_K_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_IQ2_XXS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_XXS_F32].pipeline; break;
|
|
|
|
case GGML_TYPE_IQ2_XS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_XS_F32 ].pipeline; break;
|
2024-01-30 14:14:12 +01:00
|
|
|
case GGML_TYPE_IQ3_XXS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ3_XXS_F32].pipeline; break;
|
2024-02-24 15:23:52 +01:00
|
|
|
case GGML_TYPE_IQ3_S: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ3_S_F32 ].pipeline; break;
|
2024-02-26 17:28:38 +01:00
|
|
|
case GGML_TYPE_IQ2_S: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_S_F32 ].pipeline; break;
|
2024-02-18 17:16:55 +01:00
|
|
|
case GGML_TYPE_IQ1_S: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ1_S_F32 ].pipeline; break;
|
2024-03-27 17:55:10 +01:00
|
|
|
case GGML_TYPE_IQ1_M: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ1_M_F32 ].pipeline; break;
|
2024-02-21 15:19:39 +01:00
|
|
|
case GGML_TYPE_IQ4_NL: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_NL_F32 ].pipeline; break;
|
2024-02-27 15:34:24 +01:00
|
|
|
case GGML_TYPE_IQ4_XS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_XS_F32 ].pipeline; break;
|
2024-01-18 10:03:13 +01:00
|
|
|
default: GGML_ASSERT(false && "MUL_MAT_ID not implemented");
|
|
|
|
}
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
|
2024-04-03 15:07:05 +02:00
|
|
|
[encoder setBuffer:id_src2 offset:offs_src2 atIndex:3];
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder setBytes:&ne20 length:sizeof(ne20) atIndex:4];
|
|
|
|
[encoder setBytes:&ne21 length:sizeof(ne21) atIndex:5];
|
|
|
|
[encoder setBytes:&nb21 length:sizeof(nb21) atIndex:6];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:7];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:8];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:9];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:10];
|
|
|
|
[encoder setBytes:&ne11 length:sizeof(ne11) atIndex:11];
|
|
|
|
[encoder setBytes:&ne12 length:sizeof(ne12) atIndex:12];
|
|
|
|
[encoder setBytes:&ne13 length:sizeof(ne13) atIndex:13];
|
|
|
|
[encoder setBytes:&nb10 length:sizeof(nb10) atIndex:14];
|
|
|
|
[encoder setBytes:&nb11 length:sizeof(nb11) atIndex:15];
|
|
|
|
[encoder setBytes:&nb12 length:sizeof(nb12) atIndex:16];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:17];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:18];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(nb1) atIndex:19];
|
|
|
|
|
|
|
|
[encoder setThreadgroupMemoryLength:GGML_PAD(8192 + dst_rows*4/*sizeof(ushort2)*/, 16) atIndex:0];
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne21 + 31)/32, (ne01 + 63)/64, n_as) threadsPerThreadgroup:MTLSizeMake(128, 1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
} else {
|
|
|
|
int nth0 = 32;
|
|
|
|
int nth1 = 1;
|
|
|
|
int nrows = 1;
|
|
|
|
//printf("vector: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12);
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
// use custom matrix x vector kernel
|
2024-04-03 15:07:05 +02:00
|
|
|
switch (src0t) {
|
2023-06-25 14:40:30 +02:00
|
|
|
case GGML_TYPE_F32:
|
|
|
|
{
|
2024-01-18 10:03:13 +01:00
|
|
|
GGML_ASSERT(src1t == GGML_TYPE_F32);
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F32_F32].pipeline;
|
2023-06-25 14:40:30 +02:00
|
|
|
} break;
|
|
|
|
case GGML_TYPE_F16:
|
|
|
|
{
|
2024-01-18 10:03:13 +01:00
|
|
|
GGML_ASSERT(src1t == GGML_TYPE_F32);
|
|
|
|
nth0 = 32;
|
|
|
|
nth1 = 1;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q4_0:
|
|
|
|
{
|
|
|
|
nth0 = 8;
|
|
|
|
nth1 = 8;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q4_0_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q4_1:
|
|
|
|
{
|
|
|
|
nth0 = 8;
|
|
|
|
nth1 = 8;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q4_1_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q5_0:
|
|
|
|
{
|
|
|
|
nth0 = 8;
|
|
|
|
nth1 = 8;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q5_0_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q5_1:
|
|
|
|
{
|
|
|
|
nth0 = 8;
|
|
|
|
nth1 = 8;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q5_1_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q8_0:
|
|
|
|
{
|
|
|
|
nth0 = 8;
|
|
|
|
nth1 = 8;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q8_0_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q2_K:
|
|
|
|
{
|
|
|
|
nth0 = 2;
|
|
|
|
nth1 = 32;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q2_K_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q3_K:
|
|
|
|
{
|
|
|
|
nth0 = 2;
|
|
|
|
nth1 = 32;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q3_K_F32].pipeline;
|
2023-06-25 14:40:30 +02:00
|
|
|
} break;
|
2024-01-18 10:03:13 +01:00
|
|
|
case GGML_TYPE_Q4_K:
|
|
|
|
{
|
|
|
|
nth0 = 4; //1;
|
|
|
|
nth1 = 8; //32;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q4_K_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q5_K:
|
|
|
|
{
|
|
|
|
nth0 = 2;
|
|
|
|
nth1 = 32;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q5_K_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_Q6_K:
|
|
|
|
{
|
|
|
|
nth0 = 2;
|
|
|
|
nth1 = 32;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_Q6_K_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_IQ2_XXS:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_XXS_F32].pipeline;
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_IQ2_XS:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_XS_F32].pipeline;
|
|
|
|
} break;
|
2024-01-30 14:14:12 +01:00
|
|
|
case GGML_TYPE_IQ3_XXS:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ3_XXS_F32].pipeline;
|
|
|
|
} break;
|
2024-02-24 15:23:52 +01:00
|
|
|
case GGML_TYPE_IQ3_S:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ3_S_F32].pipeline;
|
|
|
|
} break;
|
2024-02-26 17:28:38 +01:00
|
|
|
case GGML_TYPE_IQ2_S:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_S_F32].pipeline;
|
|
|
|
} break;
|
2024-02-18 17:16:55 +01:00
|
|
|
case GGML_TYPE_IQ1_S:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ1_S_F32].pipeline;
|
|
|
|
} break;
|
2024-03-27 17:55:10 +01:00
|
|
|
case GGML_TYPE_IQ1_M:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ1_M_F32].pipeline;
|
|
|
|
} break;
|
2024-02-21 15:19:39 +01:00
|
|
|
case GGML_TYPE_IQ4_NL:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_NL_F32].pipeline;
|
|
|
|
} break;
|
2024-02-27 15:34:24 +01:00
|
|
|
case GGML_TYPE_IQ4_XS:
|
|
|
|
{
|
|
|
|
nth0 = 4;
|
|
|
|
nth1 = 16;
|
2024-04-12 18:13:20 +02:00
|
|
|
#if QK_K == 64
|
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_NL_F32].pipeline;
|
|
|
|
#else
|
2024-02-27 15:34:24 +01:00
|
|
|
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_XS_F32].pipeline;
|
2024-04-12 18:13:20 +02:00
|
|
|
#endif
|
|
|
|
|
2024-02-27 15:34:24 +01:00
|
|
|
} break;
|
2024-01-18 10:03:13 +01:00
|
|
|
default:
|
|
|
|
{
|
|
|
|
GGML_METAL_LOG_ERROR("Asserting on type %d\n", (int)src2t);
|
|
|
|
GGML_ASSERT(false && "not implemented");
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
2024-04-03 15:07:05 +02:00
|
|
|
if (ggml_is_quantized(src0t)) {
|
|
|
|
GGML_ASSERT(ne00 >= nth0*nth1);
|
2023-06-25 14:40:30 +02:00
|
|
|
}
|
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
|
2024-04-03 15:07:05 +02:00
|
|
|
[encoder setBuffer:id_src2 offset:offs_src2 atIndex:3];
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder setBytes:&ne20 length:sizeof(ne20) atIndex:4];
|
|
|
|
[encoder setBytes:&ne21 length:sizeof(ne21) atIndex:5];
|
|
|
|
[encoder setBytes:&nb21 length:sizeof(nb21) atIndex:6];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:7];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:8];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:9];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(nb00) atIndex:10];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:11];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:12];
|
|
|
|
[encoder setBytes:&ne10 length:sizeof(ne10) atIndex:13];
|
|
|
|
[encoder setBytes:&ne11 length:sizeof(ne11) atIndex:14];
|
|
|
|
[encoder setBytes:&ne12 length:sizeof(ne12) atIndex:15];
|
|
|
|
[encoder setBytes:&ne13 length:sizeof(ne13) atIndex:16];
|
|
|
|
[encoder setBytes:&nb10 length:sizeof(nb10) atIndex:17];
|
|
|
|
[encoder setBytes:&nb11 length:sizeof(nb11) atIndex:18];
|
|
|
|
[encoder setBytes:&nb12 length:sizeof(nb12) atIndex:19];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:20];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:21];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(nb1) atIndex:22];
|
|
|
|
|
|
|
|
const int64_t _ne1 = 1;
|
|
|
|
const int tgz = dst_rows;
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-04-03 15:07:05 +02:00
|
|
|
if (src0t == GGML_TYPE_Q4_0 || src0t == GGML_TYPE_Q4_1 || src0t == GGML_TYPE_Q5_0 ||
|
|
|
|
src0t == GGML_TYPE_Q5_1 || src0t == GGML_TYPE_Q8_0 || src0t == GGML_TYPE_Q2_K ||
|
|
|
|
src0t == GGML_TYPE_IQ1_S || src0t == GGML_TYPE_IQ1_M || src0t == GGML_TYPE_IQ2_S) {
