Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vulkan: 64-bit im2col (#16135)

Browse Source 
* vulkan: 64-bit im2col

Add variants of the im2col shaders that use buffer_device_address/buffer_reference,
and use 64-bit address calculations. This is needed for large convolutions used in
stable-diffusion.cpp.

* fix validation error for large im2col
This commit is contained in:
Jeff Bolz
2025-09-28 08:38:37 +02:00
committed by GitHub
parent 6a2c6145a0
commit d8359f5fde
6 changed files with 117 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp
+18 -4
View File
@@ -6,8 +6,11 @@
#include "rte.comp"
#include "types.comp"
layout (push_constant) uniform parameter
{
BDA_STORAGE_T dst_addr;
uint32_t nb10;
uint32_t nb11;
uint32_t nb12;
@@ -38,8 +41,6 @@ layout (push_constant) uniform parameter
uint32_t misalign_offsets;
} p;
#include "types.comp"
uint get_aoffset() { return p.misalign_offsets >> 16; }
uint get_doffset() { return p.misalign_offsets & 0xFFFF; }
@@ -50,6 +51,10 @@ 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[];};
#if BDA
layout (buffer_reference) buffer D_ptr {D_TYPE d;};
#endif
void main() {
const uint32_t i = gl_GlobalInvocationID.x;
@@ -100,13 +105,22 @@ void main() {
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;
const BDA_OFFSET_T offset_dst = BDA_OFFSET_T(in_)*OD_OH_OW_IC_KD_KH_KW + BDA_OFFSET_T(iod)*OH_OW_IC_KD_KH_KW + BDA_OFFSET_T(ioh)*OW_IC_KD_KH_KW + BDA_OFFSET_T(iow)*IC_KD_KH_KW + iic*KD_KH_KW + ikd * KH_KW + ikh*KW + ikw;
const uint32_t offset_src = (in_*IC + iic)*nb13 + iid*nb12 + iih*nb11 + iiw*nb10;
#if BDA
D_ptr dst_addr = D_ptr(p.dst_addr + D_SIZE * offset_dst);
if (iih >= IH || iiw >= IW || iid >= ID) {
dst_addr.d = D_TYPE(0.0f);
} else {
dst_addr.d = D_TYPE(data_a[offset_src + get_aoffset()]);
}
#else
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()]);
}
#endif
}
}