extern/whisper.cpp
1
0
Fork 1
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-08-10 09:28:04 +02:00
Code Issues Packages Projects Releases Wiki Activity

sync : resolve conflicts (ggml/0)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-07-12 14:39:52 +03:00
parent 915fc153a5
commit 85dcc74b88
6 changed files with 195 additions and 113 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
ggml/CMakeLists.txt
View File

@ -360,6 +360,13 @@ write_basic_package_version_file(
VERSION ${GGML_INSTALL_VERSION}
COMPATIBILITY SameMajorVersion)
target_compile_definitions(ggml-base PRIVATE
GGML_VERSION="${GGML_INSTALL_VERSION}"
GGML_COMMIT="${GGML_BUILD_COMMIT}"
)
message(STATUS "ggml version: ${GGML_INSTALL_VERSION}")
message(STATUS "ggml commit: ${GGML_BUILD_COMMIT}")
install(FILES ${CMAKE_CURRENT_BINARY_DIR}/ggml-config.cmake
${CMAKE_CURRENT_BINARY_DIR}/ggml-version.cmake
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/ggml)

3
ggml/include/ggml.h
View File

@ -648,6 +648,9 @@ extern "C" {
// misc
GGML_API const char * ggml_version(void);
GGML_API const char * ggml_commit(void);
GGML_API void ggml_time_init(void); // call this once at the beginning of the program
GGML_API int64_t ggml_time_ms(void);
GGML_API int64_t ggml_time_us(void);