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 7)/8, _ne1, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
}
|
2024-04-03 15:07:05 +02:00
|
|
|
else if (src0t == GGML_TYPE_IQ2_XXS || src0t == GGML_TYPE_IQ2_XS) {
|
|
|
|
const int mem_size = src0t == GGML_TYPE_IQ2_XXS ? 256*8+128 : 512*8+128;
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setThreadgroupMemoryLength:mem_size atIndex:0];
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 7)/8, _ne1, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
}
|
2024-04-03 15:07:05 +02:00
|
|
|
else if (src0t == GGML_TYPE_IQ3_XXS || src0t == GGML_TYPE_IQ3_S) {
|
|
|
|
const int mem_size = src0t == GGML_TYPE_IQ3_XXS ? 256*4+128 : 512*4;
|
2024-01-30 14:14:12 +01:00
|
|
|
[encoder setThreadgroupMemoryLength:mem_size atIndex:0];
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 7)/8, _ne1, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
2024-01-30 14:14:12 +01:00
|
|
|
}
|
2024-04-03 15:07:05 +02:00
|
|
|
else if (src0t == GGML_TYPE_IQ4_NL || src0t == GGML_TYPE_IQ4_XS) {
|
2024-02-21 15:19:39 +01:00
|
|
|
const int mem_size = 32*sizeof(float);
|
|
|
|
[encoder setThreadgroupMemoryLength:mem_size atIndex:0];
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 3)/4, _ne1, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
2024-02-21 15:19:39 +01:00
|
|
|
}
|
2024-04-03 15:07:05 +02:00
|
|
|
else if (src0t == GGML_TYPE_Q4_K) {
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 3)/4, _ne1, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
}
|
2024-04-03 15:07:05 +02:00
|
|
|
else if (src0t == GGML_TYPE_Q3_K) {
|
2024-01-18 10:03:13 +01:00
|
|
|
#ifdef GGML_QKK_64
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 1)/2, _ne1, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
#else
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 3)/4, _ne1, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
#endif
|
|
|
|
}
|
2024-04-03 15:07:05 +02:00
|
|
|
else if (src0t == GGML_TYPE_Q5_K) {
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 3)/4, _ne1, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
}
|
2024-04-03 15:07:05 +02:00
|
|
|
else if (src0t == GGML_TYPE_Q6_K) {
|
2024-04-18 15:18:48 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 1)/2, _ne1, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
} else {
|
2024-04-18 15:18:48 +02:00
|
|
|
const int64_t ny = (_ne1 + nrows - 1)/nrows; // = _ne1
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ny, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
} break;
|
|
|
|
case GGML_OP_GET_ROWS:
|
|
|
|
{
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
switch (src0->type) {
|
|
|
|
case GGML_TYPE_F32: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_F32 ].pipeline; break;
|
|
|
|
case GGML_TYPE_F16: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_F16 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_Q4_0 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_1: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_Q4_1 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_Q5_0 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_1: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_Q5_1 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q8_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_Q8_0 ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q2_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_Q2_K ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q3_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_Q3_K ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_Q4_K ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_Q5_K ].pipeline; break;
|
|
|
|
case GGML_TYPE_Q6_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_Q6_K ].pipeline; break;
|
|
|
|
case GGML_TYPE_IQ2_XXS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_XXS].pipeline; break;
|
|
|
|
case GGML_TYPE_IQ2_XS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_XS ].pipeline; break;
|
2024-01-30 14:14:12 +01:00
|
|
|
case GGML_TYPE_IQ3_XXS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ3_XXS].pipeline; break;
|
2024-02-24 15:23:52 +01:00
|
|
|
case GGML_TYPE_IQ3_S: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ3_S ].pipeline; break;
|
2024-02-26 17:28:38 +01:00
|
|
|
case GGML_TYPE_IQ2_S: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_S ].pipeline; break;
|
2024-02-18 17:16:55 +01:00
|
|
|
case GGML_TYPE_IQ1_S: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ1_S ].pipeline; break;
|
2024-03-27 17:55:10 +01:00
|
|
|
case GGML_TYPE_IQ1_M: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ1_M ].pipeline; break;
|
2024-02-21 15:19:39 +01:00
|
|
|
case GGML_TYPE_IQ4_NL: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_NL ].pipeline; break;
|
2024-02-27 15:34:24 +01:00
|
|
|
case GGML_TYPE_IQ4_XS: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_XS ].pipeline; break;
|
2024-01-18 10:03:13 +01:00
|
|
|
case GGML_TYPE_I32: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_I32 ].pipeline; break;
|
|
|
|
default: GGML_ASSERT(false && "not implemented");
|
2023-09-05 12:54:40 +02:00
|
|
|
}
|
2024-01-10 15:19:19 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:3];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:4];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:5];
|
|
|
|
[encoder setBytes:&ne10 length:sizeof( int64_t) atIndex:6];
|
|
|
|
[encoder setBytes:&nb10 length:sizeof( int64_t) atIndex:7];
|
|
|
|
[encoder setBytes:&nb11 length:sizeof( int64_t) atIndex:8];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(uint64_t) atIndex:9];
|
|
|
|
[encoder setBytes:&nb2 length:sizeof(uint64_t) atIndex:10];
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne10, ne11, 1) threadsPerThreadgroup:MTLSizeMake(32, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_RMS_NORM:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(ne00 % 4 == 0);
|
|
|
|
|
|
|
|
float eps;
|
|
|
|
memcpy(&eps, dst->op_params, sizeof(float));
|
|
|
|
|
|
|
|
int nth = 32; // SIMD width
|
|
|
|
|
|
|
|
while (nth < ne00/4 && nth < 1024) {
|
|
|
|
nth *= 2;
|
|
|
|
}
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_RMS_NORM].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:3];
|
|
|
|
[encoder setBytes:&eps length:sizeof( float) atIndex:4];
|
|
|
|
[encoder setThreadgroupMemoryLength:32*sizeof(float) atIndex:0];
|
|
|
|
|
|
|
|
const int64_t nrows = ggml_nrows(src0);
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(nrows, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_GROUP_NORM:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(ne00 % 4 == 0);
|
|
|
|
|
|
|
|
//float eps;
|
|
|
|
//memcpy(&eps, dst->op_params, sizeof(float));
|
|
|
|
|
|
|
|
const float eps = 1e-6f; // TODO: temporarily hardcoded
|
|
|
|
|
|
|
|
const int32_t n_groups = ((int32_t *) dst->op_params)[0];
|
|
|
|
|
|
|
|
int nth = 32; // SIMD width
|
|
|
|
|
|
|
|
//while (nth < ne00/4 && nth < 1024) {
|
|
|
|
// nth *= 2;
|
|
|
|
//}
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GROUP_NORM].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof( int64_t) atIndex:3];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof( int64_t) atIndex:4];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(uint64_t) atIndex:5];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:6];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:7];
|
|
|
|
[encoder setBytes:&n_groups length:sizeof( int32_t) atIndex:8];
|
|
|
|
[encoder setBytes:&eps length:sizeof( float) atIndex:9];
|
|
|
|
[encoder setThreadgroupMemoryLength:32*sizeof(float) atIndex:0];
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n_groups, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_NORM:
|
|
|
|
{
|
|
|
|
float eps;
|
|
|
|
memcpy(&eps, dst->op_params, sizeof(float));
|
|
|
|
|
|
|
|
const int nth = MIN(256, ne00);
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_NORM].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:3];
|
|
|
|
[encoder setBytes:&eps length:sizeof( float) atIndex:4];
|
|
|
|
[encoder setThreadgroupMemoryLength:GGML_PAD(nth*sizeof(float), 16) atIndex:0];
|
|
|
|
|
|
|
|
const int64_t nrows = ggml_nrows(src0);
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(nrows, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_ROPE:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(ne10 == ne02);
|
|
|
|
|
|
|
|
const int nth = MIN(1024, ne00);
|
|
|
|
|
|
|
|
const int n_past = ((int32_t *) dst->op_params)[0];
|
|
|
|
const int n_dims = ((int32_t *) dst->op_params)[1];
|
|
|
|
const int mode = ((int32_t *) dst->op_params)[2];
|
|
|
|
// skip 3, n_ctx, used in GLM RoPE, unimplemented in metal
|
|
|
|
const int n_orig_ctx = ((int32_t *) dst->op_params)[4];
|
|
|
|
|
|
|
|
float freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow;
|
|
|
|
memcpy(&freq_base, (int32_t *) dst->op_params + 5, sizeof(float));
|
|
|
|
memcpy(&freq_scale, (int32_t *) dst->op_params + 6, sizeof(float));
|
|
|
|
memcpy(&ext_factor, (int32_t *) dst->op_params + 7, sizeof(float));
