ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-24 23:24:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

Attempt fix 9

Browse Source
This commit is contained in:
Saood Karim
2025-04-20 22:08:03 -05:00
parent 9aa1b15742
commit 44ef87f07a
4 changed files with 18 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/CMakeLists.txt
View File

@@ -211,8 +211,7 @@ set(GGML_PUBLIC_HEADERS
include/ggml-metal.h
include/ggml-rpc.h
include/ggml-sycl.h
include/ggml-vulkan.h
src/iqk/iqk_mul_mat.h)
include/ggml-vulkan.h)
set_target_properties(ggml PROPERTIES PUBLIC_HEADER "${GGML_PUBLIC_HEADERS}")
#if (GGML_METAL)

7
ggml/src/iqk/iqk_config.h
View File

@@ -14,6 +14,13 @@
#define IQK_IMPLEMENT
#endif
#if defined(_WIN32) && !defined(__MINGW32__)
# define IQK_API __declspec(dllexport)
#endif
#else
# define IQK_API __attribute__ ((visibility ("default")))
#endif
#ifdef _MSC_VER
#define IQK_NOINLINE __declspec(noinline)
#define IQK_ALWAYS_INLINE inline

4
ggml/src/iqk/iqk_flash_attn.cpp
View File

@@ -29,7 +29,7 @@ inline uint32_t simple_gcd(uint32_t a, uint32_t b) {
// TODO: get the ggml_type enum here without polution
//
extern "C" __attribute__ ((visibility ("default"))) bool iqk_flash_attn_noalibi(int type_q, int type_mask, float max_bias,
extern "C" IQK_API bool iqk_flash_attn_noalibi(int type_q, int type_mask, float max_bias,
int neq3, int neq2, long nbq3, long nbq2,
int nek3, int nek2, long nbk3, long nbk2,
int nev3, int nev2, long nbv3, long nbv2,
@@ -253,7 +253,7 @@ extern "C" __attribute__ ((visibility ("default"))) bool iqk_flash_attn_noalibi(
#else
extern "C" __attribute__ ((visibility ("default"))) bool iqk_flash_attn_noalibi([[maybe_unused]] int type_q, [[maybe_unused]] int type_mask, [[maybe_unused]] float max_bias,
extern "C" IQK_API bool iqk_flash_attn_noalibi([[maybe_unused]] int type_q, [[maybe_unused]] int type_mask, [[maybe_unused]] float max_bias,
[[maybe_unused]] int neq3, [[maybe_unused]] int neq2, [[maybe_unused]] long nbq3, [[maybe_unused]] long nbq2,
[[maybe_unused]] int nek3, [[maybe_unused]] int nek2, [[maybe_unused]] long nbk3, [[maybe_unused]] long nbk2,
[[maybe_unused]] int nev3, [[maybe_unused]] int nev2, [[maybe_unused]] long nbv3, [[maybe_unused]] long nbv2,

16
ggml/src/iqk/iqk_mul_mat.cpp
View File

@@ -403,7 +403,7 @@ private:
}
extern "C" __attribute__ ((visibility ("default"))) bool iqk_mul_mat(long Nx, long Ny, long ne00,
extern "C" IQK_API bool iqk_mul_mat(long Nx, long Ny, long ne00,
int typeA, const void * A, long strideA,
int typeB, const void * B, long strideB,
float * C, long stride_C, int ith, int nth) {
@@ -440,7 +440,7 @@ inline uint32_t simple_gcd(uint32_t a, uint32_t b) {
}
}
extern "C" __attribute__ ((visibility ("default"))) bool iqk_mul_mat_4d(long Nx, long Ny, long ne00,
extern "C" IQK_API bool iqk_mul_mat_4d(long Nx, long Ny, long ne00,
long ne02, long ne03, long ne12, long ne13,
long nb02, long nb03, long nb12, long nb13, long nb2, long nb3,
int typeA, const void * A, long strideA,
@@ -545,7 +545,7 @@ extern "C" __attribute__ ((visibility ("default"))) bool iqk_mul_mat_4d(long Nx,
return true;
}
extern "C" __attribute__ ((visibility ("default"))) bool iqk_mul_mat_moe(long Nx, long Ny, long ne00, int ne11,
extern "C" IQK_API bool iqk_mul_mat_moe(long Nx, long Ny, long ne00, int ne11,
int typeA, const void * A, long strideA,
int typeB, const void * B, long strideB,
float * C, long nb1, long nb2, const void * vrow_mapping, int ith, int nth) {
@@ -571,7 +571,7 @@ extern "C" __attribute__ ((visibility ("default"))) bool iqk_mul_mat_moe(long Nx
return true;
}
extern "C" __attribute__ ((visibility ("default"))) bool iqk_moe_fused_up_gate(long Nx, long Ny, long ne00, int ne11, int unary_op,
extern "C" IQK_API bool iqk_moe_fused_up_gate(long Nx, long Ny, long ne00, int ne11, int unary_op,
int typeA, const void * Aup, const void * Agate, long strideA,
int typeB, const void * B, long strideB,
float * C, long nb1, long nb2, const void * vrow_mapping, int ith, int nth) {
@@ -17550,11 +17550,11 @@ bool iqk_flash_attn_impl(int int_type_k, // type of k
#else // IQK_IMPLEMENT
extern "C" __attribute__ ((visibility ("default"))) bool iqk_mul_mat(int, long, long, long, int, const void *, long, int, const void *, long, float *, long, int, int) {
extern "C" IQK_API bool iqk_mul_mat(int, long, long, long, int, const void *, long, int, const void *, long, float *, long, int, int) {
return false;
}
extern "C" __attribute__ ((visibility ("default"))) bool iqk_mul_mat_4d(long /*Nx*/, long /*Ny*/, long /*ne00*/,
extern "C" IQK_API bool iqk_mul_mat_4d(long /*Nx*/, long /*Ny*/, long /*ne00*/,
long /*ne02*/, long /*ne03*/, long /*ne12*/, long /*ne13*/,
long /*nb02*/, long /*nb03*/, long /*nb12*/, long /*nb13*/, long /*nb2*/, long /*nb3*/,
int /*typeA*/, const void * /*A*/, long /*strideA*/,
@@ -17563,12 +17563,12 @@ extern "C" __attribute__ ((visibility ("default"))) bool iqk_mul_mat_4d(long /*N
return false;
}
extern "C" __attribute__ ((visibility ("default"))) bool iqk_mul_mat_moe(long, long, long, int, int, const void *, long, int, const void *, long, float *, long, long,
extern "C" IQK_API bool iqk_mul_mat_moe(long, long, long, int, int, const void *, long, int, const void *, long, float *, long, long,
const void *, int, int) {
return false;
}
extern "C" __attribute__ ((visibility ("default"))) bool iqk_moe_fused_up_gate(long /*Nx*/, long /*Ny*/, long /*ne00*/, int /*ne11*/, int /*unary_op*/,
extern "C" IQK_API bool iqk_moe_fused_up_gate(long /*Nx*/, long /*Ny*/, long /*ne00*/, int /*ne11*/, int /*unary_op*/,
int /*typeA*/, const void * /*Aup*/, const void * /*Agate*/, long /*strideA*/,
int /*typeB*/, const void * /*B*/, long /*strideB*/,
float * /*C*/, long /*nb1*/, long /*nb2*/, const void * /*vrow_mapping*/, int /*ith*/, int /*nth*/) {