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

ggml : fix build on Windows with Snapdragon X (#8531)

Browse source 
* Improvements for Windows with Snapdragon X

* Revert "Improvements for Windows with Snapdragon X"

This reverts commit bf21397ae5.

* Improvements for Windows with Snapdragon X

* WOA build clarifications

* WIndows on ARM build clarifications

* cmake build for Windows clarifications

* Update docs/build.md

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

---------

Co-authored-by: AndreasKunar <andreaskmsn.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Andreas (Andi) Kunar 2024-07-25 18:01:00 +02:00 • committed by GitHub
parent 88954f7fbd
commit bf5a81df37
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 18 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

12
ggml/src/ggml-aarch64.c
View file

@ -392,7 +392,7 @@ void ggml_gemv_q4_0_4x4_q8_0(int n, float * restrict s, size_t bs, const void *
#if defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8)
GGML_ASSERT(!(ggml_cpu_has_neon() && ggml_cpu_has_matmul_int8()) &&
"__ARM_NEON and __ARM_FEATURE_MATMUL_INT8 defined, use the Q4_0_4_8 quantization format for optimal performance");
#elif defined(__ARM_NEON) && defined(__aarch64__)
#elif defined(__ARM_NEON) && defined(__aarch64__) && ! ((defined(_MSC_VER)) && ! defined(__clang__))
const void * b_ptr = vx;
const void * a_ptr = vy;
float * res_ptr = s;
@ -501,7 +501,7 @@ void ggml_gemv_q4_0_4x8_q8_0(int n, float * restrict s, size_t bs, const void *
"__ARM_FEATURE_SVE defined, use the Q4_0_8_8 quantization format for optimal performance");
}
#endif
#if defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8)
#if defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8) && ! ((defined(_MSC_VER)) && ! defined(__clang__))
const void * b_ptr = vx;
const void * a_ptr = vy;
float * res_ptr = s;
@ -613,7 +613,7 @@ void ggml_gemv_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
UNUSED(ncols_interleaved);
UNUSED(blocklen);
#if defined(__ARM_FEATURE_SVE)
#if defined(__ARM_FEATURE_SVE) && ! ((defined(_MSC_VER)) && ! defined(__clang__))
if (svcntw() == 8) {
const void * b_ptr = vx;
const void * a_ptr = vy;
@ -753,7 +753,7 @@ void ggml_gemm_q4_0_4x4_q8_0(int n, float * restrict s, size_t bs, const void *
#if defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8)
GGML_ASSERT(!(ggml_cpu_has_neon() && ggml_cpu_has_matmul_int8()) &&
"__ARM_NEON and __ARM_FEATURE_MATMUL_INT8 defined, use the Q4_0_4_8 quantization format for optimal performance");
#elif defined(__ARM_NEON) && defined(__aarch64__)
#elif defined(__ARM_NEON) && defined(__aarch64__) && ! ((defined(_MSC_VER)) && ! defined(__clang__))
const void * b_ptr = vx;
const void * a_ptr = vy;
float * res_ptr = s;
@ -1271,7 +1271,7 @@ void ggml_gemm_q4_0_4x8_q8_0(int n, float * restrict s, size_t bs, const void *
"__ARM_FEATURE_SVE defined, use the Q4_0_8_8 quantization format for optimal performance");
}
#endif
#if defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8)
#if defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8) && ! ((defined(_MSC_VER)) && ! defined(__clang__))
const void * b_ptr = vx;
const void * a_ptr = vy;
float * res_ptr = s;
@ -1727,7 +1727,7 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
UNUSED(ncols_interleaved);
UNUSED(blocklen);
#if defined(__ARM_FEATURE_SVE) && defined(__ARM_FEATURE_MATMUL_INT8)
#if defined(__ARM_FEATURE_SVE) && defined(__ARM_FEATURE_MATMUL_INT8) && ! ((defined(_MSC_VER)) && ! defined(__clang__))
if (svcntw() == 8) {
const void * b_ptr = vx;
const void * a_ptr = vy;