|
|
|
|
memcpy(&attn_factor, (int32_t *) dst->op_params + 8, sizeof(float));
|
|
|
|
memcpy(&beta_fast, (int32_t *) dst->op_params + 9, sizeof(float));
|
|
|
|
memcpy(&beta_slow, (int32_t *) dst->op_params + 10, sizeof(float));
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
switch (src0->type) {
|
|
|
|
case GGML_TYPE_F32: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_ROPE_F32].pipeline; break;
|
|
|
|
case GGML_TYPE_F16: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_ROPE_F16].pipeline; break;
|
|
|
|
default: GGML_ASSERT(false);
|
|
|
|
};
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:3];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof( int64_t) atIndex:4];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof( int64_t) atIndex:5];
|
|
|
|
[encoder setBytes:&ne03 length:sizeof( int64_t) atIndex:6];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(uint64_t) atIndex:7];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:8];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:9];
|
|
|
|
[encoder setBytes:&nb03 length:sizeof(uint64_t) atIndex:10];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof( int64_t) atIndex:11];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof( int64_t) atIndex:12];
|
|
|
|
[encoder setBytes:&ne2 length:sizeof( int64_t) atIndex:13];
|
|
|
|
[encoder setBytes:&ne3 length:sizeof( int64_t) atIndex:14];
|
|
|
|
[encoder setBytes:&nb0 length:sizeof(uint64_t) atIndex:15];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(uint64_t) atIndex:16];
|
|
|
|
[encoder setBytes:&nb2 length:sizeof(uint64_t) atIndex:17];
|
|
|
|
[encoder setBytes:&nb3 length:sizeof(uint64_t) atIndex:18];
|
|
|
|
[encoder setBytes:&n_past length:sizeof( int) atIndex:19];
|
|
|
|
[encoder setBytes:&n_dims length:sizeof( int) atIndex:20];
|
|
|
|
[encoder setBytes:&mode length:sizeof( int) atIndex:21];
|
|
|
|
[encoder setBytes:&n_orig_ctx length:sizeof( int) atIndex:22];
|
|
|
|
[encoder setBytes:&freq_base length:sizeof( float) atIndex:23];
|
|
|
|
[encoder setBytes:&freq_scale length:sizeof( float) atIndex:24];
|
|
|
|
[encoder setBytes:&ext_factor length:sizeof( float) atIndex:25];
|
|
|
|
[encoder setBytes:&attn_factor length:sizeof( float) atIndex:26];
|
|
|
|
[encoder setBytes:&beta_fast length:sizeof( float) atIndex:27];
|
|
|
|
[encoder setBytes:&beta_slow length:sizeof( float) atIndex:28];
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_IM2COL:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(src0->type == GGML_TYPE_F16);
|
|
|
|
GGML_ASSERT(src1->type == GGML_TYPE_F32);
|
2024-01-31 14:35:41 +01:00
|
|
|
GGML_ASSERT( dst->type == GGML_TYPE_F16 || dst->type == GGML_TYPE_F32);
|
2024-01-18 10:03:13 +01:00
|
|
|
|
|
|
|
const int32_t s0 = ((const int32_t *)(dst->op_params))[0];
|
|
|
|
const int32_t s1 = ((const int32_t *)(dst->op_params))[1];
|
|
|
|
const int32_t p0 = ((const int32_t *)(dst->op_params))[2];
|
|
|
|
const int32_t p1 = ((const int32_t *)(dst->op_params))[3];
|
|
|
|
const int32_t d0 = ((const int32_t *)(dst->op_params))[4];
|
|
|
|
const int32_t d1 = ((const int32_t *)(dst->op_params))[5];
|
2024-01-31 14:35:41 +01:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
const bool is_2D = ((const int32_t *)(dst->op_params))[6] == 1;
|
|
|
|
|
|
|
|
const int32_t N = src1->ne[is_2D ? 3 : 2];
|
|
|
|
const int32_t IC = src1->ne[is_2D ? 2 : 1];
|
|
|
|
const int32_t IH = is_2D ? src1->ne[1] : 1;
|
|
|
|
const int32_t IW = src1->ne[0];
|
|
|
|
|
|
|
|
const int32_t KH = is_2D ? src0->ne[1] : 1;
|
|
|
|
const int32_t KW = src0->ne[0];
|
|
|
|
|
|
|
|
const int32_t OH = is_2D ? dst->ne[2] : 1;
|
|
|
|
const int32_t OW = dst->ne[1];
|
|
|
|
|
|
|
|
const int32_t CHW = IC * KH * KW;
|
|
|
|
|
|
|
|
const int32_t ofs0 = src1->nb[is_2D ? 3 : 2] / 4;
|
|
|
|
const int32_t ofs1 = src1->nb[is_2D ? 2 : 1] / 4;
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
2024-01-31 14:35:41 +01:00
|
|
|
switch (dst->type) {
|
|
|
|
case GGML_TYPE_F32: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_IM2COL_F32].pipeline; break;
|
2024-01-18 10:03:13 +01:00
|
|
|
case GGML_TYPE_F16: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_IM2COL_F16].pipeline; break;
|
|
|
|
default: GGML_ASSERT(false);
|
|
|
|
};
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ofs0 length:sizeof( int32_t) atIndex:2];
|
|
|
|
[encoder setBytes:&ofs1 length:sizeof( int32_t) atIndex:3];
|
|
|
|
[encoder setBytes:&IW length:sizeof( int32_t) atIndex:4];
|
|
|
|
[encoder setBytes:&IH length:sizeof( int32_t) atIndex:5];
|
|
|
|
[encoder setBytes:&CHW length:sizeof( int32_t) atIndex:6];
|
|
|
|
[encoder setBytes:&s0 length:sizeof( int32_t) atIndex:7];
|
|
|
|
[encoder setBytes:&s1 length:sizeof( int32_t) atIndex:8];
|
|
|
|
[encoder setBytes:&p0 length:sizeof( int32_t) atIndex:9];
|
|
|
|
[encoder setBytes:&p1 length:sizeof( int32_t) atIndex:10];
|
|
|
|
[encoder setBytes:&d0 length:sizeof( int32_t) atIndex:11];
|
|
|
|
[encoder setBytes:&d1 length:sizeof( int32_t) atIndex:12];
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(IC, OH, OW) threadsPerThreadgroup:MTLSizeMake(N, KH, KW)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_UPSCALE:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(src0->type == GGML_TYPE_F32);
|
|
|
|
|
|
|
|
const int sf = dst->op_params[0];
|
|
|
|
|
|
|
|
const id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_UPSCALE_F32].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
|
|
|
|
[encoder setBytes:&ne03 length:sizeof(ne03) atIndex:5];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(nb00) atIndex:6];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:7];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:8];
|
|
|
|
[encoder setBytes:&nb03 length:sizeof(nb03) atIndex:9];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:10];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:11];
|
|
|
|
[encoder setBytes:&ne2 length:sizeof(ne2) atIndex:12];
|
|
|
|
[encoder setBytes:&ne3 length:sizeof(ne3) atIndex:13];
|
|
|
|
[encoder setBytes:&nb0 length:sizeof(nb0) atIndex:14];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(nb1) atIndex:15];
|
|
|
|
[encoder setBytes:&nb2 length:sizeof(nb2) atIndex:16];
|
|
|
|
[encoder setBytes:&nb3 length:sizeof(nb3) atIndex:17];
|
|
|
|
[encoder setBytes:&sf length:sizeof(sf) atIndex:18];
|
|
|
|
|
|
|
|
const int nth = MIN((int) pipeline.maxTotalThreadsPerThreadgroup, ne0);
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne1, ne2, ne3) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_PAD:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(src0->type == GGML_TYPE_F32);
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_PAD_F32].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
|
|
|
|
[encoder setBytes:&ne03 length:sizeof(ne03) atIndex:5];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(nb00) atIndex:6];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:7];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:8];
|
|
|
|
[encoder setBytes:&nb03 length:sizeof(nb03) atIndex:9];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:10];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:11];
|
|
|
|
[encoder setBytes:&ne2 length:sizeof(ne2) atIndex:12];
|
|
|
|
[encoder setBytes:&ne3 length:sizeof(ne3) atIndex:13];
|
|
|
|
[encoder setBytes:&nb0 length:sizeof(nb0) atIndex:14];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(nb1) atIndex:15];
|
|
|
|
[encoder setBytes:&nb2 length:sizeof(nb2) atIndex:16];
|
|
|
|
[encoder setBytes:&nb3 length:sizeof(nb3) atIndex:17];
|
|
|
|
|
|
|
|
const int nth = MIN(1024, ne0);
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne1, ne2, ne3) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
2024-03-03 13:23:52 +01:00
|
|
|
case GGML_OP_ARANGE:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(dst->type == GGML_TYPE_F32);
|
|
|
|
|
|
|
|
float start;
|
|
|
|
float step;
|
|
|
|
|
|
|
|
memcpy(&start, ((int32_t *) dst->op_params) + 0, sizeof(float));
|
|
|
|
memcpy(&step, ((int32_t *) dst->op_params) + 2, sizeof(float));
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_ARANGE_F32].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:0];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:1];
|
|
|
|
[encoder setBytes:&start length:sizeof(start) atIndex:2];
|
|
|
|
[encoder setBytes:&step length:sizeof(step) atIndex:3];
|
|
|
|
|
|
|
|
const int nth = MIN(1024, ne0);
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(1, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
|
|
|
case GGML_OP_TIMESTEP_EMBEDDING:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(src0->type == GGML_TYPE_F32);
|
|
|
|
|
|
|
|
const int dim = dst->op_params[0];
|
|
|
|
const int max_period = dst->op_params[1];
|
|
|
|
|
|
|
|
const int half = dim / 2;
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_TIMESTEP_EMBEDDING_F32].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(nb1) atIndex:2];
|
|
|
|
[encoder setBytes:&dim length:sizeof(dim) atIndex:3];