214
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -425,13 +425,14 @@ struct vk_device_struct {
vk_pipeline pipeline_div_norepeat[2][2][2];
vk_pipeline pipeline_concat_f32, pipeline_concat_f16, pipeline_concat_i32;
vk_pipeline pipeline_upscale_f32;
vk_pipeline pipeline_upscale_nearest_f32, pipeline_upscale_bilinear_f32, pipeline_upscale_bilinear_ac_f32;
vk_pipeline pipeline_scale_f32;
vk_pipeline pipeline_sqr_f32;
vk_pipeline pipeline_sin_f32;
vk_pipeline pipeline_cos_f32;
vk_pipeline pipeline_clamp_f32;
vk_pipeline pipeline_pad_f32;
vk_pipeline pipeline_roll_f32;
vk_pipeline pipeline_repeat_f32, pipeline_repeat_back_f32;
vk_pipeline pipeline_cpy_f32_f32, pipeline_cpy_f32_f16, pipeline_cpy_f16_f16, pipeline_cpy_f16_f32, pipeline_cpy_f32_bf16;
vk_pipeline pipeline_contig_cpy_f32_f32, pipeline_contig_cpy_f32_f16, pipeline_contig_cpy_f16_f16, pipeline_contig_cpy_f16_f32, pipeline_contig_cpy_f32_bf16;
@ -694,6 +695,37 @@ struct vk_op_unary_push_constants {
};
static_assert(sizeof(vk_op_unary_push_constants) <= 128, "sizeof(vk_op_unary_push_constants) must be <= 128");
static vk_op_unary_push_constants vk_op_unary_push_constants_init(const ggml_tensor * src0, const ggml_tensor * dst, int64_t ne = 0) {
GGML_ASSERT(ne != 0 || (ggml_nelements(src0) == ggml_nelements(dst)));
ne = ne != 0 ? ne : ggml_nelements(dst);
GGML_ASSERT(ne <= (int64_t)std::numeric_limits<uint32_t>::max());
vk_op_unary_push_constants p{};
p.ne = (uint32_t)ne;
size_t src0_tsize = ggml_type_size(src0->type);
p.ne00 = (uint32_t)src0->ne[0];
p.ne01 = (uint32_t)src0->ne[1];
p.ne02 = (uint32_t)src0->ne[2];
p.ne03 = (uint32_t)src0->ne[3];
p.nb00 = (uint32_t)(src0->nb[0] / src0_tsize);
p.nb01 = (uint32_t)(src0->nb[1] / src0_tsize);
p.nb02 = (uint32_t)(src0->nb[2] / src0_tsize);
p.nb03 = (uint32_t)(src0->nb[3] / src0_tsize);
size_t dst_tsize = ggml_type_size(dst->type);
p.ne10 = (uint32_t)dst->ne[0];
p.ne11 = (uint32_t)dst->ne[1];
p.ne12 = (uint32_t)dst->ne[2];
p.ne13 = (uint32_t)dst->ne[3];
p.nb10 = (uint32_t)(dst->nb[0] / dst_tsize);
p.nb11 = (uint32_t)(dst->nb[1] / dst_tsize);
p.nb12 = (uint32_t)(dst->nb[2] / dst_tsize);
p.nb13 = (uint32_t)(dst->nb[3] / dst_tsize);
return p; // fastdiv values and offsets are initialized later in ggml_vk_op
}
// See https://gmplib.org/~tege/divcnst-pldi94.pdf figure 4.1.
// Precompute mp (m' in the paper) and L such that division
// can be computed using a multiply (high 32b of 64b result)
@ -863,6 +895,7 @@ struct vk_op_conv2d_dw_push_constants {
struct vk_op_upscale_push_constants {
uint32_t ne; uint32_t a_offset; uint32_t d_offset;
uint32_t ne00; uint32_t ne01;
uint32_t nb00; uint32_t nb01; uint32_t nb02; uint32_t nb03;
uint32_t ne10; uint32_t ne11; uint32_t ne12; uint32_t ne13;
float sf0; float sf1; float sf2; float sf3;
@ -2824,7 +2857,9 @@ static void ggml_vk_load_shaders(vk_device& device) {
ggml_vk_create_pipeline(device, device->pipeline_concat_f16, "concat_f16", concat_f16_len, concat_f16_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_concat_i32, "concat_i32", concat_i32_len, concat_i32_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_upscale_f32, "upscale_f32", upscale_f32_len, upscale_f32_data, "main", 2, sizeof(vk_op_upscale_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_upscale_nearest_f32, "upscale_f32", upscale_f32_len, upscale_f32_data, "main", 2, sizeof(vk_op_upscale_push_constants), {512, 1, 1}, {GGML_SCALE_MODE_NEAREST}, 1);
ggml_vk_create_pipeline(device, device->pipeline_upscale_bilinear_f32, "upscale_f32", upscale_f32_len, upscale_f32_data, "main", 2, sizeof(vk_op_upscale_push_constants), {512, 1, 1}, {GGML_SCALE_MODE_BILINEAR}, 1);
ggml_vk_create_pipeline(device, device->pipeline_upscale_bilinear_ac_f32, "upscale_f32", upscale_f32_len, upscale_f32_data, "main", 2, sizeof(vk_op_upscale_push_constants), {512, 1, 1}, {GGML_SCALE_MODE_BILINEAR | GGML_SCALE_FLAG_ALIGN_CORNERS}, 1);
ggml_vk_create_pipeline(device, device->pipeline_scale_f32, "scale_f32", scale_f32_len, scale_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
@ -2836,6 +2871,8 @@ static void ggml_vk_load_shaders(vk_device& device) {
ggml_vk_create_pipeline(device, device->pipeline_pad_f32, "pad_f32", pad_f32_len, pad_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_roll_f32, "roll_f32", roll_f32_len, roll_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_repeat_f32, "repeat_f32", repeat_f32_len, repeat_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_repeat_back_f32, "repeat_back_f32", repeat_back_f32_len, repeat_back_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
