Skip to content

Commit 4b1cdc1

Browse files
YixingZhang007igcbot
YixingZhang007
authored and
igcbot
committed
Resolve incorrect result issue for Accumulator 32x64 checked_store
Update the memory location calculation in DEFINE_STORE_CHECKED_LARGE_IMPL_16 within Source/IGC/BiFModule/Languages/OpenCL/PreRelease/IBiF_matrix.cl to resolve the incorrect results produced by the built-in DEFINE_STORE_CHECKED_LARGE(Accumulator_RowMajor, int, int, 32, 64, 128).
1 parent cc0aa07 commit 4b1cdc1

File tree

1 file changed

+16
-16
lines changed

1 file changed

+16
-16
lines changed

‎IGC/BiFModule/Languages/OpenCL/PreRelease/IBiF_matrix.cl‎

Lines changed: 16 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -1644,22 +1644,22 @@ DEFINE_LOAD_CHECKED_LARGE(Accumulator_RowMajor, int, int, 32, 64, 128, 4)
16441644
__private char *c13 = src + 13 * 8 * (sizeof (contrib_type)); \
16451645
__private char *c14 = src + 14 * 8 * (sizeof (contrib_type)); \
16461646
__private char *c15 = src + 15 * 8 * (sizeof (contrib_type)); \
1647-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c0, y + 0 * column_stride, x + 0 * row_stride, height, width, stride, cacheOpt); \
1648-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c1, y + 1 * column_stride, x + 0 * row_stride, height, width, stride, cacheOpt); \
1649-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c2, y + 2 * column_stride, x + 0 * row_stride, height, width, stride, cacheOpt); \
1650-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c3, y + 3 * column_stride, x + 0 * row_stride, height, width, stride, cacheOpt); \
1651-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c4, y + 0 * column_stride, x + 1 * row_stride, height, width, stride, cacheOpt); \
1652-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c5, y + 1 * column_stride, x + 1 * row_stride, height, width, stride, cacheOpt); \
1653-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c6, y + 2 * column_stride, x + 1 * row_stride, height, width, stride, cacheOpt); \
1654-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c7, y + 3 * column_stride, x + 1 * row_stride, height, width, stride, cacheOpt); \
1655-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c8, y + 0 * column_stride, x + 2 * row_stride, height, width, stride, cacheOpt); \
1656-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c9, y + 1 * column_stride, x + 2 * row_stride, height, width, stride, cacheOpt); \
1657-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c10, y + 2 * column_stride, x + 2 * row_stride, height, width, stride, cacheOpt); \
1658-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c11, y + 3 * column_stride, x + 2 * row_stride, height, width, stride, cacheOpt); \
1659-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c12, y + 0 * column_stride, x + 3 * row_stride, height, width, stride, cacheOpt); \
1660-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c13, y + 1 * column_stride, x + 3 * row_stride, height, width, stride, cacheOpt); \
1661-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c14, y + 2 * column_stride, x + 3 * row_stride, height, width, stride, cacheOpt); \
1662-
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c15, y + 3 * column_stride, x + 3 * row_stride, height, width, stride, cacheOpt); \
1647+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c0, y + 0 * row_stride, x + 0 * column_stride, height, width, stride, cacheOpt); \
1648+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c1, y + 1 * row_stride, x + 0 * column_stride, height, width, stride, cacheOpt); \
1649+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c2, y + 2 * row_stride, x + 0 * column_stride, height, width, stride, cacheOpt); \
1650+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c3, y + 3 * row_stride, x + 0 * column_stride, height, width, stride, cacheOpt); \
1651+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c4, y + 0 * row_stride, x + 1 * column_stride, height, width, stride, cacheOpt); \
1652+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c5, y + 1 * row_stride, x + 1 * column_stride, height, width, stride, cacheOpt); \
1653+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c6, y + 2 * row_stride, x + 1 * column_stride, height, width, stride, cacheOpt); \
1654+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c7, y + 3 * row_stride, x + 1 * column_stride, height, width, stride, cacheOpt); \
1655+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c8, y + 0 * row_stride, x + 2 * column_stride, height, width, stride, cacheOpt); \
1656+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c9, y + 1 * row_stride, x + 2 * column_stride, height, width, stride, cacheOpt); \
1657+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c10, y + 2 * row_stride, x + 2 * column_stride, height, width, stride, cacheOpt); \
1658+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c11, y + 3 * row_stride, x + 2 * column_stride, height, width, stride, cacheOpt); \
1659+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c12, y + 0 * row_stride, x + 3 * column_stride, height, width, stride, cacheOpt); \
1660+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c13, y + 1 * row_stride, x + 3 * column_stride, height, width, stride, cacheOpt); \
1661+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c14, y + 2 * row_stride, x + 3 * column_stride, height, width, stride, cacheOpt); \
1662+
__builtin_spriv_OpJointMatrixStoreCheckedINTEL_Accumulator_RowMajor_SG16_##store_shape##_i32_8_pi64_v8i8(mem, c15, y + 3 * row_stride, x + 3 * column_stride, height, width, stride, cacheOpt); \
16631663
}
16641664

16651665
#define DEFINE_STORE_LARGE_IMPL_(layout, sg, elem_type, elem_bitwidth, contrib_type, shape, WI_rows, num_stores, row_stride, column_stride, store_shape) \

0 commit comments

Comments
 (0)