|
|
|
|
[encoder setBytes:&max_period length:sizeof(max_period) atIndex:4];
|
|
|
|
|
|
|
|
const int nth = MIN(1024, half);
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne00, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
2024-01-18 10:03:13 +01:00
|
|
|
case GGML_OP_ARGSORT:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(src0->type == GGML_TYPE_F32);
|
|
|
|
GGML_ASSERT( dst->type == GGML_TYPE_I32);
|
|
|
|
|
|
|
|
const int nrows = ggml_nrows(src0);
|
|
|
|
|
|
|
|
enum ggml_sort_order order = (enum ggml_sort_order) dst->op_params[0];
|
|
|
|
|
2024-04-03 15:07:05 +02:00
|
|
|
// bitonic sort requires the number of elements to be power of 2
|
|
|
|
int64_t ne00_padded = 1;
|
|
|
|
while (ne00_padded < ne00) {
|
|
|
|
ne00_padded *= 2;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Metal kernels require the buffer size to be multiple of 16 bytes
|
|
|
|
// https://developer.apple.com/documentation/metal/mtlcomputecommandencoder/1443142-setthreadgroupmemorylength
|
|
|
|
const int mem_size = GGML_PAD(ne00_padded*sizeof(int32_t), 16);
|
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
switch (order) {
|
2024-02-25 11:09:09 +01:00
|
|
|
case GGML_SORT_ORDER_ASC: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_ARGSORT_F32_I32_ASC].pipeline; break;
|
|
|
|
case GGML_SORT_ORDER_DESC: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_ARGSORT_F32_I32_DESC].pipeline; break;
|
2024-01-18 10:03:13 +01:00
|
|
|
default: GGML_ASSERT(false);
|
|
|
|
};
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
2024-04-03 15:07:05 +02:00
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
|
|
|
|
[encoder setBytes:&ne00_padded length:sizeof( int64_t) atIndex:3];
|
|
|
|
[encoder setThreadgroupMemoryLength:mem_size atIndex:0];
|
2024-01-18 10:03:13 +01:00
|
|
|
|
2024-04-03 15:07:05 +02:00
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(1, nrows, 1) threadsPerThreadgroup:MTLSizeMake(ne00_padded, 1, 1)];
|
2024-01-18 10:03:13 +01:00
|
|
|
} break;
|
|
|
|
case GGML_OP_LEAKY_RELU:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(src0->type == GGML_TYPE_F32);
|
|
|
|
|
|
|
|
float slope;
|
|
|
|
memcpy(&slope, dst->op_params, sizeof(float));
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_LEAKY_RELU_F32].pipeline;
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&slope length:sizeof(slope) atIndex:2];
|
|
|
|
|
|
|
|
const int64_t n = ggml_nelements(dst);
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
|
|
|
|
} break;
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
case GGML_OP_FLASH_ATTN_EXT:
|
|
|
|
{
|
2024-05-14 18:09:30 +02:00
|
|
|
GGML_ASSERT(ne00 % 4 == 0);
|
|
|
|
GGML_ASSERT(ne11 % 32 == 0);
|
|
|
|
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
GGML_ASSERT(src0->type == GGML_TYPE_F32);
|
|
|
|
|
2024-05-14 18:09:30 +02:00
|
|
|
GGML_ASSERT(ggml_are_same_shape (src1, src2));
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
|
2024-05-14 18:09:30 +02:00
|
|
|
struct ggml_tensor * src3 = gf->nodes[i]->src[3];
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
|
|
|
|
size_t offs_src3 = 0;
|
|
|
|
|
|
|
|
id<MTLBuffer> id_src3 = src3 ? ggml_metal_get_buffer(src3, &offs_src3) : nil;
|
|
|
|
|
|
|
|
GGML_ASSERT(!src3 || src3->type == GGML_TYPE_F16);
|
|
|
|
GGML_ASSERT(!src3 || src3->ne[1] >= GGML_PAD(src0->ne[1], 8) &&
|
|
|
|
"the Flash-Attention Metal kernel requires the mask to be padded to 8 and at least n_queries big");
|
|
|
|
|
2024-05-14 18:09:30 +02:00
|
|
|
const uint64_t nb20 = src2 ? src2->nb[0] : 0; GGML_UNUSED(nb20);
|
|
|
|
const uint64_t nb21 = src2 ? src2->nb[1] : 0;
|
|
|
|
const uint64_t nb22 = src2 ? src2->nb[2] : 0;
|
|
|
|
const uint64_t nb23 = src2 ? src2->nb[3] : 0;
|
|
|
|
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
const int64_t ne30 = src3 ? src3->ne[0] : 0; GGML_UNUSED(ne30);
|
2024-05-11 09:32:41 +02:00
|
|
|
//const int64_t ne31 = src3 ? src3->ne[1] : 0;
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
const int64_t ne32 = src3 ? src3->ne[2] : 0; GGML_UNUSED(ne32);
|
|
|
|
const int64_t ne33 = src3 ? src3->ne[3] : 0; GGML_UNUSED(ne33);
|
|
|
|
|
|
|
|
const uint64_t nb30 = src3 ? src3->nb[0] : 0; GGML_UNUSED(nb30);
|
|
|
|
const uint64_t nb31 = src3 ? src3->nb[1] : 0;
|
|
|
|
const uint64_t nb32 = src3 ? src3->nb[2] : 0; GGML_UNUSED(nb32);
|
|
|
|
const uint64_t nb33 = src3 ? src3->nb[3] : 0; GGML_UNUSED(nb33);
|
|
|
|
|
|
|
|
const enum ggml_type src2t = src2 ? src2->type : GGML_TYPE_COUNT; GGML_UNUSED(src2t);
|
|
|
|
|
|
|
|
float scale;
|
2024-05-11 09:32:41 +02:00
|
|
|
float max_bias;
|
|
|
|
|
|
|
|
memcpy(&scale, ((int32_t *) dst->op_params) + 0, sizeof(scale));
|
|
|
|
memcpy(&max_bias, ((int32_t *) dst->op_params) + 1, sizeof(max_bias));
|
|
|
|
|
|
|
|
const uint32_t n_head = src0->ne[2];
|
|
|
|
const uint32_t n_head_log2 = 1u << (uint32_t) floorf(log2f((float) n_head));
|
|
|
|
|
|
|
|
const float m0 = powf(2.0f, -(max_bias ) / n_head_log2);
|
|
|
|
const float m1 = powf(2.0f, -(max_bias / 2.0f) / n_head_log2);
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
bool use_vec_kernel = false;
|
|
|
|
|
|
|
|
if (ne01 >= 4 || (ne00%128 != 0)) {
|
|
|
|
switch (ne00) {
|
|
|
|
case 64: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H64 ].pipeline; break;
|
|
|
|
case 80: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H80 ].pipeline; break;
|
|
|
|
case 96: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H96 ].pipeline; break;
|
|
|
|
case 112: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H112].pipeline; break;
|
|
|
|
case 128: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H128].pipeline; break;
|
|
|
|
case 256: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_F16_H256].pipeline; break;
|
|
|
|
default:
|
|
|
|
{
|
|
|
|
GGML_METAL_LOG_ERROR("unsupported size: %lld\n", ne00);
|
|
|
|
GGML_METAL_LOG_ERROR("add template specialization for this size\n");
|
|
|
|
GGML_ASSERT(false && "add template specialization for this size");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
use_vec_kernel = true;
|
|
|
|
|
|
|
|
switch (ne00) {
|
|
|
|
case 128: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_VEC_F16_H128].pipeline; break;
|
|
|
|
case 256: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_FLASH_ATTN_EXT_VEC_F16_H256].pipeline; break;
|
|
|
|
default:
|
|
|
|
{
|
|
|
|
GGML_METAL_LOG_ERROR("unsupported size: %lld\n", ne00);
|
|
|
|
GGML_METAL_LOG_ERROR("add template specialization for this size\n");
|
|
|
|
GGML_ASSERT(false && "add template specialization for this size");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
[encoder setComputePipelineState:pipeline];
|
2024-05-11 09:32:41 +02:00
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
|
|
|
|
[encoder setBuffer:id_src2 offset:offs_src2 atIndex:2];
|
2024-05-14 18:09:30 +02:00
|
|
|
if (id_src3) {
|
|
|
|
[encoder setBuffer:id_src3 offset:offs_src3 atIndex:3];
|
|
|
|
} else {
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:3];
|
|
|
|
}
|
2024-05-11 09:32:41 +02:00
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:4];
|
2024-05-14 18:09:30 +02:00
|
|
|
[encoder setBytes:&ne01 length:sizeof( int64_t) atIndex:5];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof( int64_t) atIndex:6];
|
|
|
|
[encoder setBytes:&ne03 length:sizeof( int64_t) atIndex:7];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:8];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:9];
|
|
|
|
[encoder setBytes:&nb03 length:sizeof(uint64_t) atIndex:10];
|
|
|
|
[encoder setBytes:&ne11 length:sizeof( int64_t) atIndex:11];
|
|
|
|
[encoder setBytes:&ne12 length:sizeof( int64_t) atIndex:12];
|
|
|
|
[encoder setBytes:&ne13 length:sizeof( int64_t) atIndex:13];
|
|
|
|
[encoder setBytes:&nb11 length:sizeof(uint64_t) atIndex:14];
|
|
|
|
[encoder setBytes:&nb12 length:sizeof(uint64_t) atIndex:15];
|
|
|
|
[encoder setBytes:&nb13 length:sizeof(uint64_t) atIndex:16];
|
|
|
|
[encoder setBytes:&nb21 length:sizeof(uint64_t) atIndex:17];
|
|
|
|
[encoder setBytes:&nb22 length:sizeof(uint64_t) atIndex:18];
|
|
|
|
[encoder setBytes:&nb23 length:sizeof(uint64_t) atIndex:19];
|
|
|
|
[encoder setBytes:&nb31 length:sizeof(uint64_t) atIndex:20];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof( int64_t) atIndex:21];
|
|
|
|
[encoder setBytes:&ne2 length:sizeof( int64_t) atIndex:22];
|
|
|
|
[encoder setBytes:&scale length:sizeof( float) atIndex:23];
|
|
|
|
[encoder setBytes:&max_bias length:sizeof( float) atIndex:24];
|
|
|
|
[encoder setBytes:&m0 length:sizeof(m0) atIndex:25];
|
|
|
|
[encoder setBytes:&m1 length:sizeof(m1) atIndex:26];
|
|
|
|
[encoder setBytes:&n_head_log2 length:sizeof(n_head_log2) atIndex:27];
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
|
|
|
|
if (!use_vec_kernel) {
|
|
|
|
// half8x8 kernel
|
|
|
|
const int64_t nqptg = 8; // queries per threadgroup !! sync with kernel template arguments !!
|
|
|
|
const int64_t ncpsg = 32; // cache values per simdgroup !! sync with kernel template arguments !!