@ -6502,8 +6539,16 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
}
return nullptr;
case GGML_OP_UPSCALE:
if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32 && dst->op_params[0] == GGML_SCALE_MODE_NEAREST) {
return ctx->device->pipeline_upscale_f32;
if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
int mode = ggml_get_op_params_i32(dst, 0);
switch (mode) {
case GGML_SCALE_MODE_NEAREST:
return ctx->device->pipeline_upscale_nearest_f32;
case GGML_SCALE_MODE_BILINEAR:
return ctx->device->pipeline_upscale_bilinear_f32;
case GGML_SCALE_MODE_BILINEAR | GGML_SCALE_FLAG_ALIGN_CORNERS:
return ctx->device->pipeline_upscale_bilinear_ac_f32;
}
}
return nullptr;
case GGML_OP_SCALE:
@ -6536,6 +6581,11 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
return ctx->device->pipeline_pad_f32;
}
return nullptr;
case GGML_OP_ROLL:
if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
return ctx->device->pipeline_roll_f32;
}
return nullptr;
case GGML_OP_REPEAT:
if (ggml_type_size(src0->type) == sizeof(float) && ggml_type_size(dst->type) == sizeof(float)) {
return ctx->device->pipeline_repeat_f32;
@ -7085,6 +7135,7 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
case GGML_OP_COS:
case GGML_OP_CLAMP:
case GGML_OP_PAD:
case GGML_OP_ROLL:
case GGML_OP_REPEAT:
case GGML_OP_REPEAT_BACK:
case GGML_OP_CPY:
@ -7546,14 +7597,21 @@ static void ggml_vk_concat(ggml_backend_vk_context * ctx, vk_context& subctx, co
static void ggml_vk_upscale(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t mode = (uint32_t)ggml_get_op_params_i32(dst, 0);
const float sf0 = (float)dst->ne[0] / src0->ne[0];
const float sf1 = (float)dst->ne[1] / src0->ne[1];
const float sf2 = (float)dst->ne[2] / src0->ne[2];
const float sf3 = (float)dst->ne[3] / src0->ne[3];
float sf0 = (float)dst->ne[0] / src0->ne[0];
float sf1 = (float)dst->ne[1] / src0->ne[1];
float sf2 = (float)dst->ne[2] / src0->ne[2];
float sf3 = (float)dst->ne[3] / src0->ne[3];
if (mode & GGML_SCALE_FLAG_ALIGN_CORNERS) {
sf0 = (float)(dst->ne[0] - 1) / (src0->ne[0] - 1);
sf1 = (float)(dst->ne[1] - 1) / (src0->ne[1] - 1);
}
ggml_vk_op_f32<vk_op_upscale_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_UPSCALE, {
(uint32_t)ggml_nelements(dst), 0, 0,
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1],
(uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t)dst->ne[0], (uint32_t)dst->ne[1], (uint32_t)dst->ne[2],(uint32_t)dst->ne[3],
sf0, sf1, sf2, sf3,
@ -7561,123 +7619,64 @@ static void ggml_vk_upscale(ggml_backend_vk_context * ctx, vk_context& subctx, c
}
static void ggml_vk_scale(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
float * op_params = (float *)dst->op_params;
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
vk_op_unary_push_constants p = vk_op_unary_push_constants_init(src0, dst);
p.param1 = ggml_get_op_params_f32(dst, 0);
p.param2 = ggml_get_op_params_f32(dst, 1);
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_SCALE, {
(uint32_t)ggml_nelements(src0),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
op_params[0], op_params[1],
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
}, dryrun);
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_SCALE, std::move(p), dryrun);
}
static void ggml_vk_sqr(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_SQR, {
(uint32_t)ggml_nelements(src0),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
0.0f, 0.0f,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
}, dryrun);
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_SQR, vk_op_unary_push_constants_init(src0, dst), dryrun);
}
static void ggml_vk_sin(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_SIN, {
(uint32_t)ggml_nelements(src0),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
0.0f, 0.0f,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
}, dryrun);
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_SIN, vk_op_unary_push_constants_init(src0, dst), dryrun);
}
static void ggml_vk_cos(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_COS, {
(uint32_t)ggml_nelements(src0),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
0.0f, 0.0f,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
}, dryrun);
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_COS, vk_op_unary_push_constants_init(src0, dst), dryrun);
}
static void ggml_vk_clamp(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
float * op_params = (float *)dst->op_params;
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
