CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-28 08:31:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml : fix loongarch lsx compilation error (#15864)

Browse Source
This commit is contained in:
junchao-zhao
2025-09-25 17:22:55 +08:00
committed by GitHub
parent 4cdd0bb453
commit aa719c2f88
2 changed files with 20 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
ggml/src/ggml-cpu/arch/loongarch/quants.c
View File

@@ -105,6 +105,18 @@ static inline float hsum_float_4x4(const __m128 a, const __m128 b, const __m128
return ((v4f32)res)[0]; return ((v4f32)res)[0];
} }
// multiply int8_t, add results pairwise twice
static inline __m128i mul_sum_i8_pairs(const __m128i x, const __m128i y) {
// Get absolute values of x vectors
const __m128i ax = __lsx_vsigncov_b(x, x);
// Sign the values of the y vectors
const __m128i sy = __lsx_vsigncov_b(x, y);
// Perform multiplication and create 16-bit values
const __m128i dot = lsx_maddubs_h(ax, sy);
const __m128i ones = __lsx_vreplgr2vr_h(1);
return lsx_madd_h(ones, dot);
}
#endif #endif
#if defined(__loongarch_asx) #if defined(__loongarch_asx)
@@ -323,18 +335,6 @@ static inline __m256i lasx_xvandi_b_bit(__m256i a, const unsigned int b) {
} }
} }
// multiply int8_t, add results pairwise twice
static inline __m128i mul_sum_i8_pairs(const __m128i x, const __m128i y) {
// Get absolute values of x vectors
const __m128i ax = __lsx_vsigncov_b(x, x);
// Sign the values of the y vectors
const __m128i sy = __lsx_vsigncov_b(x, y);
// Perform multiplication and create 16-bit values
const __m128i dot = lsx_maddubs_h(ax, sy);
const __m128i ones = __lsx_vreplgr2vr_h(1);
return lsx_madd_h(ones, dot);
}
// horizontally add 8 floats // horizontally add 8 floats
static inline float hsum_float_8(const __m256 x) { static inline float hsum_float_8(const __m256 x) {
__m128 res = lasx_extractf128(x, 1); __m128 res = lasx_extractf128(x, 1);

16
ggml/src/ggml-cpu/simd-mappings.h
View File

@@ -998,9 +998,9 @@ static inline void __lasx_f32cx8_store(ggml_fp16_t * x, __m256 y) {
#define GGML_F32_EPR 4 #define GGML_F32_EPR 4
#define GGML_F32x4 __m128 #define GGML_F32x4 __m128
#define GGML_F32x4_ZERO __lsx_vldi(0) #define GGML_F32x4_ZERO (__m128)__lsx_vldi(0)
#define GGML_F32x4_SET1(x) __lsx_vinsgr2vr_w(__lsx_vldi(0),(x), 0) #define GGML_F32x4_SET1(x) (__m128)__lsx_vinsgr2vr_w(__lsx_vldi(0),(x), 0)
#define GGML_F32x4_LOAD(x) __lsx_vld((x), 0) #define GGML_F32x4_LOAD(x) (__m128)__lsx_vld((x), 0)
#define GGML_F32x4_STORE(x, y) __lsx_vst(y, x, 0) #define GGML_F32x4_STORE(x, y) __lsx_vst(y, x, 0)
#define GGML_F32x4_FMA(a, b, c) __lsx_vfmadd_s(b, c, a) #define GGML_F32x4_FMA(a, b, c) __lsx_vfmadd_s(b, c, a)
#define GGML_F32x4_ADD __lsx_vfadd_s #define GGML_F32x4_ADD __lsx_vfadd_s
@@ -1022,7 +1022,7 @@ static inline void __lasx_f32cx8_store(ggml_fp16_t * x, __m256 y) {
__m128i tmp = __lsx_vsrli_d((__m128i) x[0], 32); \ __m128i tmp = __lsx_vsrli_d((__m128i) x[0], 32); \
tmp = (__m128i) __lsx_vfadd_s((__m128) tmp, x[0]); \ tmp = (__m128i) __lsx_vfadd_s((__m128) tmp, x[0]); \
tmp = __lsx_vpickev_w(__lsx_vldi(0), tmp); \ tmp = __lsx_vpickev_w(__lsx_vldi(0), tmp); \
const __m128 t0 = __lsx_vshuf4i_w(tmp, 0x88); \ const __m128 t0 = (__m128)__lsx_vshuf4i_w(tmp, 0x88); \
tmp = __lsx_vsrli_d((__m128i) t0, 32); \ tmp = __lsx_vsrli_d((__m128i) t0, 32); \
tmp = (__m128i) __lsx_vfadd_s((__m128) tmp, t0); \ tmp = (__m128i) __lsx_vfadd_s((__m128) tmp, t0); \
tmp = __lsx_vpickev_w(__lsx_vldi(0), tmp); \ tmp = __lsx_vpickev_w(__lsx_vldi(0), tmp); \
@@ -1052,7 +1052,7 @@ static inline __m128 __lsx_f16x4_load(const ggml_fp16_t * x) {
tmp[2] = GGML_CPU_FP16_TO_FP32(x[2]); tmp[2] = GGML_CPU_FP16_TO_FP32(x[2]);
tmp[3] = GGML_CPU_FP16_TO_FP32(x[3]); tmp[3] = GGML_CPU_FP16_TO_FP32(x[3]);
return __lsx_vld(tmp, 0); return (__m128)__lsx_vld(tmp, 0);
} }
static inline void __lsx_f16x4_store(ggml_fp16_t * x, __m128 y) { static inline void __lsx_f16x4_store(ggml_fp16_t * x, __m128 y) {
@@ -1067,9 +1067,9 @@ static inline void __lsx_f16x4_store(ggml_fp16_t * x, __m128 y) {
} }
#define GGML_F32Cx4 __m128 #define GGML_F32Cx4 __m128
#define GGML_F32Cx4_ZERO __lsx_vldi(0) #define GGML_F32Cx4_ZERO (__m128)__lsx_vldi(0)
#define GGML_F32Cx4_SET1(x) __lsx_vinsgr2vr_w(__lsx_vldi(0),(x), 0) #define GGML_F32Cx4_SET1(x) (__m128)__lsx_vinsgr2vr_w(__lsx_vldi(0),(x), 0)
#define GGML_F32Cx4_LOAD(x) __lsx_f16x4_load(x) #define GGML_F32Cx4_LOAD(x) (__m128)__lsx_f16x4_load(x)
#define GGML_F32Cx4_STORE(x, y) __lsx_f16x4_store(x, y) #define GGML_F32Cx4_STORE(x, y) __lsx_f16x4_store(x, y)
#define GGML_F32Cx4_FMA GGML_F32x4_FMA #define GGML_F32Cx4_FMA GGML_F32x4_FMA
#define GGML_F32Cx4_ADD __lsx_vfadd_s #define GGML_F32Cx4_ADD __lsx_vfadd_s