|
|
|
|
|
|
|
|
GGML_ASSERT(nqptg <= 32);
|
|
|
|
GGML_ASSERT(nqptg % 8 == 0);
|
|
|
|
GGML_ASSERT(ncpsg % 32 == 0);
|
|
|
|
|
|
|
|
int64_t nsgmax = 2;
|
|
|
|
|
|
|
|
while (true) {
|
|
|
|
const size_t smem = nqptg*(ne00 + 2*nsgmax*(ncpsg + nqptg))*(sizeof(float)/2);
|
|
|
|
if (smem > ctx->device.maxThreadgroupMemoryLength) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
nsgmax *= 2;
|
|
|
|
}
|
|
|
|
nsgmax /= 2;
|
|
|
|
|
|
|
|
// simdgroups per threadgroup (a.k.a. warps)
|
|
|
|
const int64_t nsg = ne01 <= nqptg ? MAX(4, MIN(nsgmax, MIN(ne11/ncpsg, (int64_t) pipeline.maxTotalThreadsPerThreadgroup/32))) : 4;
|
|
|
|
|
|
|
|
const size_t smem = nqptg*(ne00 + 2*nsg*(ncpsg + nqptg))*(sizeof(float)/2);
|
|
|
|
|
|
|
|
//printf("smem: %zu, max: %zu\n", smem, ctx->device.maxThreadgroupMemoryLength);
|
|
|
|
GGML_ASSERT(smem <= ctx->device.maxThreadgroupMemoryLength);
|
|
|
|
|
|
|
|
[encoder setThreadgroupMemoryLength:GGML_PAD(smem, 16) atIndex:0];
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + nqptg - 1)/nqptg, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(32, nsg, 1)];
|
|
|
|
} else {
|
|
|
|
// half1x4 kernel
|
|
|
|
const int64_t nqptg = 1; // queries per threadgroup !! sync with kernel template arguments !!
|
|
|
|
const int64_t ncpsg = 32; // cache values per simdgroup !! sync with kernel template arguments !!
|
|
|
|
|
|
|
|
GGML_ASSERT(nqptg <= 32);
|
|
|
|
GGML_ASSERT(nqptg % 1 == 0);
|
|
|
|
GGML_ASSERT(ncpsg % 32 == 0);
|
|
|
|
|
|
|
|
// simdgroups per threadgroup (a.k.a. warps)
|
|
|
|
const int64_t nsgt = MAX(2, MIN(ne11/ncpsg, (int64_t) pipeline.maxTotalThreadsPerThreadgroup/32));
|
|
|
|
|
|
|
|
int64_t nsg = 1;
|
|
|
|
while (nsg <= nsgt) {
|
|
|
|
nsg *= 2;
|
|
|
|
}
|
|
|
|
nsg /= 2;
|
|
|
|
|
|
|
|
const size_t smem = (nqptg*(ne00 + 2*nsg*(ncpsg + nqptg)) + nsg*ne00)*(sizeof(float)/2);
|
|
|
|
|
|
|
|
//printf("smem: %zu, max: %zu\n", smem, ctx->device.maxThreadgroupMemoryLength);
|
|
|
|
GGML_ASSERT(smem <= ctx->device.maxThreadgroupMemoryLength);
|
|
|
|
[encoder setThreadgroupMemoryLength:GGML_PAD(smem, 16) atIndex:0];
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + nqptg - 1)/nqptg, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(32, nsg, 1)];
|
|
|
|
}
|
|
|
|
} break;
|
2024-01-18 10:03:13 +01:00
|
|
|
case GGML_OP_DUP:
|
|
|
|
case GGML_OP_CPY:
|
|
|
|
case GGML_OP_CONT:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(ne00 % ggml_blck_size(src0->type) == 0);
|
|
|
|
|
|
|
|
int nth = MIN(1024, ne00/ggml_blck_size(src0->type));
|
|
|
|
|
|
|
|
id<MTLComputePipelineState> pipeline = nil;
|
|
|
|
|
|
|
|
switch (src0t) {
|
|
|
|
case GGML_TYPE_F32:
|
|
|
|
{
|
|
|
|
GGML_ASSERT(ne0 % ggml_blck_size(dst->type) == 0);
|
|
|
|
|
|
|
|
switch (dstt) {
|
2024-03-27 17:55:10 +01:00
|
|
|
case GGML_TYPE_F16: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F32_F16].pipeline; break;
|
|
|
|
case GGML_TYPE_F32: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F32_F32].pipeline; break;
|
|
|
|
case GGML_TYPE_Q8_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F32_Q8_0].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F32_Q4_0].pipeline; break;
|
|
|
|
case GGML_TYPE_Q4_1: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F32_Q4_1].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_0: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F32_Q5_0].pipeline; break;
|
|
|
|
case GGML_TYPE_Q5_1: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F32_Q5_1].pipeline; break;
|
|
|
|
case GGML_TYPE_IQ4_NL: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F32_IQ4_NL].pipeline; break;
|
2024-01-18 10:03:13 +01:00
|
|
|
default: GGML_ASSERT(false && "not implemented");
|
|
|
|
};
|
|
|
|
} break;
|
|
|
|
case GGML_TYPE_F16:
|
|
|
|
{
|
|
|
|
switch (dstt) {
|
|
|
|
case GGML_TYPE_F16: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F16_F16].pipeline; break;
|
|
|
|
case GGML_TYPE_F32: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_CPY_F16_F32].pipeline; break;
|
|
|
|
default: GGML_ASSERT(false && "not implemented");
|
|
|
|
};
|
|
|
|
} break;
|
|
|
|
default: GGML_ASSERT(false && "not implemented");
|
|
|
|
}
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[encoder setComputePipelineState:pipeline];
|
|
|
|
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
|
|
|
|
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
|
|
|
|
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
|
|
|
|
[encoder setBytes:&ne01 length:sizeof( int64_t) atIndex:3];
|
|
|
|
[encoder setBytes:&ne02 length:sizeof( int64_t) atIndex:4];
|
|
|
|
[encoder setBytes:&ne03 length:sizeof( int64_t) atIndex:5];
|
|
|
|
[encoder setBytes:&nb00 length:sizeof(uint64_t) atIndex:6];
|
|
|
|
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:7];
|
|
|
|
[encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:8];
|
|
|
|
[encoder setBytes:&nb03 length:sizeof(uint64_t) atIndex:9];
|
|
|
|
[encoder setBytes:&ne0 length:sizeof( int64_t) atIndex:10];
|
|
|
|
[encoder setBytes:&ne1 length:sizeof( int64_t) atIndex:11];
|
|
|
|
[encoder setBytes:&ne2 length:sizeof( int64_t) atIndex:12];
|
|
|
|
[encoder setBytes:&ne3 length:sizeof( int64_t) atIndex:13];
|
|
|
|
[encoder setBytes:&nb0 length:sizeof(uint64_t) atIndex:14];
|
|
|
|
[encoder setBytes:&nb1 length:sizeof(uint64_t) atIndex:15];
|
|
|
|
[encoder setBytes:&nb2 length:sizeof(uint64_t) atIndex:16];
|
|
|
|
[encoder setBytes:&nb3 length:sizeof(uint64_t) atIndex:17];
|
|
|
|
|
|
|
|
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
|
|
|
|
} break;
|
|
|
|
default:
|
|
|
|
{
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: node %3d, op = %8s not implemented\n", __func__, i, ggml_op_name(dst->op));
|
|
|
|
GGML_ASSERT(false);
|
|
|
|
}
|
2023-06-25 14:40:30 +02:00
|
|
|
}
|
|
|
|
|
2024-01-29 10:22:23 +01:00
|
|
|
if (should_capture) {
|
|
|
|
[encoder popDebugGroup];
|
|
|
|
}
|
2024-01-18 10:03:13 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
[encoder endEncoding];
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
[command_buffer commit];
|
|
|
|
});
|
2023-06-25 14:40:30 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
// Wait for completion and check status of each command buffer
|
2023-06-25 14:40:30 +02:00
|
|
|
// needed to detect if the device ran out-of-memory for example (#1881)
|
2023-09-05 12:54:40 +02:00
|
|
|
|
2024-01-18 10:03:13 +01:00
|
|
|
for (int i = 0; i < n_cb; ++i) {
|
|
|
|
id<MTLCommandBuffer> command_buffer = command_buffers[i];
|
|
|
|
[command_buffer waitUntilCompleted];
|
|
|
|
|
|
|
|
MTLCommandBufferStatus status = [command_buffer status];
|
2023-06-25 14:40:30 +02:00
|
|
|
if (status != MTLCommandBufferStatusCompleted) {
|
2023-11-03 20:35:05 +01:00
|
|
|
GGML_METAL_LOG_INFO("%s: command buffer %d failed with status %lu\n", __func__, i, status);
|
2024-04-30 11:34:50 +02:00
|
|
|
if (status == MTLCommandBufferStatusError) {
|
2024-04-30 17:14:02 +02:00
|
|
|
NSString * error_code = [command_buffer error].localizedDescription;
|
|
|
|
GGML_METAL_LOG_INFO("error: %s\n", [error_code UTF8String]);
|
2024-04-30 11:34:50 +02:00
|
|
|
}
|
|
|
|
|
2024-03-04 10:05:42 +01:00
|
|
|
return GGML_STATUS_FAILED;
|
2023-06-25 14:40:30 +02:00
|
|
|
}
|
|
|
|
}
|
2023-09-05 12:54:40 +02:00
|
|
|
|
2024-01-29 10:22:23 +01:00
|
|
|
if (should_capture) {
|
|
|
|
[[MTLCaptureManager sharedCaptureManager] stopCapture];
|
|
|
|
}
|
|
|
|
|
2024-02-10 11:53:28 +01:00
|
|
|
}
|
2024-03-04 10:05:42 +01:00
|
|
|
return GGML_STATUS_SUCCESS;
|
2023-06-25 14:40:30 +02:00
|
|
|
}
|
2023-11-03 20:35:05 +01:00
|
|
|
|
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
|
|
|
|
// backend interface
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
// default buffer
|
2023-12-07 21:27:19 +01:00
|
|
|
static id<MTLDevice> g_backend_device = nil;
|
|
|
|
static int g_backend_device_ref_count = 0;
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2023-12-07 21:27:19 +01:00
|
|
|
static id<MTLDevice> ggml_backend_metal_get_device(void) {
|
|
|
|
if (g_backend_device == nil) {
|
|
|
|
g_backend_device = MTLCreateSystemDefaultDevice();
|
|
|
|
}
|
|
|
|
|
|
|
|
g_backend_device_ref_count++;
|
|
|
|
|
|
|
|
return g_backend_device;
