mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-12-25 02:44:36 +00:00
ggml : fix error C2078: too many initializers
for MSVC ARM64 (#5404)
This commit is contained in:
parent
44fbe34360
commit
b2f87cb64d
@ -268,6 +268,17 @@ static inline float hsum_float_4x4(const __m128 a, const __m128 b, const __m128
|
||||
#endif // defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__)
|
||||
|
||||
#if defined(__ARM_NEON)
|
||||
|
||||
#ifdef _MSC_VER
|
||||
|
||||
#define ggml_vld1q_u32(w,x,y,z) { ((w) + ((uint64_t)(x) << 32)), ((y) + ((uint64_t)(z) << 32)) }
|
||||
|
||||
#else
|
||||
|
||||
#define ggml_vld1q_u32(w,x,y,z) { (w), (x), (y), (z) }
|
||||
|
||||
#endif
|
||||
|
||||
#if !defined(__aarch64__)
|
||||
|
||||
// 64-bit compatibility
|
||||
@ -8698,10 +8709,10 @@ void ggml_vec_dot_iq3_xxs_q8_K(const int n, float * restrict s, const void * res
|
||||
for (int ib32 = 0; ib32 < QK_K/32; ib32 += 2) {
|
||||
q8b = ggml_vld1q_s8_x4(q8); q8 += 64;
|
||||
memcpy(aux32, gas, 2*sizeof(uint32_t)); gas += 2*sizeof(uint32_t);
|
||||
const uint32x4_t aux32x4_0 = {iq3xxs_grid[q3[ 0]], iq3xxs_grid[q3[ 1]], iq3xxs_grid[q3[ 2]], iq3xxs_grid[q3[ 3]]};
|
||||
const uint32x4_t aux32x4_1 = {iq3xxs_grid[q3[ 4]], iq3xxs_grid[q3[ 5]], iq3xxs_grid[q3[ 6]], iq3xxs_grid[q3[ 7]]};
|
||||
const uint32x4_t aux32x4_2 = {iq3xxs_grid[q3[ 8]], iq3xxs_grid[q3[ 9]], iq3xxs_grid[q3[10]], iq3xxs_grid[q3[11]]};
|
||||
const uint32x4_t aux32x4_3 = {iq3xxs_grid[q3[12]], iq3xxs_grid[q3[13]], iq3xxs_grid[q3[14]], iq3xxs_grid[q3[15]]};
|
||||
const uint32x4_t aux32x4_0 = ggml_vld1q_u32(iq3xxs_grid[q3[ 0]], iq3xxs_grid[q3[ 1]], iq3xxs_grid[q3[ 2]], iq3xxs_grid[q3[ 3]]);
|
||||
const uint32x4_t aux32x4_1 = ggml_vld1q_u32(iq3xxs_grid[q3[ 4]], iq3xxs_grid[q3[ 5]], iq3xxs_grid[q3[ 6]], iq3xxs_grid[q3[ 7]]);
|
||||
const uint32x4_t aux32x4_2 = ggml_vld1q_u32(iq3xxs_grid[q3[ 8]], iq3xxs_grid[q3[ 9]], iq3xxs_grid[q3[10]], iq3xxs_grid[q3[11]]);
|
||||
const uint32x4_t aux32x4_3 = ggml_vld1q_u32(iq3xxs_grid[q3[12]], iq3xxs_grid[q3[13]], iq3xxs_grid[q3[14]], iq3xxs_grid[q3[15]]);
|
||||
q3 += 16;
|
||||
q3s.val[0] = vcombine_s8(vld1_s8((const void *)(signs64 + ((aux32[0] >> 0) & 127))), vld1_s8((const void *)(signs64 + ((aux32[0] >> 7) & 127))));
|
||||
q3s.val[1] = vcombine_s8(vld1_s8((const void *)(signs64 + ((aux32[0] >> 14) & 127))), vld1_s8((const void *)(signs64 + ((aux32[0] >> 21) & 127))));
|
||||
|
Loading…
Reference in New Issue
Block a user