vk_op_unary_push_constants p = vk_op_unary_push_constants_init(src0, dst);
p.param1 = ggml_get_op_params_f32(dst, 0);
p.param2 = ggml_get_op_params_f32(dst, 1);
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_CLAMP, {
(uint32_t)ggml_nelements(src0),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
op_params[0], op_params[1],
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
}, dryrun);
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_CLAMP, std::move(p), dryrun);
}
static void ggml_vk_pad(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
vk_op_unary_push_constants p = vk_op_unary_push_constants_init(src0, dst, ggml_nelements(dst));
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_PAD, std::move(p), dryrun);
}
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_PAD, {
(uint32_t)ggml_nelements(dst),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
0.0f, 0.0f,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
}, dryrun);
static void ggml_vk_roll(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
const int32_t s0 = ggml_get_op_params_i32(dst, 0);
const int32_t s1 = ggml_get_op_params_i32(dst, 1);
const int32_t s2 = ggml_get_op_params_i32(dst, 2);
const int32_t s3 = ggml_get_op_params_i32(dst, 3);
const uint32_t s01_packed = ((s0 + 0x8000) << 16) | (s1 + 0x8000);
const uint32_t s23_packed = ((s2 + 0x8000) << 16) | (s3 + 0x8000);
vk_op_unary_push_constants p = vk_op_unary_push_constants_init(src0, dst);
memcpy(&p.param1, &s01_packed, sizeof(float));
memcpy(&p.param2, &s23_packed, sizeof(float));
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_ROLL, std::move(p), dryrun);
}
static void ggml_vk_repeat(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_REPEAT, {
(uint32_t)ggml_nelements(dst),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
0.0f, 0.0f,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
}, dryrun);
vk_op_unary_push_constants p = vk_op_unary_push_constants_init(src0, dst, ggml_nelements(dst));
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_REPEAT, std::move(p), dryrun);
}
static void ggml_vk_repeat_back(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_REPEAT_BACK, {
(uint32_t)ggml_nelements(dst),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
0.0f, 0.0f,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
}, dryrun);
vk_op_unary_push_constants p = vk_op_unary_push_constants_init(src0, dst, ggml_nelements(dst));
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_REPEAT_BACK, std::move(p), dryrun);
}
static void ggml_vk_cpy(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
uint32_t ne = (uint32_t)ggml_nelements(src0);
if (ggml_is_quantized(src0->type) && ggml_is_quantized(dst->type)) {
// Convert from number of logical elements to 2- or 4-byte units.
@ -7689,14 +7688,8 @@ static void ggml_vk_cpy(ggml_backend_vk_context * ctx, vk_context& subctx, const
}
}
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_CPY, {
ne,
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
0.0f, 0.0f,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
}, dryrun);
vk_op_unary_push_constants p = vk_op_unary_push_constants_init(src0, dst, ne);
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_CPY, std::move(p), dryrun);
}
static void ggml_vk_set_rows(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, bool dryrun = false) {
@ -9033,6 +9026,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
case GGML_OP_COS:
case GGML_OP_CLAMP:
case GGML_OP_PAD:
case GGML_OP_ROLL:
case GGML_OP_CPY:
case GGML_OP_SET_ROWS:
case GGML_OP_CONT:
@ -9204,6 +9198,10 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
case GGML_OP_PAD:
ggml_vk_pad(ctx, compute_ctx, src0, node, dryrun);
break;
case GGML_OP_ROLL:
ggml_vk_roll(ctx, compute_ctx, src0, node, dryrun);
break;
case GGML_OP_CPY:
case GGML_OP_CONT:
@ -9428,6 +9426,7 @@ static bool ggml_vk_compute_forward(ggml_backend_vk_context * ctx, ggml_cgraph *
case GGML_OP_COS:
case GGML_OP_CLAMP:
case GGML_OP_PAD:
case GGML_OP_ROLL:
case GGML_OP_CPY:
case GGML_OP_SET_ROWS:
case GGML_OP_CONT:
@ -10594,13 +10593,12 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
case GGML_OP_CLAMP:
return op->src[0]->type == GGML_TYPE_F32;
case GGML_OP_UPSCALE:
return op->op_params[0] == GGML_SCALE_MODE_NEAREST;
case GGML_OP_ACC:
case GGML_OP_CONCAT:
case GGML_OP_SCALE:
case GGML_OP_PAD:
case GGML_OP_ROLL:
case GGML_OP_DIAG_MASK_INF:
return true;
case GGML_OP_SOFT_MAX:
case GGML_OP_SOFT_MAX_BACK:
case GGML_OP_ARGSORT:

74
ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp
View File

@ -3,6 +3,7 @@
layout (push_constant) uniform parameter
{
uint ne; uint a_offset; uint d_offset;
uint ne00; uint ne01;
uint nb00; uint nb01; uint nb02; uint nb03;
uint ne10; uint ne11; uint ne12; uint ne13;
float sf0; float sf1; float sf2; float sf3;
@ -15,6 +16,61 @@ layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
layout (binding = 0) readonly buffer A {A_TYPE data_a[];};
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
// from ggml.h: enum ggml_scale_mode, enum ggml_scale_flag
#define NEAREST 0
#define BILINEAR 1
#define ALIGN_CORNERS (1 << 8)
layout (constant_id = 0) const uint scale_mode = 0;
float fetch_nearest(uint i10, uint i11, uint i12, uint i13) {
const uint i00 = uint(i10 / p.sf0);
const uint i01 = uint(i11 / p.sf1);
const uint i02 = uint(i12 / p.sf2);
const uint i03 = uint(i13 / p.sf3);
return data_a[p.a_offset + i03 * p.nb03 + i02 * p.nb02 + i01 * p.nb01 + i00 * p.nb00];
}
float fetch_bilinear(ivec2 c0, ivec2 c1, vec2 d, uint i12, uint i13) {
const uint i02 = uint(i12 / p.sf2);
const uint i03 = uint(i13 / p.sf3);
const uint base = p.a_offset + i03 * p.nb03 + i02 * p.nb02;
const float v00 = data_a[base + c0.y * p.nb01 + c0.x * p.nb00];
const float v01 = data_a[base + c0.y * p.nb01 + c1.x * p.nb00];
const float v10 = data_a[base + c1.y * p.nb01 + c0.x * p.nb00];
const float v11 = data_a[base + c1.y * p.nb01 + c1.x * p.nb00];
return
v00 * (1.0-d.x) * (1.0-d.y) +
v01 * d.x * (1.0-d.y) +
v10 * (1.0-d.x) * d.y +
v11 * d.x * d.y;
}
float interpolate_bilinear(uint i10, uint i11, uint i12, uint i13) {
const ivec2 ne0 = ivec2(p.ne00, p.ne01);
const vec2 c = (vec2(i10, i11) + 0.5) / vec2(p.sf0, p.sf1) - 0.5;
const vec2 c0f = floor(c);
const vec2 d = c - c0f;
const ivec2 c0 = max(ivec2(c0f), 0);
const ivec2 c1 = min(ivec2(c0f + 1), ne0 - 1);
return fetch_bilinear(c0, c1, d, i12, i13);
}
float interpolate_bilinear_align_corners(uint i10, uint i11, uint i12, uint i13) {
const vec2 c = vec2(i10, i11) / vec2(p.sf0, p.sf1);
const vec2 c0f = floor(c);
const vec2 d = c - c0f;
const ivec2 c0 = ivec2(c0f);
const ivec2 c1 = c0 + 1;
return fetch_bilinear(c0, c1, d, i12, i13);
}
void main() {
const uint idx = gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
@ -27,10 +83,18 @@ void main() {
const uint i12 = (idx / (p.ne10 * p.ne11)) % p.ne12;
const uint i13 = (idx / (p.ne10 * p.ne11 * p.ne12)) % p.ne13;
const uint i00 = uint(i10 / p.sf0);
const uint i01 = uint(i11 / p.sf1);
const uint i02 = uint(i12 / p.sf2);
const uint i03 = uint(i13 / p.sf3);
float result;
switch (scale_mode) {
case NEAREST:
result = fetch_nearest(i10, i11, i12, i13);
break;
case BILINEAR:
result = interpolate_bilinear(i10, i11, i12, i13);
break;
case BILINEAR | ALIGN_CORNERS:
result = interpolate_bilinear_align_corners(i10, i11, i12, i13);
break;
}
data_d[p.d_offset + idx] = D_TYPE(data_a[p.a_offset + i03 * p.nb03 + i02 * p.nb02 + i01 * p.nb01 + i00 * p.nb00]);
data_d[p.d_offset + idx] = D_TYPE(result);
}

2
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
View File

@ -653,6 +653,8 @@ void process_shaders() {
string_to_spv("conv2d_dw_whcn_f32", "conv2d_dw.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"WHCN", "1"}}));
string_to_spv("conv2d_dw_cwhn_f32", "conv2d_dw.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"CWHN", "1"}}));
string_to_spv("roll_f32", "roll.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
for (auto &c : compiles) {
c.wait();
}

8
ggml/src/ggml.c
View File

@ -473,6 +473,14 @@ bool ggml_guid_matches(ggml_guid_t guid_a, ggml_guid_t guid_b) {
return memcmp(guid_a, guid_b, sizeof(ggml_guid)) == 0;
}
const char * ggml_version(void) {
return GGML_VERSION;
}
const char * ggml_commit(void) {
return GGML_COMMIT;
}
//
// timing
//