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
|
|
|
|
2023-12-07 21:27:19 +01:00
|
|
|
static void ggml_backend_metal_free_device(void) {
|
|
|
|
assert(g_backend_device_ref_count > 0);
|
|
|
|
|
|
|
|
g_backend_device_ref_count--;
|
|
|
|
|
|
|
|
if (g_backend_device_ref_count == 0) {
|
|
|
|
[g_backend_device release];
|
|
|
|
g_backend_device = nil;
|
|
|
|
}
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static const char * ggml_backend_metal_buffer_get_name(ggml_backend_buffer_t buffer) {
|
2024-01-12 20:07:38 +01:00
|
|
|
return "Metal";
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-01-12 20:07:38 +01:00
|
|
|
UNUSED(buffer);
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static void ggml_backend_metal_buffer_free_buffer(ggml_backend_buffer_t buffer) {
|
2023-12-07 21:27:19 +01:00
|
|
|
struct ggml_backend_metal_buffer_context * ctx = (struct ggml_backend_metal_buffer_context *)buffer->context;
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
for (int i = 0; i < ctx->n_buffers; i++) {
|
|
|
|
[ctx->buffers[i].metal release];
|
|
|
|
}
|
2023-12-07 21:27:19 +01:00
|
|
|
ggml_backend_metal_free_device();
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
if (ctx->owned) {
|
2024-05-08 21:08:10 +02:00
|
|
|
#if TARGET_OS_OSX
|
|
|
|
vm_deallocate((vm_map_t)mach_task_self(), (vm_address_t)ctx->all_data, ctx->all_size);
|
|
|
|
#else
|
2023-12-22 16:53:39 +01:00
|
|
|
free(ctx->all_data);
|
2024-05-08 21:08:10 +02:00
|
|
|
#endif
|
2023-12-22 16:53:39 +01:00
|
|
|
}
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
free(ctx);
|
2023-12-07 21:27:19 +01:00
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static void * ggml_backend_metal_buffer_get_base(ggml_backend_buffer_t buffer) {
|
2024-01-12 20:07:38 +01:00
|
|
|
struct ggml_backend_metal_buffer_context * ctx = (struct ggml_backend_metal_buffer_context *)buffer->context;
|
|
|
|
|
|
|
|
return ctx->all_data;
|
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static void ggml_backend_metal_buffer_set_tensor(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size) {
|
2023-12-07 21:27:19 +01:00
|
|
|
memcpy((char *)tensor->data + offset, data, size);
|
|
|
|
|
|
|
|
UNUSED(buffer);
|
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static void ggml_backend_metal_buffer_get_tensor(ggml_backend_buffer_t buffer, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size) {
|
2023-12-07 21:27:19 +01:00
|
|
|
memcpy(data, (const char *)tensor->data + offset, size);
|
|
|
|
|
|
|
|
UNUSED(buffer);
|
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static bool ggml_backend_metal_buffer_cpy_tensor(ggml_backend_buffer_t buffer, const struct ggml_tensor * src, struct ggml_tensor * dst) {
|
2024-01-12 20:07:38 +01:00
|
|
|
if (ggml_backend_buffer_is_host(src->buffer)) {
|
|
|
|
memcpy(dst->data, src->data, ggml_nbytes(src));
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
return false;
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2023-11-03 20:35:05 +01:00
|
|
|
UNUSED(buffer);
|
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static void ggml_backend_metal_buffer_clear(ggml_backend_buffer_t buffer, uint8_t value) {
|
2023-12-22 16:53:39 +01:00
|
|
|
struct ggml_backend_metal_buffer_context * ctx = (struct ggml_backend_metal_buffer_context *)buffer->context;
|
|
|
|
|
|
|
|
memset(ctx->all_data, value, ctx->all_size);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct ggml_backend_buffer_i ggml_backend_metal_buffer_i = {
|
2024-01-12 20:07:38 +01:00
|
|
|
/* .get_name = */ ggml_backend_metal_buffer_get_name,
|
2023-12-07 21:27:19 +01:00
|
|
|
/* .free_buffer = */ ggml_backend_metal_buffer_free_buffer,
|
|
|
|
/* .get_base = */ ggml_backend_metal_buffer_get_base,
|
|
|
|
/* .init_tensor = */ NULL,
|
|
|
|
/* .set_tensor = */ ggml_backend_metal_buffer_set_tensor,
|
|
|
|
/* .get_tensor = */ ggml_backend_metal_buffer_get_tensor,
|
2024-01-12 20:07:38 +01:00
|
|
|
/* .cpy_tensor = */ ggml_backend_metal_buffer_cpy_tensor,
|
2023-12-22 16:53:39 +01:00
|
|
|
/* .clear = */ ggml_backend_metal_buffer_clear,
|
2024-01-12 20:07:38 +01:00
|
|
|
/* .reset = */ NULL,
|
2023-11-03 20:35:05 +01:00
|
|
|
};
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
// default buffer type
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static const char * ggml_backend_metal_buffer_type_get_name(ggml_backend_buffer_type_t buft) {
|
2024-01-12 20:07:38 +01:00
|
|
|
return "Metal";
|
|
|
|
|
|
|
|
UNUSED(buft);
|
|
|
|
}
|
|
|
|
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
static void ggml_backend_metal_log_allocated_size(id<MTLDevice> device, size_t size_aligned) {
|
|
|
|
#ifndef GGML_METAL_NDEBUG
|
2024-01-16 14:33:02 +01:00
|
|
|
#if TARGET_OS_OSX || (TARGET_OS_IOS && __clang_major__ >= 15)
|
|
|
|
if (@available(macOS 10.12, iOS 16.0, *)) {
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
GGML_METAL_LOG_INFO("%s: allocated buffer, size = %8.2f MiB, (%8.2f / %8.2f)",
|
|
|
|
__func__,
|
|
|
|
size_aligned / 1024.0 / 1024.0,
|
2024-01-16 14:33:02 +01:00
|
|
|
device.currentAllocatedSize / 1024.0 / 1024.0,
|
|
|
|
device.recommendedMaxWorkingSetSize / 1024.0 / 1024.0);
|
|
|
|
|
|
|
|
if (device.currentAllocatedSize > device.recommendedMaxWorkingSetSize) {
|
|
|
|
GGML_METAL_LOG_WARN("%s: warning: current allocated size is greater than the recommended max working set size\n", __func__);
|
|
|
|
} else {
|
|
|
|
GGML_METAL_LOG_INFO("\n");
|
|
|
|
}
|
|
|
|
} else {
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
GGML_METAL_LOG_INFO("%s: allocated buffer, size = %8.2f MiB, (%8.2f)\n",
|
|
|
|
__func__,
|
|
|
|
size_aligned / 1024.0 / 1024.0,
|
|
|
|
device.currentAllocatedSize / 1024.0 / 1024.0);
|
2024-01-16 14:33:02 +01:00
|
|
|
}
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
#endif
|
2024-01-16 14:33:02 +01:00
|
|
|
#endif
|
|
|
|
UNUSED(device);
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
UNUSED(size_aligned);
|
2024-01-16 14:33:02 +01:00
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static ggml_backend_buffer_t ggml_backend_metal_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
|
2023-12-07 21:27:19 +01:00
|
|
|
struct ggml_backend_metal_buffer_context * ctx = malloc(sizeof(struct ggml_backend_metal_buffer_context));
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2023-12-07 21:27:19 +01:00
|
|
|
const size_t size_page = sysconf(_SC_PAGESIZE);
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2023-12-07 21:27:19 +01:00
|
|
|
size_t size_aligned = size;
|
|
|
|
if ((size_aligned % size_page) != 0) {
|
|
|
|
size_aligned += (size_page - (size_aligned % size_page));
|
|
|
|
}
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
id<MTLDevice> device = ggml_backend_metal_get_device();
|
|
|
|
|
|
|
|
ctx->all_data = ggml_metal_host_malloc(size_aligned);
|
|
|
|
ctx->all_size = size_aligned;
|
|
|
|
ctx->owned = true;
|
|
|
|
ctx->n_buffers = 1;
|
|
|
|
|
2024-05-08 21:08:10 +02:00
|
|
|
if (ctx->all_data != NULL) {
|
|
|
|
ctx->buffers[0].data = ctx->all_data;
|
|
|
|
ctx->buffers[0].size = size;
|
|
|
|
ctx->buffers[0].metal = [device newBufferWithBytesNoCopy:ctx->all_data
|
|
|
|
length:size_aligned
|
|
|
|
options:MTLResourceStorageModeShared
|
|
|
|
deallocator:nil];
|
|
|
|
}
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-05-08 21:08:10 +02:00
|
|
|
if (ctx->all_data == NULL || ctx->buffers[0].metal == nil) {
|
2023-12-22 16:53:39 +01:00
|
|
|
GGML_METAL_LOG_ERROR("%s: error: failed to allocate buffer, size = %8.2f MiB\n", __func__, size_aligned / 1024.0 / 1024.0);
|
|
|
|
free(ctx);
|
|
|
|
ggml_backend_metal_free_device();
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
//ggml_backend_metal_log_allocated_size(device, size_aligned);
|
2023-12-22 16:53:39 +01:00
|
|
|
|
|
|
|
return ggml_backend_buffer_init(buft, ggml_backend_metal_buffer_i, ctx, size);
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static size_t ggml_backend_metal_buffer_type_get_alignment(ggml_backend_buffer_type_t buft) {
|
2023-11-03 20:35:05 +01:00
|
|
|
return 32;
|
2023-12-07 21:27:19 +01:00
|
|
|
UNUSED(buft);
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
|
|
|
|
2024-01-29 09:05:13 +01:00
|
|
|
