tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-06 01:05:03 -04:00
Code Issues Packages Projects Releases Wiki Activity

HIP: Ignore unsupported unroll transformation in fattn-vec (#14931)

Browse Source 
llvm with the amdgcn target dose not support unrolling loops with conditional break statements, when those statements can not be resolved at compile time. Similar to other places in GGML lets simply ignore this warning.
This commit is contained in:
uvos
2025-07-29 17:43:43 +02:00
committed by GitHub
parent 1a67fcc306
commit c7aa1364fd
2 changed files with 18 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
ggml/src/ggml-cuda/fattn-vec-f16.cuh
View File

@@ -1,6 +1,12 @@
#include "common.cuh"
#include "fattn-common.cuh"
// Currenlty llvm with the amdgcn target dose not support unrolling loops
// that contain a break that can not be resolved at compile time.
#ifdef __clang__
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wpass-failed"
#endif // __clang__
template<int D, int ncols, ggml_type type_K, ggml_type type_V, bool use_logit_softcap> // D == head size
#ifndef GGML_USE_HIP
__launch_bounds__(D, 1)
@@ -341,6 +347,9 @@ static __global__ void flash_attn_vec_ext_f16(
NO_DEVICE_CODE;
#endif // defined(FLASH_ATTN_AVAILABLE) && defined(FP16_AVAILABLE)
}
#ifdef __clang__
#pragma clang diagnostic pop
#endif // __clang__
template <int D, int cols_per_block, ggml_type type_K, ggml_type type_V, bool use_logit_softcap>
void ggml_cuda_flash_attn_ext_vec_f16_case_impl(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {

9
ggml/src/ggml-cuda/fattn-vec-f32.cuh
View File

@@ -1,6 +1,12 @@
#include "common.cuh"
#include "fattn-common.cuh"
// Currenlty llvm with the amdgcn target dose not support unrolling loops
// that contain a break that can not be resolved at compile time.
#ifdef __clang__
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wpass-failed"
#endif // __clang__
template<int D, int ncols, ggml_type type_K, ggml_type type_V, bool use_logit_softcap> // D == head size
#ifndef GGML_USE_HIP
__launch_bounds__(D, 1)
@@ -336,6 +342,9 @@ static __global__ void flash_attn_vec_ext_f32(
NO_DEVICE_CODE;
#endif // FLASH_ATTN_AVAILABLE
}
#ifdef __clang__
#pragma clang diagnostic pop
#endif // __clang__
template <int D, int cols_per_block, ggml_type type_K, ggml_type type_V, bool use_logit_softcap>
void ggml_cuda_flash_attn_ext_vec_f32_case_impl(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {