CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml : fix UB in IQ2_S and IQ3_S (#6012)

Browse Source
This commit is contained in:
Georgi Gerganov
2024-03-12 13:49:55 +02:00
committed by GitHub
parent 48358b2e5b
commit 184215e783
1 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
ggml-quants.c
View File

@@ -9025,7 +9025,7 @@ void ggml_vec_dot_iq2_s_q8_K(int n, float * restrict s, size_t bs, const void *
vld1_s8((const int8_t *)(iq2s_grid + (qs[7] | ((qh[ib32+1] << 2) & 0x300))))); vld1_s8((const int8_t *)(iq2s_grid + (qs[7] | ((qh[ib32+1] << 2) & 0x300)))));
qs += 8; qs += 8;
vs.val[0] = vreinterpretq_u8_u32(vdupq_n_u32(signs[0] | (signs[1] << 16))); vs.val[0] = vreinterpretq_u8_u32(vdupq_n_u32(signs[0] | ((uint32_t) signs[1] << 16)));
vs.val[1] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2); vs.val[1] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2);
vs.val[0] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2); vs.val[0] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2);
vs.val[0] = vceqq_u8(vs.val[0], mask2); vs.val[0] = vceqq_u8(vs.val[0], mask2);
@@ -9034,7 +9034,7 @@ void ggml_vec_dot_iq2_s_q8_K(int n, float * restrict s, size_t bs, const void *
q2s.val[0] = vmulq_s8(vreinterpretq_s8_u8(vorrq_u8(vs.val[0], m1)), q2s.val[0]); q2s.val[0] = vmulq_s8(vreinterpretq_s8_u8(vorrq_u8(vs.val[0], m1)), q2s.val[0]);
q2s.val[1] = vmulq_s8(vreinterpretq_s8_u8(vorrq_u8(vs.val[1], m1)), q2s.val[1]); q2s.val[1] = vmulq_s8(vreinterpretq_s8_u8(vorrq_u8(vs.val[1], m1)), q2s.val[1]);
vs.val[0] = vreinterpretq_u8_u32(vdupq_n_u32(signs[2] | (signs[3] << 16))); vs.val[0] = vreinterpretq_u8_u32(vdupq_n_u32(signs[2] | ((uint32_t) signs[3] << 16)));
vs.val[1] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2); vs.val[1] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2);
vs.val[0] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2); vs.val[0] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2);
vs.val[0] = vceqq_u8(vs.val[0], mask2); vs.val[0] = vceqq_u8(vs.val[0], mask2);
@@ -9105,12 +9105,12 @@ void ggml_vec_dot_iq2_s_q8_K(int n, float * restrict s, size_t bs, const void *
iq2s_grid[qs[4] | ((qh[ib32+1] << 8) & 0x300)]); iq2s_grid[qs[4] | ((qh[ib32+1] << 8) & 0x300)]);
qs += 8; qs += 8;
__m256i aux256 = _mm256_set1_epi32(signs[0] | (signs[1] << 16)); __m256i aux256 = _mm256_set1_epi32(signs[0] | ((uint32_t) signs[1] << 16));
aux256 = _mm256_and_si256(_mm256_shuffle_epi8(aux256,mask1), mask2); aux256 = _mm256_and_si256(_mm256_shuffle_epi8(aux256,mask1), mask2);
const __m256i s2_1 = _mm256_cmpeq_epi8(aux256, mask2); const __m256i s2_1 = _mm256_cmpeq_epi8(aux256, mask2);
const __m256i q8s_1 = _mm256_sub_epi8(_mm256_xor_si256(s2_1, q8_1), s2_1); const __m256i q8s_1 = _mm256_sub_epi8(_mm256_xor_si256(s2_1, q8_1), s2_1);
aux256 = _mm256_set1_epi32(signs[2] | (signs[3] << 16)); aux256 = _mm256_set1_epi32(signs[2] | ((uint32_t) signs[3] << 16));
aux256 = _mm256_and_si256(_mm256_shuffle_epi8(aux256,mask1), mask2); aux256 = _mm256_and_si256(_mm256_shuffle_epi8(aux256,mask1), mask2);
const __m256i s2_2 = _mm256_cmpeq_epi8(aux256, mask2); const __m256i s2_2 = _mm256_cmpeq_epi8(aux256, mask2);
const __m256i q8s_2 = _mm256_sub_epi8(_mm256_xor_si256(s2_2, q8_2), s2_2); const __m256i q8s_2 = _mm256_sub_epi8(_mm256_xor_si256(s2_2, q8_2), s2_2);
@@ -9386,7 +9386,7 @@ void ggml_vec_dot_iq3_s_q8_K (int n, float * restrict s, size_t bs, const void *
iq3s_grid[idx.index[6]], iq3s_grid[idx.index[7]]); iq3s_grid[idx.index[6]], iq3s_grid[idx.index[7]]);
vs.val[0] = vreinterpretq_u8_u32(vdupq_n_u32(signs[0] | (signs[1] << 16))); vs.val[0] = vreinterpretq_u8_u32(vdupq_n_u32(signs[0] | ((uint32_t) signs[1] << 16)));
vs.val[1] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2); vs.val[1] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2);
vs.val[0] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2); vs.val[0] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2);
vs.val[0] = vorrq_u8(vceqq_u8(vs.val[0], mask2), m1); vs.val[0] = vorrq_u8(vceqq_u8(vs.val[0], mask2), m1);
@@ -9395,7 +9395,7 @@ void ggml_vec_dot_iq3_s_q8_K (int n, float * restrict s, size_t bs, const void *
q3s.val[0] = vmulq_s8(vreinterpretq_s8_u8(vs.val[0]), vreinterpretq_s8_u32(aux32x4_0)); q3s.val[0] = vmulq_s8(vreinterpretq_s8_u8(vs.val[0]), vreinterpretq_s8_u32(aux32x4_0));
q3s.val[1] = vmulq_s8(vreinterpretq_s8_u8(vs.val[1]), vreinterpretq_s8_u32(aux32x4_1)); q3s.val[1] = vmulq_s8(vreinterpretq_s8_u8(vs.val[1]), vreinterpretq_s8_u32(aux32x4_1));
vs.val[0] = vreinterpretq_u8_u32(vdupq_n_u32(signs[2] | (signs[3] << 16))); vs.val[0] = vreinterpretq_u8_u32(vdupq_n_u32(signs[2] | ((uint32_t) signs[3] << 16)));
vs.val[1] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2); vs.val[1] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[1]), mask2);
vs.val[0] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2); vs.val[0] = vandq_u8(ggml_vqtbl1q_u8(vs.val[0], mask1.val[0]), mask2);
vs.val[0] = vorrq_u8(vceqq_u8(vs.val[0], mask2), m1); vs.val[0] = vorrq_u8(vceqq_u8(vs.val[0], mask2), m1);