GGML_CALL static size_t ggml_backend_metal_buffer_type_get_max_size(ggml_backend_buffer_type_t buft) {
|
|
|
|
id<MTLDevice> device = ggml_backend_metal_get_device();
|
|
|
|
size_t max_size = device.maxBufferLength;
|
|
|
|
ggml_backend_metal_free_device();
|
|
|
|
|
|
|
|
return max_size;
|
|
|
|
|
|
|
|
UNUSED(buft);
|
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static bool ggml_backend_metal_buffer_type_supports_backend(ggml_backend_buffer_type_t buft, ggml_backend_t backend) {
|
2023-12-07 21:27:19 +01:00
|
|
|
return ggml_backend_is_metal(backend) || ggml_backend_is_cpu(backend);
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
UNUSED(buft);
|
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static bool ggml_backend_metal_buffer_type_is_host(ggml_backend_buffer_type_t buft) {
|
2023-12-22 16:53:39 +01:00
|
|
|
return true;
|
|
|
|
|
|
|
|
UNUSED(buft);
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL ggml_backend_buffer_type_t ggml_backend_metal_buffer_type(void) {
|
2023-12-07 21:27:19 +01:00
|
|
|
static struct ggml_backend_buffer_type ggml_backend_buffer_type_metal = {
|
|
|
|
/* .iface = */ {
|
2024-01-12 20:07:38 +01:00
|
|
|
/* .get_name = */ ggml_backend_metal_buffer_type_get_name,
|
2023-12-07 21:27:19 +01:00
|
|
|
/* .alloc_buffer = */ ggml_backend_metal_buffer_type_alloc_buffer,
|
|
|
|
/* .get_alignment = */ ggml_backend_metal_buffer_type_get_alignment,
|
2024-01-29 09:05:13 +01:00
|
|
|
/* .get_max_size = */ ggml_backend_metal_buffer_type_get_max_size,
|
2023-12-07 21:27:19 +01:00
|
|
|
/* .get_alloc_size = */ NULL, // defaults to ggml_nbytes
|
|
|
|
/* .supports_backend = */ ggml_backend_metal_buffer_type_supports_backend,
|
2023-12-22 16:53:39 +01:00
|
|
|
/* .is_host = */ ggml_backend_metal_buffer_type_is_host,
|
2023-12-07 21:27:19 +01:00
|
|
|
},
|
|
|
|
/* .context = */ NULL,
|
|
|
|
};
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2023-12-07 21:27:19 +01:00
|
|
|
return &ggml_backend_buffer_type_metal;
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
// buffer from ptr
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data, size_t size, size_t max_size) {
|
2023-12-22 16:53:39 +01:00
|
|
|
struct ggml_backend_metal_buffer_context * ctx = malloc(sizeof(struct ggml_backend_metal_buffer_context));
|
|
|
|
|
|
|
|
ctx->all_data = data;
|
|
|
|
ctx->all_size = size;
|
|
|
|
ctx->owned = false;
|
|
|
|
ctx->n_buffers = 0;
|
|
|
|
|
|
|
|
const size_t size_page = sysconf(_SC_PAGESIZE);
|
2024-01-12 20:07:38 +01:00
|
|
|
|
|
|
|
// page-align the data ptr
|
|
|
|
{
|
|
|
|
const uintptr_t offs = (uintptr_t) data % size_page;
|
|
|
|
data = (void *) ((char *) data - offs);
|
|
|
|
size += offs;
|
|
|
|
}
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
size_t size_aligned = size;
|
|
|
|
if ((size_aligned % size_page) != 0) {
|
|
|
|
size_aligned += (size_page - (size_aligned % size_page));
|
|
|
|
}
|
|
|
|
|
|
|
|
id<MTLDevice> device = ggml_backend_metal_get_device();
|
|
|
|
|
|
|
|
// the buffer fits into the max buffer size allowed by the device
|
|
|
|
if (size_aligned <= device.maxBufferLength) {
|
|
|
|
ctx->buffers[ctx->n_buffers].data = data;
|
|
|
|
ctx->buffers[ctx->n_buffers].size = size;
|
|
|
|
|
|
|
|
ctx->buffers[ctx->n_buffers].metal = [device newBufferWithBytesNoCopy:data length:size_aligned options:MTLResourceStorageModeShared deallocator:nil];
|
|
|
|
|
|
|
|
if (ctx->buffers[ctx->n_buffers].metal == nil) {
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: failed to allocate buffer, size = %8.2f MiB\n", __func__, size_aligned / 1024.0 / 1024.0);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
ggml_backend_metal_log_allocated_size(device, size_aligned);
|
2023-12-22 16:53:39 +01:00
|
|
|
|
|
|
|
++ctx->n_buffers;
|
|
|
|
} else {
|
|
|
|
// this overlap between the views will guarantee that the tensor with the maximum size will fully fit into
|
|
|
|
// one of the views
|
|
|
|
const size_t size_ovlp = ((max_size + size_page - 1) / size_page + 1) * size_page; // round-up 2 pages just in case
|
|
|
|
const size_t size_step = device.maxBufferLength - size_ovlp;
|
|
|
|
const size_t size_view = device.maxBufferLength;
|
|
|
|
|
|
|
|
for (size_t i = 0; i < size; i += size_step) {
|
|
|
|
const size_t size_step_aligned = (i + size_view <= size) ? size_view : (size_aligned - i);
|
|
|
|
|
|
|
|
ctx->buffers[ctx->n_buffers].data = (void *) ((uint8_t *) data + i);
|
|
|
|
ctx->buffers[ctx->n_buffers].size = size_step_aligned;
|
|
|
|
|
|
|
|
ctx->buffers[ctx->n_buffers].metal = [device newBufferWithBytesNoCopy:(void *) ((uint8_t *) data + i) length:size_step_aligned options:MTLResourceStorageModeShared deallocator:nil];
|
|
|
|
|
|
|
|
if (ctx->buffers[ctx->n_buffers].metal == nil) {
|
|
|
|
GGML_METAL_LOG_ERROR("%s: error: failed to allocate buffer, size = %8.2f MiB\n", __func__, size_step_aligned / 1024.0 / 1024.0);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
ggml : add Flash Attention (llama/5021)
* ggml : add ggml_flash_attn_ext API
* ggml : fix GQA support in ggml_flash_attn_ext
* ggml : online attention (CPU)
* metal : initial implementation
* metal : f16 precision
* metal : reduce branches
* metal : specialize for head size
* wip : 8 rows per simd group
* wip : 4 rows per simd group
* wip : template for rows per warp
* metal : parallelize across KV size
* metal : parallel reduce across heads
* metal : efficient flash_attn_f16 implementation
* metal : avoid redundant loads of the attention
* metal : scale and mask in matrix form
* metal : fix comment
* llama : avoid ggml_cast, use F32 query
* metal : add parallel reduce version (disabled)
* metal : move output into local memory + optimize
- the result from each simdgroup now stays in the registers
- significantly reduced SRAM usage
- more efficient skipping of -INF blocks
- avoid simdgroup barrier in hot loop
- add comments
* metal : add tests, fix scaling, support C > 32
* metal : improve precision
* ggml : fix f16 mad
* metal : minor
* metal : support Q > 8
* tests : add ATTN tests
* metal : disable buffer allocation logs
* tests : more
* metal : faster inner loop for C == 32
* metal : fix array initialization
* tests : ifdef
* ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext
* ggml : fix ggml_soft_max mask requirement
* cuda : fix soft_max to use correct mask size
* cuda : add flash_attn kernel (wip)
* metal : optimize softmax for C > 32
* metal : optimize softmax
* tests : minor fix
* cuda : avoid zeroing fragments
* tests : update dims
* cuda : fix __hisinf() result check
* cuda : avoid warp_reduce for smax
* cuda : use int instead of int64_t
Noticeably improves performance (thanks to Johannes)
* cuda : make loops use the same loop values
Thanks Johannes again for the tip
* cuda : unroll some of the loops
* cuda : avoid __hisinf branches
* cuda : use half2 in softmax
* cuda : switch to 1 warp for bs > 16
* cuda : speed-up reduce part of the kernel
* cuda : unroll Q*K^T loop
* cuda : fix -INF block check
* cuda : simplify softmax
* cuda : fix matrix names
* cuda : minor
* llama : adapt to F16 KQ_pos
* llama : adapt new models to F16 KQ_mask
* ggml : fix F16 store (ARM NEON)
* llama : fix type of KQ_mask and KQ_pos
* ggml : fix CPU soft_max
* tests : add hs=256
* cuda : fix build
* metal : improve perf via smaller int registers
* cuda : adapt soft_max to F16 mask and pos
* CUDA: faster FlashAttention, kernel for bs == 1
* 16 cols for Phi-2
* no vec for hs, no hs==256 ncols==32 for Volta
* adjust kernel selection logic
* 4 warps, 256 stride for all D
* no ncols == 64
* Multiple parallel blocks for batch size 1
* fix compile warnings
* fix excessive KQ_b loads
* fix cmake build
* fix KV cache padding, NaN from INFINITY (llama/6438)
* llama : flash_attn cparam + fix defrag
* server: support flash_attn param
* server: bench: enable flash_attn param
* CUDA: refactor host code, dyn. par. blocks
* fix flash_attn_vec_f16 race condition
* flush softmax exp below threshold to 0
* store temp KQ in registers
* Calculate KQ as FP32 if KQV has GGML_PREC_F32
* Add __hgt2_mask implementation for CUDA 11
* fix KQ FP32 precision fpr parallel_blocks > 1
* llama-bench : add -fa,--flash-attn arg
* metal : add BS=1 kernel for flash attention (llama/6508)
