vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

opencl: rename kernel files ggml-opencl2 -> ggml-opencl

Browse source
This commit is contained in:
Li He 2024-12-10 23:24:34 -08:00 • committed by Max Krasnyansky
parent 34f2fc15ea
commit 97a12703dd
11 changed files with 90 additions and 83 deletions
Download patch file Download diff file Expand all files Collapse all files

137
ggml/src/ggml-opencl/CMakeLists.txt
View file

@ -27,16 +27,16 @@ if (OpenCL_FOUND)
if (GGML_OPENCL_EMBED_KERNELS) if (GGML_OPENCL_EMBED_KERNELS)
add_compile_definitions(GGML_OPENCL_EMBED_KERNELS) add_compile_definitions(GGML_OPENCL_EMBED_KERNELS)
set(OPENCL2_CL_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl2.cl.h") set(OPENCL_CL_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl.cl.h")
set(OPENCL2_MM_CL_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl2_mm.cl.h") set(OPENCL_MM_CL_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl_mm.cl.h")
set(OPENCL2_CVT_CL_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl2_cvt.cl.h") set(OPENCL_CVT_CL_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl_cvt.cl.h")
set(OPENCL2_GEMV_NOSHUFFLE_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl2_gemv_noshuffle.cl.h") set(OPENCL_GEMV_NOSHUFFLE_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl_gemv_noshuffle.cl.h")
set(OPENCL2_GEMV_NOSHUFFLE_GENERAL_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl2_gemv_noshuffle_general.cl.h") set(OPENCL_GEMV_NOSHUFFLE_GENERAL_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl_gemv_noshuffle_general.cl.h")
set(OPENCL2_MUL_MAT_Ab_Bi_8x4_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl2_mul_mat_Ab_Bi_8x4.cl.h") set(OPENCL_MUL_MAT_Ab_Bi_8x4_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl_mul_mat_Ab_Bi_8x4.cl.h")
set(OPENCL2_TRANSPOSE_16_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl2_transpose_16.cl.h") set(OPENCL_TRANSPOSE_16_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl_transpose_16.cl.h")
set(OPENCL2_TRANSPOSE_32_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl2_transpose_32.cl.h") set(OPENCL_TRANSPOSE_32_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl_transpose_32.cl.h")
set(OPENCL2_TRANSPOSE_32_16_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl2_transpose_32_16.cl.h") set(OPENCL_TRANSPOSE_32_16_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-opencl_transpose_32_16.cl.h")
set(EMBED_KERNEL_SCRIPT "${CMAKE_CURRENT_SOURCE_DIR}/kernels/embed_kernel.py") set(EMBED_KERNEL_SCRIPT "${CMAKE_CURRENT_SOURCE_DIR}/kernels/embed_kernel.py")
file(MAKE_DIRECTORY "${CMAKE_BINARY_DIR}/autogenerated") file(MAKE_DIRECTORY "${CMAKE_BINARY_DIR}/autogenerated")
@ -45,101 +45,108 @@ if (OpenCL_FOUND)
# Python must be accessible from command line # Python must be accessible from command line
add_custom_command( add_custom_command(
OUTPUT ${OPENCL2_CL_SOURCE_EMBED} OUTPUT ${OPENCL_CL_SOURCE_EMBED}
COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT} COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT}
${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl2.cl ${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl.cl
${OPENCL2_CL_SOURCE_EMBED} ${OPENCL_CL_SOURCE_EMBED}
DEPENDS kernels/ggml-opencl2.cl ${EMBED_KERNEL_SCRIPT} DEPENDS kernels/ggml-opencl.cl ${EMBED_KERNEL_SCRIPT}
COMMENT "Generate ggml-opencl2.cl.h" COMMENT "Generate ggml-opencl2.cl.h"
) )
add_custom_command( add_custom_command(
OUTPUT ${OPENCL2_MM_CL_SOURCE_EMBED} OUTPUT ${OPENCL_MM_CL_SOURCE_EMBED}
COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT} COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT}
${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl2_mm.cl ${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl_mm.cl
${OPENCL2_MM_CL_SOURCE_EMBED} ${OPENCL_MM_CL_SOURCE_EMBED}
