diff options
Diffstat (limited to 'multiply.h')
| -rw-r--r-- | multiply.h | 8 |
1 files changed, 4 insertions, 4 deletions
@@ -135,9 +135,9 @@ INTGEMM_PACK0123(INTGEMM_AVX512BW, __m512i) const int simd_width = width / (sizeof(Integer) / sizeof(int16_t)); \ typename WriteC::WriteCSubType write_C(C); \ const Integer *B0_col = reinterpret_cast<const Integer *>(B); \ - for (int B0_colidx = 0; B0_colidx < B_cols; B0_col += 8 * simd_width, B0_colidx += 8) { \ + for (Index B0_colidx = 0; B0_colidx < B_cols; B0_col += 8 * simd_width, B0_colidx += 8) { \ /* Process one row of A at a time. Doesn't seem to be faster to do multiple rows of A at once.*/ \ - for (int A_rowidx = 0; A_rowidx < A_rows; ++A_rowidx) { \ + for (Index A_rowidx = 0; A_rowidx < A_rows; ++A_rowidx) { \ const Integer *A_row = reinterpret_cast<const Integer*>(A + A_rowidx * width); \ /* These will be packed 32-bit integers containing sums for each row of B multiplied by the row of A. \ Iterate over shared (inner) dimension.*/ \ @@ -340,9 +340,9 @@ template <class WriteC> target static void Multiply(const int8_t *A, const int8_ const Integer *B0_col = reinterpret_cast<const Integer*>(B); \ typename WriteC::WriteCSubType c_writer(C); \ /*Go over 8 columns of B at a time.*/ \ - for (int B0_colidx = 0; B0_colidx != B_cols; B0_col += 8 * simd_width, B0_colidx += 8) { \ + for (Index B0_colidx = 0; B0_colidx != B_cols; B0_col += 8 * simd_width, B0_colidx += 8) { \ /*Process one row of A at a time. Doesn't seem to be faster to do multiple rows of A at once.*/ \ - for (int A_rowidx = 0; A_rowidx < A_rows; ++A_rowidx) { \ + for (Index A_rowidx = 0; A_rowidx < A_rows; ++A_rowidx) { \ /*Iterate over shared (inner) dimension.*/ \ const Integer *A_live = reinterpret_cast<const Integer *>(A + A_rowidx * width); \ const Integer *A_end = A_live + simd_width; \ |