* metal : add BS=1 kernel for flash attention (wip)
* metal : support more than 1 warps
* metal : opts
* metal : opt
* metal : switch to parallel reduce
* metal : reduce registers
* metal : simplify
* metal : initial FA vec kernel
* metal : use F32 attention accumulators
* batched-bench : add fattn arg
* llama : simplify llama_build_kv_store
ggml-ci
* llama : adapt build_olmo to changes
* ggml : fix arm fp16 store on windows
* metal : clean-up
* metal : clean-up kernel code
* metal : minor
* tests : remove benchmarks
ggml-ci
* ggml : fix avx512 const correctness
ggml-ci
* ggml : fix soft_max with bias on CPU
ggml-ci
* common : print --flash-attn in help
* ggml : fix num dimensions in ggml_flash_attn_ext
* llama : force disable flash attention for incompatible models
* ggml : ggml_soft_max support F16/F32 mask/pos
ggml-ci
* cuda : uint -> uint32_t
* cuda : "constexpr dim3" -> "const dim3"
ggml-ci
* cuda : try to fix __hgt2_mask
ggml-ci
* ggml : add TODO's for F16/F32 mask/pos support in other backends
* llama : replace bool need_kq_pos with use_alibi
* llama : prep ALiBi support for BERT models
ggml-ci
* llama : fix n_batch requirements
ggml-ci
* cont
* server : add help for --flash-attn arg
* llama : disable FA for AMD
* tests : remove TMP_ATTN_BENCH
ggml-ci
* llama : support save/load state with FA enabled
ggml-ci
* ci : add CUDA save-load-state tests
ggml-ci
* llama : llama_kv_cache_clear zeroes data + fix save-load seq
ggml-ci
* llama : fix copy-paste errors, add TODO
* llama : disallow incompatible states
* llama : update llama_state_get_size after v_trans field
* metal : remove tmp log
* llama : add static reminder for llama_state_get_size
* metal : fix max nsg
ggml-ci
* ci : fix arg order
ggml-ci
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
2024-04-30 11:16:08 +02:00
|
|
|
ggml_backend_metal_log_allocated_size(device, size_step_aligned);
|
|
|
|
|
2023-12-22 16:53:39 +01:00
|
|
|
if (i + size_step < size) {
|
|
|
|
GGML_METAL_LOG_INFO("\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
++ctx->n_buffers;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return ggml_backend_buffer_init(ggml_backend_metal_buffer_type(), ggml_backend_metal_buffer_i, ctx, size);
|
|
|
|
}
|
|
|
|
|
|
|
|
// backend
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static const char * ggml_backend_metal_name(ggml_backend_t backend) {
|
2023-12-07 21:27:19 +01:00
|
|
|
return "Metal";
|
|
|
|
|
2023-11-03 20:35:05 +01:00
|
|
|
UNUSED(backend);
|
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static void ggml_backend_metal_free(ggml_backend_t backend) {
|
2023-12-07 21:27:19 +01:00
|
|
|
struct ggml_metal_context * ctx = (struct ggml_metal_context *)backend->context;
|
|
|
|
ggml_metal_free(ctx);
|
|
|
|
free(backend);
|
|
|
|
}
|
2023-11-03 20:35:05 +01:00
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static ggml_backend_buffer_type_t ggml_backend_metal_get_default_buffer_type(ggml_backend_t backend) {
|
2023-12-07 21:27:19 +01:00
|
|
|
return ggml_backend_metal_buffer_type();
|
2023-11-03 20:35:05 +01:00
|
|
|
|
|
|
|
UNUSED(backend);
|
|
|
|
}
|
|
|
|
|
2024-03-04 10:05:42 +01:00
|
|
|
GGML_CALL static enum ggml_status ggml_backend_metal_graph_compute(ggml_backend_t backend, struct ggml_cgraph * cgraph) {
|
2023-11-03 20:35:05 +01:00
|
|
|
struct ggml_metal_context * metal_ctx = (struct ggml_metal_context *)backend->context;
|
|
|
|
|
2024-01-03 14:39:43 +01:00
|
|
|
return ggml_metal_graph_compute(metal_ctx, cgraph);
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL static bool ggml_backend_metal_supports_op(ggml_backend_t backend, const struct ggml_tensor * op) {
|
2024-01-13 17:03:45 +01:00
|
|
|
struct ggml_metal_context * metal_ctx = (struct ggml_metal_context *)backend->context;
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-01-13 17:03:45 +01:00
|
|
|
return ggml_metal_supports_op(metal_ctx, op);
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
|
|
|
|
2024-01-12 20:07:38 +01:00
|
|
|
static struct ggml_backend_i ggml_backend_metal_i = {
|
2023-12-07 21:27:19 +01:00
|
|
|
/* .get_name = */ ggml_backend_metal_name,
|
|
|
|
/* .free = */ ggml_backend_metal_free,
|
|
|
|
/* .get_default_buffer_type = */ ggml_backend_metal_get_default_buffer_type,
|
|
|
|
/* .set_tensor_async = */ NULL,
|
|
|
|
/* .get_tensor_async = */ NULL,
|
2024-01-12 20:07:38 +01:00
|
|
|
/* .cpy_tensor_async = */ NULL,
|
2023-12-22 16:53:39 +01:00
|
|
|
/* .synchronize = */ NULL,
|
|
|
|
/* .graph_plan_create = */ NULL,
|
2023-12-07 21:27:19 +01:00
|
|
|
/* .graph_plan_free = */ NULL,
|
|
|
|
/* .graph_plan_compute = */ NULL,
|
|
|
|
/* .graph_compute = */ ggml_backend_metal_graph_compute,
|
|
|
|
/* .supports_op = */ ggml_backend_metal_supports_op,
|
2024-03-27 17:55:10 +01:00
|
|
|
/* .offload_op = */ NULL,
|
2024-03-13 18:54:21 +01:00
|
|
|
/* .event_new = */ NULL,
|
|
|
|
/* .event_free = */ NULL,
|
|
|
|
/* .event_record = */ NULL,
|
|
|
|
/* .event_wait = */ NULL,
|
|
|
|
/* .event_synchronize = */ NULL,
|
2023-11-03 20:35:05 +01:00
|
|
|
};
|
|
|
|
|
2024-01-13 19:45:45 +01:00
|
|
|
void ggml_backend_metal_log_set_callback(ggml_log_callback log_callback, void * user_data) {
|
|
|
|
ggml_metal_log_callback = log_callback;
|
|
|
|
ggml_metal_log_user_data = user_data;
|
|
|
|
}
|
|
|
|
|
2024-02-24 17:27:36 +01:00
|
|
|
static ggml_guid_t ggml_backend_metal_guid(void) {
|
|
|
|
static ggml_guid guid = { 0x81, 0xa1, 0x8b, 0x1e, 0x71, 0xec, 0x79, 0xed, 0x2b, 0x85, 0xdc, 0x8a, 0x61, 0x98, 0x30, 0xe6 };
|
|
|
|
return &guid;
|
|
|
|
}
|
|
|
|
|
2023-11-03 20:35:05 +01:00
|
|
|
ggml_backend_t ggml_backend_metal_init(void) {
|
2023-12-07 21:27:19 +01:00
|
|
|
struct ggml_metal_context * ctx = ggml_metal_init(GGML_DEFAULT_N_THREADS);
|
|
|
|
|
|
|
|
if (ctx == NULL) {
|
|
|
|
return NULL;
|
|
|
|
}
|
2023-11-03 20:35:05 +01:00
|
|
|
|
|
|
|
ggml_backend_t metal_backend = malloc(sizeof(struct ggml_backend));
|
|
|
|
|
|
|
|
*metal_backend = (struct ggml_backend) {
|
2024-02-24 17:27:36 +01:00
|
|
|
/* .guid = */ ggml_backend_metal_guid(),
|
2024-01-12 20:07:38 +01:00
|
|
|
/* .interface = */ ggml_backend_metal_i,
|
2023-11-03 20:35:05 +01:00
|
|
|
/* .context = */ ctx,
|
|
|
|
};
|
|
|
|
|
|
|
|
return metal_backend;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool ggml_backend_is_metal(ggml_backend_t backend) {
|
2024-02-24 17:27:36 +01:00
|
|
|
return backend != NULL && ggml_guid_matches(backend->guid, ggml_backend_metal_guid());
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
void ggml_backend_metal_set_n_cb(ggml_backend_t backend, int n_cb) {
|
2023-12-07 21:27:19 +01:00
|
|
|
GGML_ASSERT(ggml_backend_is_metal(backend));
|
|
|
|
|
2023-11-03 20:35:05 +01:00
|
|
|
struct ggml_metal_context * ctx = (struct ggml_metal_context *)backend->context;
|
|
|
|
|
2024-01-13 19:45:45 +01:00
|
|
|
ctx->n_cb = MIN(n_cb, GGML_METAL_MAX_BUFFERS);
|
2023-11-03 20:35:05 +01:00
|
|
|
}
|
2023-11-24 11:37:08 +01:00
|
|
|
|
|
|
|
bool ggml_backend_metal_supports_family(ggml_backend_t backend, int family) {
|
2023-12-07 21:27:19 +01:00
|
|
|
GGML_ASSERT(ggml_backend_is_metal(backend));
|
|
|
|
|
2023-11-24 11:37:08 +01:00
|
|
|
struct ggml_metal_context * ctx = (struct ggml_metal_context *)backend->context;
|
|
|
|
|
2023-12-07 21:27:19 +01:00
|
|
|
return [ctx->device supportsFamily:(MTLGPUFamilyApple1 + family - 1)];
|
|
|
|
}
|
|
|
|
|
2024-01-29 10:22:23 +01:00
|
|
|
void ggml_backend_metal_capture_next_compute(ggml_backend_t backend) {
|
|
|
|
GGML_ASSERT(ggml_backend_is_metal(backend));
|
|
|
|
|
|
|
|
struct ggml_metal_context * ctx = (struct ggml_metal_context *)backend->context;
|
|
|
|
ctx->should_capture_next_compute = true;
|
|
|
|
}
|
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL ggml_backend_t ggml_backend_reg_metal_init(const char * params, void * user_data); // silence warning
|
2023-12-07 21:27:19 +01:00
|
|
|
|
2024-01-16 12:16:33 +01:00
|
|
|
GGML_CALL ggml_backend_t ggml_backend_reg_metal_init(const char * params, void * user_data) {
|
2023-12-07 21:27:19 +01:00
|
|
|
return ggml_backend_metal_init();
|
|
|
|
|
|
|
|
GGML_UNUSED(params);
|
|
|
|
GGML_UNUSED(user_data);
|
2023-11-24 11:37:08 +01:00
|
|
|
}
|