DEPENDS kernels/ggml-opencl2_mm.cl ${EMBED_KERNEL_SCRIPT} DEPENDS kernels/ggml-opencl_mm.cl ${EMBED_KERNEL_SCRIPT}
COMMENT "Generate ggml-opencl2_mm.cl.h" COMMENT "Generate ggml-opencl_mm.cl.h"
) )
add_custom_command( add_custom_command(
OUTPUT ${OPENCL2_CVT_CL_SOURCE_EMBED} OUTPUT ${OPENCL_CVT_CL_SOURCE_EMBED}
COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT} COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT}
${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl2_cvt.cl ${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl_cvt.cl
${OPENCL2_CVT_CL_SOURCE_EMBED} ${OPENCL_CVT_CL_SOURCE_EMBED}
DEPENDS kernels/ggml-opencl2_cvt.cl ${EMBED_KERNEL_SCRIPT} DEPENDS kernels/ggml-opencl_cvt.cl ${EMBED_KERNEL_SCRIPT}
COMMENT "Generate ggml-opencl2_cvt.cl.h" COMMENT "Generate ggml-opencl_cvt.cl.h"
) )
add_custom_command( add_custom_command(
OUTPUT ${OPENCL2_GEMV_NOSHUFFLE_SOURCE_EMBED} OUTPUT ${OPENCL_GEMV_NOSHUFFLE_SOURCE_EMBED}
COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT} COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT}
${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl2_gemv_noshuffle.cl ${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl_gemv_noshuffle.cl
${OPENCL2_GEMV_NOSHUFFLE_SOURCE_EMBED} ${OPENCL_GEMV_NOSHUFFLE_SOURCE_EMBED}
DEPENDS kernels/ggml-opencl2_gemv_noshuffle.cl ${EMBED_KERNEL_SCRIPT} DEPENDS kernels/ggml-opencl_gemv_noshuffle.cl ${EMBED_KERNEL_SCRIPT}
COMMENT "Generate ggml-opencl2_gemv_noshuffle.cl.h" COMMENT "Generate ggml-opencl_gemv_noshuffle.cl.h"
) )
add_custom_command( add_custom_command(
OUTPUT ${OPENCL2_GEMV_NOSHUFFLE_GENERAL_SOURCE_EMBED} OUTPUT ${OPENCL_GEMV_NOSHUFFLE_GENERAL_SOURCE_EMBED}
COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT} COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT}
${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl2_gemv_noshuffle_general.cl ${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl_gemv_noshuffle_general.cl
${OPENCL2_GEMV_NOSHUFFLE_GENERAL_SOURCE_EMBED} ${OPENCL_GEMV_NOSHUFFLE_GENERAL_SOURCE_EMBED}
DEPENDS kernels/ggml-opencl2_gemv_noshuffle_general.cl ${EMBED_KERNEL_SCRIPT} DEPENDS kernels/ggml-opencl_gemv_noshuffle_general.cl ${EMBED_KERNEL_SCRIPT}
COMMENT "Generate ggml-opencl2_gemv_noshuffle_general.cl.h" COMMENT "Generate ggml-opencl_gemv_noshuffle_general.cl.h"
) )
add_custom_command( add_custom_command(
OUTPUT ${OPENCL2_MUL_MAT_Ab_Bi_8x4_SOURCE_EMBED} OUTPUT ${OPENCL_MUL_MAT_Ab_Bi_8x4_SOURCE_EMBED}
COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT} COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT}
${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl2_mul_mat_Ab_Bi_8x4.cl ${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl_mul_mat_Ab_Bi_8x4.cl
${OPENCL2_MUL_MAT_Ab_Bi_8x4_SOURCE_EMBED} ${OPENCL_MUL_MAT_Ab_Bi_8x4_SOURCE_EMBED}
DEPENDS kernels/ggml-opencl2_mul_mat_Ab_Bi_8x4.cl ${EMBED_KERNEL_SCRIPT} DEPENDS kernels/ggml-opencl_mul_mat_Ab_Bi_8x4.cl ${EMBED_KERNEL_SCRIPT}
COMMENT "Generate ggml-opencl2_mul_mat_Ab_Bi_8x4.cl.cl.h" COMMENT "Generate ggml-opencl_mul_mat_Ab_Bi_8x4.cl.cl.h"
) )
add_custom_command( add_custom_command(
OUTPUT ${OPENCL2_TRANSPOSE_16_SOURCE_EMBED} OUTPUT ${OPENCL_TRANSPOSE_16_SOURCE_EMBED}
COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT} COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT}
${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl2_transpose_16.cl ${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl_transpose_16.cl
${OPENCL2_TRANSPOSE_16_SOURCE_EMBED} ${OPENCL_TRANSPOSE_16_SOURCE_EMBED}
DEPENDS kernels/ggml-opencl2_transpose_16.cl ${EMBED_KERNEL_SCRIPT} DEPENDS kernels/ggml-opencl_transpose_16.cl ${EMBED_KERNEL_SCRIPT}
COMMENT "Generate ggml-opencl2_transpose_16.cl.h" COMMENT "Generate ggml-opencl_transpose_16.cl.h"
) )
