CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-11 10:36:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: support im2col_3d (#15795)

Browse Source
This commit is contained in:
Jeff Bolz
2025-09-07 13:50:26 -05:00
committed by GitHub
parent d36e61c580
commit 3976dfbe00
4 changed files with 290 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

112
ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp Normal file
View File

@@ -0,0 +1,112 @@
#version 450
#extension GL_EXT_shader_16bit_storage : require
#extension GL_EXT_control_flow_attributes : require
#extension GL_EXT_shader_explicit_arithmetic_types_int32 : require
#include "rte.comp"
layout (push_constant) uniform parameter
{
uint32_t nb10;
uint32_t nb11;
uint32_t nb12;
uint32_t nb13;
uint32_t s0;
uint32_t s1;
uint32_t s2;
uint32_t p0;
uint32_t p1;
uint32_t p2;
uint32_t d0;
uint32_t d1;
uint32_t d2;
uint32_t IW;
uint32_t IH;
uint32_t ID;
uint32_t IC;
uint32_t KW;
uint32_t OH;
uint32_t KD_KH_KW;
uint32_t KH_KW;
uint32_t IC_KD_KH_KW;
uint32_t N_OD_OH;
uint32_t OD_OH;
uint32_t OD_OH_OW_IC_KD_KH_KW;
uint32_t OH_OW_IC_KD_KH_KW;
uint32_t OW_IC_KD_KH_KW;
uint32_t misalign_offsets;
} p;
#include "types.comp"
uint get_aoffset() { return p.misalign_offsets >> 16; }
uint get_doffset() { return p.misalign_offsets & 0xFFFF; }
layout(constant_id = 0) const uint BLOCK_SIZE = 32;
layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
layout (binding = 0) readonly buffer X {A_TYPE data_a[];};
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
void main() {
const uint32_t i = gl_GlobalInvocationID.x;
uint32_t nb10 = p.nb10;
uint32_t nb11 = p.nb11;
uint32_t nb12 = p.nb12;
uint32_t nb13 = p.nb13;
uint32_t s0 = p.s0;
uint32_t s1 = p.s1;
uint32_t s2 = p.s2;
uint32_t p0 = p.p0;
uint32_t p1 = p.p1;
uint32_t p2 = p.p2;
uint32_t d0 = p.d0;
uint32_t d1 = p.d1;
uint32_t d2 = p.d2;
uint32_t IW = p.IW;
uint32_t IH = p.IH;
uint32_t ID = p.ID;
uint32_t IC = p.IC;
uint32_t KW = p.KW;
uint32_t OH = p.OH;
uint32_t KD_KH_KW = p.KD_KH_KW;
uint32_t KH_KW = p.KH_KW;
uint32_t IC_KD_KH_KW = p.IC_KD_KH_KW;
uint32_t N_OD_OH = p.N_OD_OH;
uint32_t OD_OH = p.OD_OH;
uint32_t OD_OH_OW_IC_KD_KH_KW = p.OD_OH_OW_IC_KD_KH_KW;
uint32_t OH_OW_IC_KD_KH_KW = p.OH_OW_IC_KD_KH_KW;
uint32_t OW_IC_KD_KH_KW = p.OW_IC_KD_KH_KW;
if (i >= IC_KD_KH_KW) {
return;
}
const uint32_t iic = i / KD_KH_KW;
const uint32_t ikd = (i - iic * KD_KH_KW) / KH_KW;
const uint32_t ikh = (i - iic * KD_KH_KW - ikd * KH_KW) / KW;
const uint32_t ikw = i % KW;
const uint32_t iow = gl_GlobalInvocationID.y;
for (uint32_t iz = gl_GlobalInvocationID.z; iz < N_OD_OH; iz += gl_NumWorkGroups.z) {
const uint32_t in_ = iz / OD_OH;
const uint32_t iod = (iz - in_*OD_OH) / OH;
const uint32_t ioh = iz % OH;
const uint32_t iiw = iow * s0 + ikw * d0 - p0;
const uint32_t iih = ioh * s1 + ikh * d1 - p1;
const uint32_t iid = iod * s2 + ikd * d2 - p2;
const uint32_t offset_dst = in_*OD_OH_OW_IC_KD_KH_KW + iod*OH_OW_IC_KD_KH_KW + ioh*OW_IC_KD_KH_KW + iow*IC_KD_KH_KW + iic*KD_KH_KW + ikd * KH_KW + ikh*KW + ikw;
if (iih >= IH || iiw >= IW || iid >= ID) {
data_d[offset_dst + get_doffset()] = D_TYPE(0.0f);
} else {
const uint32_t offset_src = (in_*IC + iic)*nb13 + iid*nb12 + iih*nb11 + iiw*nb10;
data_d[offset_dst + get_doffset()] = D_TYPE(data_a[offset_src + get_aoffset()]);
}
}
}

4
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
View File

@@ -713,6 +713,10 @@ void process_shaders() {
string_to_spv("im2col_f32_f16", "im2col.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}}));
string_to_spv("im2col_f32_f16_rte", "im2col.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}, {"RTE16", "1"}}));
string_to_spv("im2col_3d_f32", "im2col_3d.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
string_to_spv("im2col_3d_f32_f16", "im2col_3d.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}}));
string_to_spv("im2col_3d_f32_f16_rte", "im2col_3d.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}, {"RTE16", "1"}}));
string_to_spv("timestep_embedding_f32", "timestep_embedding.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
string_to_spv("conv_transpose_1d_f32", "conv_transpose_1d.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}});