root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-12 03:31:46 +00:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

Refactor Vulkan backend to allow multiple contexts (#7961)

Browse Source 
* Refactor Vulkan backend to allow multiple contexts

* Fix too many shader groups called validation error in llama3 on AMD and Intel GPUs

* Fix Vulkan debug build error
This commit is contained in:
0cc4m 2024-06-23 10:21:25 +02:00 committed by GitHub
parent b5a5f34efa
commit 45c0e2e4c1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
8 changed files with 21021 additions and 20675 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39661
ggml-vulkan-shaders.hpp
View File

File diff suppressed because it is too large Load Diff

2017
ggml-vulkan.cpp
View File

File diff suppressed because it is too large Load Diff

2
vulkan-shaders/mul_mat_vec.comp
View File

@ -13,7 +13,7 @@ layout (constant_id = 0) const uint BLOCK_SIZE = 32;
shared FLOAT_TYPE tmp[BLOCK_SIZE]; shared FLOAT_TYPE tmp[BLOCK_SIZE];
void main() { void main() {
const uint row = gl_WorkGroupID.x; const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
const uint tid = gl_LocalInvocationID.x; const uint tid = gl_LocalInvocationID.x;
uint a_offset, b_offset, d_offset; uint a_offset, b_offset, d_offset;

2
vulkan-shaders/mul_mat_vec_q2_k.comp
View File

@ -7,7 +7,7 @@ layout(local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
shared FLOAT_TYPE tmp[32]; shared FLOAT_TYPE tmp[32];
void main() { void main() {
const uint row = gl_WorkGroupID.x; const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
uint a_offset, b_offset, d_offset; uint a_offset, b_offset, d_offset;
get_offsets(a_offset, b_offset, d_offset); get_offsets(a_offset, b_offset, d_offset);

2
vulkan-shaders/mul_mat_vec_q3_k.comp
View File

@ -7,7 +7,7 @@ layout(local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
shared FLOAT_TYPE tmp[32]; shared FLOAT_TYPE tmp[32];
void main() { void main() {
const uint row = gl_WorkGroupID.x; const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
uint a_offset, b_offset, d_offset; uint a_offset, b_offset, d_offset;
get_offsets(a_offset, b_offset, d_offset); get_offsets(a_offset, b_offset, d_offset);

2
vulkan-shaders/mul_mat_vec_q4_k.comp
View File

@ -7,7 +7,7 @@ layout(local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
shared FLOAT_TYPE tmp[32]; shared FLOAT_TYPE tmp[32];
void main() { void main() {
const uint row = gl_WorkGroupID.x; const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
uint a_offset, b_offset, d_offset; uint a_offset, b_offset, d_offset;
get_offsets(a_offset, b_offset, d_offset); get_offsets(a_offset, b_offset, d_offset);

2
vulkan-shaders/mul_mat_vec_q5_k.comp
View File

@ -7,7 +7,7 @@ layout(local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
shared FLOAT_TYPE tmp[32]; shared FLOAT_TYPE tmp[32];
void main() { void main() {
const uint row = gl_WorkGroupID.x; const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
uint a_offset, b_offset, d_offset; uint a_offset, b_offset, d_offset;
get_offsets(a_offset, b_offset, d_offset); get_offsets(a_offset, b_offset, d_offset);

2
vulkan-shaders/mul_mat_vec_q6_k.comp
View File

@ -7,7 +7,7 @@ layout(local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
shared FLOAT_TYPE tmp[32]; shared FLOAT_TYPE tmp[32];
void main() { void main() {
const uint row = gl_WorkGroupID.x; const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
uint a_offset, b_offset, d_offset; uint a_offset, b_offset, d_offset;
get_offsets(a_offset, b_offset, d_offset); get_offsets(a_offset, b_offset, d_offset);