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

make, cmake : enable CUDA errors on warnings

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-02-19 14:10:45 +02:00
parent 46fd273724
commit c3520e9327
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
3 changed files with 34 additions and 25 deletions
Download patch file Download diff file Expand all files Collapse all files

4
CMakeLists.txt
View file

@ -774,6 +774,10 @@ set(CUDA_CXX_FLAGS "")
if (LLAMA_CUBLAS)
set(CUDA_FLAGS -use_fast_math)
if (LLAMA_FATAL_WARNINGS)
list(APPEND CUDA_FLAGS -Werror all-warnings)
endif()
if (LLAMA_ALL_WARNINGS AND NOT MSVC)
set(NVCC_CMD ${CMAKE_CUDA_COMPILER} .c)
if (NOT CMAKE_CUDA_HOST_COMPILER STREQUAL "")

5
Makefile
View file

@ -217,7 +217,7 @@ MK_CFLAGS += $(WARN_FLAGS) -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmis
MK_CXXFLAGS += $(WARN_FLAGS) -Wmissing-declarations -Wmissing-noreturn
ifeq ($(LLAMA_FATAL_WARNINGS),1)
MK_CFLAGS += -Werror
MK_CFLAGS += -Werror
MK_CXXFLAGS += -Werror
endif
@ -385,6 +385,9 @@ ifdef LLAMA_CUBLAS
MK_LDFLAGS += -lcuda -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64 -L/opt/cuda/lib64 -L$(CUDA_PATH)/targets/x86_64-linux/lib -L/usr/local/cuda/targets/aarch64-linux/lib -L/usr/lib/wsl/lib
OBJS += ggml-cuda.o
MK_NVCCFLAGS += -use_fast_math
ifdef LLAMA_FATAL_WARNINGS
MK_NVCCFLAGS += -Werror all-warnings
endif # LLAMA_FATAL_WARNINGS
ifndef JETSON_EOL_MODULE_DETECT
MK_NVCCFLAGS += --forward-unknown-to-host-compiler
endif # JETSON_EOL_MODULE_DETECT

50
ggml-cuda.cu
View file

@ -651,18 +651,18 @@ static __device__ __forceinline__ float2 warp_reduce_sum(float2 a) {
return a;
}
static __device__ __forceinline__ half2 warp_reduce_sum(half2 a) {
#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL
#pragma unroll
for (int mask = 16; mask > 0; mask >>= 1) {
a = __hadd2(a, __shfl_xor_sync(0xffffffff, a, mask, 32));
}
return a;
#else
(void) a;
NO_DEVICE_CODE;
#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL
}
//static __device__ __forceinline__ half2 warp_reduce_sum(half2 a) {
//#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL
//#pragma unroll
// for (int mask = 16; mask > 0; mask >>= 1) {
// a = __hadd2(a, __shfl_xor_sync(0xffffffff, a, mask, 32));
// }
// return a;
//#else
// (void) a;
// NO_DEVICE_CODE;
//#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL
//}
static __device__ __forceinline__ float warp_reduce_max(float x) {
#pragma unroll
@ -672,18 +672,18 @@ static __device__ __forceinline__ float warp_reduce_max(float x) {
return x;
}
static __device__ __forceinline__ half2 warp_reduce_max(half2 x) {
#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL && CUDART_VERSION >= CUDART_HMAX
#pragma unroll
for (int mask = 16; mask > 0; mask >>= 1) {
x = __hmax2(x, __shfl_xor_sync(0xffffffff, x, mask, 32));
}
return x;
#else
(void) x;
NO_DEVICE_CODE;
#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL && CUDART_VERSION >= CUDART_HMAX
}
//static __device__ __forceinline__ half2 warp_reduce_max(half2 x) {
//#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL && CUDART_VERSION >= CUDART_HMAX
//#pragma unroll
// for (int mask = 16; mask > 0; mask >>= 1) {
// x = __hmax2(x, __shfl_xor_sync(0xffffffff, x, mask, 32));
// }
// return x;
//#else
// (void) x;
// NO_DEVICE_CODE;
//#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL && CUDART_VERSION >= CUDART_HMAX
//}
static __device__ __forceinline__ float op_repeat(const float a, const float b) {
return b;
@ -4641,10 +4641,12 @@ static __device__ __forceinline__ float vec_dot_iq2_xs_q8_1(
const float d = (float)bq2->d * __low2float(bq8_1[ib32].ds) * 0.25f;
return d * ((0.5f + ls1) * sumi1 + (0.5f + ls2) * sumi2);
#else
(void) ksigns64;
assert(false);
return 0.f;
#endif
#else
(void) ksigns64;
assert(false);
return 0.f;
#endif