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

arm64 support for windows (#3007)

Browse source 
Co-authored-by: Cebtenzzre <cebtenzzre@gmail.com>
This commit is contained in:
Eric Sommerlade 2023-09-13 02:54:20 +01:00 committed by GitHub
parent 4f7cd6ba9c
commit b52b29ab9d
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
4 changed files with 20 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

17
CMakeLists.txt
View file

@ -461,6 +461,13 @@ endif()
# TODO: probably these flags need to be tweaked on some architectures # TODO: probably these flags need to be tweaked on some architectures
# feel free to update the Makefile for your architecture and send a pull request or issue # feel free to update the Makefile for your architecture and send a pull request or issue
message(STATUS "CMAKE_SYSTEM_PROCESSOR: ${CMAKE_SYSTEM_PROCESSOR}") message(STATUS "CMAKE_SYSTEM_PROCESSOR: ${CMAKE_SYSTEM_PROCESSOR}")
if (MSVC)
string(TOLOWER "${CMAKE_GENERATOR_PLATFORM}" CMAKE_GENERATOR_PLATFORM_LWR)
message(STATUS "CMAKE_GENERATOR_PLATFORM: ${CMAKE_GENERATOR_PLATFORM}")
else ()
set(CMAKE_GENERATOR_PLATFORM_LWR "")
endif ()
if (NOT MSVC) if (NOT MSVC)
if (LLAMA_STATIC) if (LLAMA_STATIC)
add_link_options(-static) add_link_options(-static)
@ -476,10 +483,14 @@ if (NOT MSVC)
endif() endif()
endif() endif()
if ((${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm") OR (${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64")) if ((${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm") OR (${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64") OR ("${CMAKE_GENERATOR_PLATFORM_LWR}" MATCHES "arm64"))
message(STATUS "ARM detected") message(STATUS "ARM detected")
if (MSVC) if (MSVC)
# TODO: arm msvc? add_compile_definitions(__ARM_NEON)
add_compile_definitions(__ARM_FEATURE_FMA)
add_compile_definitions(__ARM_FEATURE_DOTPROD)
# add_compile_definitions(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) # MSVC doesn't support vdupq_n_f16, vld1q_f16, vst1q_f16
add_compile_definitions(__aarch64__) # MSVC defines _M_ARM64 instead
else() else()
if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "armv6") if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "armv6")
# Raspberry Pi 1, Zero # Raspberry Pi 1, Zero
@ -494,7 +505,7 @@ if ((${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm") OR (${CMAKE_SYSTEM_PROCESSOR} MATC
add_compile_options(-mfp16-format=ieee -mno-unaligned-access) add_compile_options(-mfp16-format=ieee -mno-unaligned-access)
endif() endif()
endif() endif()
elseif (${CMAKE_SYSTEM_PROCESSOR} MATCHES "^(x86_64|i686|AMD64)$") elseif (${CMAKE_SYSTEM_PROCESSOR} MATCHES "^(x86_64|i686|AMD64)$" OR "${CMAKE_GENERATOR_PLATFORM_LWR}" MATCHES "^(x86_64|i686|amd64|x64)$" )
message(STATUS "x86 detected") message(STATUS "x86 detected")
if (MSVC) if (MSVC)
if (LLAMA_AVX512) if (LLAMA_AVX512)

2
ggml.c
View file

@ -283,7 +283,7 @@ typedef double ggml_float;
// 16-bit float // 16-bit float
// on Arm, we use __fp16 // on Arm, we use __fp16
// on x86, we use uint16_t // on x86, we use uint16_t
#ifdef __ARM_NEON #if defined(__ARM_NEON) && !defined(_MSC_VER)
// if YCM cannot find <arm_neon.h>, make a symbolic link to it, for example: // if YCM cannot find <arm_neon.h>, make a symbolic link to it, for example:
// //

2
ggml.h
View file

@ -270,7 +270,7 @@ extern "C" {
#if defined(__ARM_NEON) && defined(__CUDACC__) #if defined(__ARM_NEON) && defined(__CUDACC__)
typedef half ggml_fp16_t; typedef half ggml_fp16_t;
#elif defined(__ARM_NEON) #elif defined(__ARM_NEON) && !defined(_MSC_VER)
typedef __fp16 ggml_fp16_t; typedef __fp16 ggml_fp16_t;
#else #else
typedef uint16_t ggml_fp16_t; typedef uint16_t ggml_fp16_t;

5
k_quants.c
View file

@ -2609,7 +2609,10 @@ void ggml_vec_dot_q4_K_q8_K(const int n, float * restrict s, const void * restri
memcpy(utmp, x[i].scales, 12); memcpy(utmp, x[i].scales, 12);
const uint32x2_t mins8 = {utmp[1] & kmask1, ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4)}; uint32x2_t mins8 = { 0 };
mins8 = vset_lane_u32(utmp[1] & kmask1, mins8, 0);
mins8 = vset_lane_u32(((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4), mins8, 1);
utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4); utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
utmp[0] &= kmask1; utmp[0] &= kmask1;