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

chore : Fix the error when compiling rocm build on windows using cmake

Browse source 
(#9666)

Fix the compilation error "call to undeclared function '_mm256_dpbusd_epi32'". The function _mm256_dpbusd_epi32 is defined in avxintrin.h, while _mm256_dpbusd_epi32 is defined in avx512vlvnniintrin.h. Therefore, __AVX__, __AVX512VNNI__, and __AVX512VL__ need to be defined.

According to (#7743), DGGML_OPENMP=OFF is needed to add, so adding it in doc.
This commit is contained in:
cocochick 2024-11-15 20:37:38 +08:00
parent 18429220bd
commit 68ca77a148
4 changed files with 4 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

2
docs/build.md
View file

@ -259,7 +259,7 @@ You can download it from your Linux distro's package manager or from here: [ROCm
- Using `CMake` for Windows (using x64 Native Tools Command Prompt for VS, and assuming a gfx1100-compatible AMD GPU):
```bash
set PATH=%HIP_PATH%\bin;%PATH%
cmake -S . -B build -G Ninja -DAMDGPU_TARGETS=gfx1100 -DGGML_HIP=ON -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release
cmake -S . -B build -G Ninja -DGGML_OPENMP=OFF -DAMDGPU_TARGETS=gfx1100 -DGGML_HIP=ON -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release
cmake --build build
```
Make sure that `AMDGPU_TARGETS` is set to the GPU arch you want to compile for. The above example uses `gfx1100` that corresponds to Radeon RX 7900XTX/XT/GRE. You can find a list of targets [here](https://llvm.org/docs/AMDGPUUsage.html#processors)

2
ggml/src/ggml-cpu/ggml-cpu-aarch64.c
View file

@ -161,7 +161,7 @@ static inline __m256i sum_i16_pairs_int32x8(const __m256i x) {
}
static inline __m256i mul_sum_us8_pairs_int32x8(const __m256i ax, const __m256i sy) {
#if defined(__AVXVNNI__) || (defined(__AVX512VNNI__) && defined(__AVX512VL__))
#if defined(__AVX__) && defined(__AVX512VNNI__) && defined(__AVX512VL__)
const __m256i zero = _mm256_setzero_si256();
return _mm256_dpbusd_epi32(zero, ax, sy);
#else

2
ggml/src/ggml-cpu/ggml-cpu-quants.c
View file

@ -103,7 +103,7 @@ static inline __m256 sum_i16_pairs_float(const __m256i x) {
}
static inline __m256 mul_sum_us8_pairs_float(const __m256i ax, const __m256i sy) {
#if defined(__AVXVNNI__) || (defined(__AVX512VNNI__) && defined(__AVX512VL__))
#if defined(__AVX__) && defined(__AVX512VNNI__) && defined(__AVX512VL__)
const __m256i zero = _mm256_setzero_si256();
const __m256i summed_pairs = _mm256_dpbusd_epi32(zero, ax, sy);
return _mm256_cvtepi32_ps(summed_pairs);

2
ggml/src/ggml-cpu/llamafile/sgemm.cpp
View file

@ -992,7 +992,7 @@ class tinyBLAS_Q0_AVX {
inline __m256 updot(__m256i u, __m256i s) {
__m256i res;
#if defined(__AVXVNNI__) || (defined(__AVX512VNNI__) && defined(__AVX512VL__))
#if defined(__AVX__) && defined(__AVX512VNNI__) && defined(__AVX512VL__)
res = _mm256_dpbusd_epi32(_mm256_setzero_si256(), u, s);
#else
res = _mm256_madd_epi16(_mm256_set1_epi16(1), _mm256_maddubs_epi16(u, s));