add_custom_command( add_custom_command(
OUTPUT ${OPENCL2_TRANSPOSE_32_SOURCE_EMBED} OUTPUT ${OPENCL_TRANSPOSE_32_SOURCE_EMBED}
COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT} COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT}
${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl2_transpose_32.cl ${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl_transpose_32.cl
${OPENCL2_TRANSPOSE_32_SOURCE_EMBED} ${OPENCL_TRANSPOSE_32_SOURCE_EMBED}
DEPENDS kernels/ggml-opencl2_transpose_32.cl ${EMBED_KERNEL_SCRIPT} DEPENDS kernels/ggml-opencl_transpose_32.cl ${EMBED_KERNEL_SCRIPT}
COMMENT "Generate ggml-opencl2_transpose_32.cl.h" COMMENT "Generate ggml-opencl_transpose_32.cl.h"
) )
add_custom_command( add_custom_command(
OUTPUT ${OPENCL2_TRANSPOSE_32_16_SOURCE_EMBED} OUTPUT ${OPENCL_TRANSPOSE_32_16_SOURCE_EMBED}
COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT} COMMAND ${Python3_EXECUTABLE} ${EMBED_KERNEL_SCRIPT}
${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl2_transpose_32_16.cl ${CMAKE_CURRENT_SOURCE_DIR}/kernels/ggml-opencl_transpose_32_16.cl
${OPENCL2_TRANSPOSE_32_16_SOURCE_EMBED} ${OPENCL_TRANSPOSE_32_16_SOURCE_EMBED}
DEPENDS kernels/ggml-opencl2_transpose_32_16.cl ${EMBED_KERNEL_SCRIPT} DEPENDS kernels/ggml-opencl_transpose_32_16.cl ${EMBED_KERNEL_SCRIPT}
COMMENT "Generate ggml-opencl2_transpose_32_16.cl.h" COMMENT "Generate ggml-opencl_transpose_32_16.cl.h"
) )
target_sources(${TARGET_NAME} PRIVATE target_sources(${TARGET_NAME} PRIVATE
${OPENCL2_CL_SOURCE_EMBED} ${OPENCL_CL_SOURCE_EMBED}
${OPENCL2_MM_CL_SOURCE_EMBED} ${OPENCL_MM_CL_SOURCE_EMBED}
${OPENCL2_CVT_CL_SOURCE_EMBED} ${OPENCL_CVT_CL_SOURCE_EMBED}
${OPENCL2_GEMV_NOSHUFFLE_SOURCE_EMBED} ${OPENCL_GEMV_NOSHUFFLE_SOURCE_EMBED}
${OPENCL2_GEMV_NOSHUFFLE_GENERAL_SOURCE_EMBED} ${OPENCL_GEMV_NOSHUFFLE_GENERAL_SOURCE_EMBED}
${OPENCL2_MUL_MAT_Ab_Bi_8x4_SOURCE_EMBED} ${OPENCL_MUL_MAT_Ab_Bi_8x4_SOURCE_EMBED}
${OPENCL2_TRANSPOSE_16_SOURCE_EMBED} ${OPENCL_TRANSPOSE_16_SOURCE_EMBED}
${OPENCL2_TRANSPOSE_32_SOURCE_EMBED} ${OPENCL_TRANSPOSE_32_SOURCE_EMBED}
${OPENCL2_TRANSPOSE_32_16_SOURCE_EMBED}) ${OPENCL_TRANSPOSE_32_16_SOURCE_EMBED})
else () else ()
# copy ggml-opencl.cl to bin directory # copy ggml-opencl.cl to bin directory
configure_file(kernels/ggml-opencl2.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl2.cl COPYONLY) configure_file(kernels/ggml-opencl.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl.cl COPYONLY)
configure_file(kernels/ggml-opencl2_mm.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl2_mm.cl COPYONLY) configure_file(kernels/ggml-opencl_mm.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl_mm.cl COPYONLY)
configure_file(kernels/ggml-opencl2_cvt.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl2_cvt.cl COPYONLY) configure_file(kernels/ggml-opencl_cvt.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl_cvt.cl COPYONLY)
configure_file(kernels/ggml-opencl_gemv_noshuffle.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl_gemv_noshuffle.cl COPYONLY)
configure_file(kernels/ggml-opencl_gemv_noshuffle_general.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl_gemv_noshuffle_general.cl COPYONLY)
configure_file(kernels/ggml-opencl_mul_mat_Ab_Bi_8x4.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl_mul_mat_Ab_Bi_8x4.cl COPYONLY)
configure_file(kernels/ggml-opencl_transpose_16.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl_transpose_16.cl COPYONLY)
configure_file(kernels/ggml-opencl_transpose_32.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl_transpose_32.cl COPYONLY)
configure_file(kernels/ggml-opencl_transpose_32_16.cl ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-opencl_transpose_32_16.cl COPYONLY)
endif () endif ()
else () else ()
message(WARNING "OpenCL not found") message(WARNING "OpenCL not found")

36
ggml/src/ggml-opencl/ggml-opencl.cpp
View file

@ -539,10 +539,10 @@ static ggml_backend_opencl2_context * ggml_cl2_init(ggml_backend_dev_t dev) {
#ifdef GGML_OPENCL_EMBED_KERNELS #ifdef GGML_OPENCL_EMBED_KERNELS
const std::string kernel_src { const std::string kernel_src {
#include "ggml-opencl2.cl.h" #include "ggml-opencl.cl.h"
}; };
#else #else
const std::string kernel_src = read_file("ggml-opencl2.cl"); const std::string kernel_src = read_file("ggml-opencl.cl");
#endif #endif
std::string compile_opts = std::string compile_opts =
@ -597,10 +597,10 @@ static ggml_backend_opencl2_context * ggml_cl2_init(ggml_backend_dev_t dev) {
// Load additional mulmat kernels. // Load additional mulmat kernels.
#ifdef GGML_OPENCL_EMBED_KERNELS #ifdef GGML_OPENCL_EMBED_KERNELS
const std::string kernel_src_1 { const std::string kernel_src_1 {
#include "ggml-opencl2_mm.cl.h" #include "ggml-opencl_mm.cl.h"
}; };
#else #else
const std::string kernel_src_1 = read_file("ggml-opencl2_mm.cl"); const std::string kernel_src_1 = read_file("ggml-opencl_mm.cl");
#endif #endif
backend_ctx->program_1 = build_program_from_source(context, device, kernel_src_1.c_str(), compile_opts); backend_ctx->program_1 = build_program_from_source(context, device, kernel_src_1.c_str(), compile_opts);
@ -613,10 +613,10 @@ static ggml_backend_opencl2_context * ggml_cl2_init(ggml_backend_dev_t dev) {
// Load additional data conversion kernels. // Load additional data conversion kernels.
#ifdef GGML_OPENCL_EMBED_KERNELS #ifdef GGML_OPENCL_EMBED_KERNELS
const std::string kernel_src_2 { const std::string kernel_src_2 {
#include "ggml-opencl2_cvt.cl.h" #include "ggml-opencl_cvt.cl.h"
}; };
#else #else
const std::string kernel_src_2 = read_file("ggml-opencl2_cvt.cl"); const std::string kernel_src_2 = read_file("ggml-opencl_cvt.cl");
#endif #endif
backend_ctx->program_2 = build_program_from_source(context, device, kernel_src_2.c_str(), compile_opts); backend_ctx->program_2 = build_program_from_source(context, device, kernel_src_2.c_str(), compile_opts);
@ -626,30 +626,30 @@ static ggml_backend_opencl2_context * ggml_cl2_init(ggml_backend_dev_t dev) {
#ifdef GGML_OPENCL_USE_ADRENO_KERNELS #ifdef GGML_OPENCL_USE_ADRENO_KERNELS
#ifdef GGML_OPENCL_EMBED_KERNELS #ifdef GGML_OPENCL_EMBED_KERNELS
const std::string transpose_32_src { const std::string transpose_32_src {
#include "ggml-opencl2_transpose_32.cl.h" #include "ggml-opencl_transpose_32.cl.h"
}; };
#else #else
const std::string transpose_32_src = read_file("ggml-opencl2_transpose_32.cl"); const std::string transpose_32_src = read_file("ggml-opencl_transpose_32.cl");
#endif #endif
backend_ctx->program_transpose_32 = build_program_from_source(context, device, transpose_32_src.c_str(), compile_opts); backend_ctx->program_transpose_32 = build_program_from_source(context, device, transpose_32_src.c_str(), compile_opts);
CL_CHECK((backend_ctx->kernel_transpose_32 = clCreateKernel(backend_ctx->program_transpose_32, "kernel_transpose_32", &err), err)); CL_CHECK((backend_ctx->kernel_transpose_32 = clCreateKernel(backend_ctx->program_transpose_32, "kernel_transpose_32", &err), err));
#ifdef GGML_OPENCL_EMBED_KERNELS #ifdef GGML_OPENCL_EMBED_KERNELS
const std::string transpose_32_16_src { const std::string transpose_32_16_src {
#include "ggml-opencl2_transpose_32_16.cl.h" #include "ggml-opencl_transpose_32_16.cl.h"
}; };
#else #else
const std::string transpose_32_16_src = read_file("ggml-opencl2_transpose_32_16.cl"); const std::string transpose_32_16_src = read_file("ggml-opencl_transpose_32_16.cl");
#endif #endif
backend_ctx->program_transpose_32_16 = build_program_from_source(context, device, transpose_32_16_src.c_str(), compile_opts); backend_ctx->program_transpose_32_16 = build_program_from_source(context, device, transpose_32_16_src.c_str(), compile_opts);
CL_CHECK((backend_ctx->kernel_transpose_32_16 = clCreateKernel(backend_ctx->program_transpose_32_16, "kernel_transpose_32_16", &err), err)); CL_CHECK((backend_ctx->kernel_transpose_32_16 = clCreateKernel(backend_ctx->program_transpose_32_16, "kernel_transpose_32_16", &err), err));
#ifdef GGML_OPENCL_EMBED_KERNELS #ifdef GGML_OPENCL_EMBED_KERNELS
const std::string transpose_16_src { const std::string transpose_16_src {
#include "ggml-opencl2_transpose_16.cl.h" #include "ggml-opencl_transpose_16.cl.h"
}; };
#else #else
const std::string transpose_16_src = read_file("ggml-opencl2_transpose_16.cl"); const std::string transpose_16_src = read_file("ggml-opencl_transpose_16.cl");
#endif #endif
backend_ctx->program_transpose_16 = build_program_from_source(context, device, transpose_16_src.c_str(), compile_opts); backend_ctx->program_transpose_16 = build_program_from_source(context, device, transpose_16_src.c_str(), compile_opts);
CL_CHECK((backend_ctx->kernel_transpose_16 = clCreateKernel(backend_ctx->program_transpose_16, "kernel_transpose_16", &err), err)); CL_CHECK((backend_ctx->kernel_transpose_16 = clCreateKernel(backend_ctx->program_transpose_16, "kernel_transpose_16", &err), err));
@ -664,10 +664,10 @@ static ggml_backend_opencl2_context * ggml_cl2_init(ggml_backend_dev_t dev) {
} }
#ifdef GGML_OPENCL_EMBED_KERNELS #ifdef GGML_OPENCL_EMBED_KERNELS
const std::string kernel_src_CL_gemv_general { const std::string kernel_src_CL_gemv_general {
#include "ggml-opencl2_gemv_noshuffle_general.cl.h" #include "ggml-opencl_gemv_noshuffle_general.cl.h"
}; };
#else #else
const std::string kernel_src_CL_gemv_general = read_file("ggml-opencl2_gemv_noshuffle_general.cl"); const std::string kernel_src_CL_gemv_general = read_file("ggml-opencl_gemv_noshuffle_general.cl");
#endif #endif
backend_ctx->program_CL_gemv_general = build_program_from_source( backend_ctx->program_CL_gemv_general = build_program_from_source(
@ -686,10 +686,10 @@ static ggml_backend_opencl2_context * ggml_cl2_init(ggml_backend_dev_t dev) {
} }
#ifdef GGML_OPENCL_EMBED_KERNELS #ifdef GGML_OPENCL_EMBED_KERNELS
const std::string kernel_src_CL_gemv { const std::string kernel_src_CL_gemv {
#include "ggml-opencl2_gemv_noshuffle.cl.h" #include "ggml-opencl_gemv_noshuffle.cl.h"
}; };
#else #else
const std::string kernel_src_CL_gemv = read_file("ggml-opencl2_gemv_noshuffle.cl"); const std::string kernel_src_CL_gemv = read_file("ggml-opencl_gemv_noshuffle.cl");
#endif #endif
backend_ctx->program_CL_gemv_4096_1_4096 = build_program_from_source( backend_ctx->program_CL_gemv_4096_1_4096 = build_program_from_source(
@ -743,10 +743,10 @@ static ggml_backend_opencl2_context * ggml_cl2_init(ggml_backend_dev_t dev) {
// Gemm // Gemm
#ifdef GGML_OPENCL_EMBED_KERNELS #ifdef GGML_OPENCL_EMBED_KERNELS
const std::string kernel_src_CL_gemm { const std::string kernel_src_CL_gemm {
#include "ggml-opencl2_mul_mat_Ab_Bi_8x4.cl.h" #include "ggml-opencl_mul_mat_Ab_Bi_8x4.cl.h"
}; };
#else #else
const std::string kernel_src_CL_gemm = read_file("ggml-opencl2_mul_mat_Ab_Bi_8x4.cl"); const std::string kernel_src_CL_gemm = read_file("ggml-opencl_mul_mat_Ab_Bi_8x4.cl");
#endif #endif
backend_ctx->program_CL_gemm = build_program_from_source(context, device, kernel_src_CL_gemm.c_str(), compile_opts); backend_ctx->program_CL_gemm = build_program_from_source(context, device, kernel_src_CL_gemm.c_str(), compile_opts);
CL_CHECK((backend_ctx->CL_mul_mat_Ab_Bi_8x4 = clCreateKernel(backend_ctx->program_CL_gemm, "kernel_mul_mat_Ab_Bi_8x4", &err), err)); CL_CHECK((backend_ctx->CL_mul_mat_Ab_Bi_8x4 = clCreateKernel(backend_ctx->program_CL_gemm, "kernel_mul_mat_Ab_Bi_8x4", &err), err));

0
ggml/src/ggml-opencl/kernels/ggml-opencl2.cl → ggml/src/ggml-opencl/kernels/ggml-opencl.cl
View file

0
ggml/src/ggml-opencl/kernels/ggml-opencl2_cvt.cl → ggml/src/ggml-opencl/kernels/ggml-opencl_cvt.cl
View file

0
ggml/src/ggml-opencl/kernels/ggml-opencl2_gemv_noshuffle.cl → ggml/src/ggml-opencl/kernels/ggml-opencl_gemv_noshuffle.cl
View file

0
ggml/src/ggml-opencl/kernels/ggml-opencl2_gemv_noshuffle_general.cl → ggml/src/ggml-opencl/kernels/ggml-opencl_gemv_noshuffle_general.cl
View file

0
ggml/src/ggml-opencl/kernels/ggml-opencl2_mm.cl → ggml/src/ggml-opencl/kernels/ggml-opencl_mm.cl
View file

0
ggml/src/ggml-opencl/kernels/ggml-opencl2_mul_mat_Ab_Bi_8x4.cl → ggml/src/ggml-opencl/kernels/ggml-opencl_mul_mat_Ab_Bi_8x4.cl
View file

0
ggml/src/ggml-opencl/kernels/ggml-opencl2_transpose_16.cl → ggml/src/ggml-opencl/kernels/ggml-opencl_transpose_16.cl
View file

0
ggml/src/ggml-opencl/kernels/ggml-opencl2_transpose_32.cl → ggml/src/ggml-opencl/kernels/ggml-opencl_transpose_32.cl
View file

0
ggml/src/ggml-opencl/kernels/ggml-opencl2_transpose_32_16.cl → ggml/src/ggml-opencl/kernels/ggml-opencl_transpose_32_16.cl
View file