/dports/math/gemmlowp/gemmlowp-dc69acd/meta/ |
H A D | legacy_operations_common.h | 20 Quantized8BitOperation(std::int32_t lhs_offset, std::int32_t rhs_offset, in Quantized8BitOperation() argument 23 : lhs_offset(lhs_offset), in Quantized8BitOperation() 30 std::int32_t lhs_offset; 39 FloatOperation(std::int32_t lhs_offset, std::int32_t rhs_offset, in FloatOperation() argument 41 : lhs_offset(lhs_offset), in FloatOperation() 46 std::int32_t lhs_offset; 53 Int32Operation(std::int32_t lhs_offset, std::int32_t rhs_offset) in Int32Operation() argument 54 : lhs_offset(lhs_offset), rhs_offset(rhs_offset) {} in Int32Operation() 57 std::int32_t lhs_offset;
|
H A D | legacy_multi_thread_gemv.h | 43 gemv_q8(scratch, lhs, rhs, n, k, lhs_offset, rhs_offset, sum_offset, in ExecuteMatrixMatrix() 55 GemvFloatOperation(std::int32_t lhs_offset, std::int32_t rhs_offset, in GemvFloatOperation() argument 57 : FloatOperation(lhs_offset, rhs_offset, result_offset) {} in GemvFloatOperation() 63 gemv_f(scratch, lhs, rhs, n, k, lhs_offset, rhs_offset, result_offset, in ExecuteMatrixMatrix() 75 GemvInt32Operation(std::int32_t lhs_offset, std::int32_t rhs_offset) in GemvInt32Operation() argument 76 : Int32Operation(lhs_offset, rhs_offset) {} in GemvInt32Operation() 82 gemv_i32(scratch, lhs, rhs, n, k, lhs_offset, rhs_offset, result); in ExecuteMatrixMatrix() 102 std::int32_t k, std::int32_t lhs_offset, in multi_thread_gemv_q8() argument 107 internal::GemvQuantized8BitOperation operation(lhs_offset, rhs_offset, in multi_thread_gemv_q8() 126 std::int32_t k, std::int32_t lhs_offset, in multi_thread_gemv_f() argument [all …]
|
H A D | legacy_multi_thread_gemm.h | 81 gemm_q8_strided(scratch, lhs, rhs, m, n, k, lhs_offset, rhs_offset, in ExecuteCacheFriendlyMatrixMatrix() 93 GemmFloatOperation(std::int32_t lhs_offset, std::int32_t rhs_offset, in GemmFloatOperation() argument 95 : FloatOperation(lhs_offset, rhs_offset, result_offset) {} in GemmFloatOperation() 111 gemm_f_strided(scratch, lhs, rhs, m, n, k, lhs_offset, rhs_offset, in ExecuteCacheFriendlyMatrixMatrix() 123 GemmInt32Operation(std::int32_t lhs_offset, std::int32_t rhs_offset) in GemmInt32Operation() argument 124 : Int32Operation(lhs_offset, rhs_offset) {} in GemmInt32Operation() 176 internal::GemmQuantized8BitOperation operation(lhs_offset, rhs_offset, in multi_thread_gemm_q8() 205 lhs_offset, result_offset, result); in multi_thread_gemm_f() 234 lhs_offset, rhs_offset, result); in multi_thread_gemm_i32() 238 rhs_offset, lhs_offset, result); in multi_thread_gemm_i32() [all …]
|
H A D | legacy_single_thread_gemm.h | 31 std::int32_t k, std::int32_t lhs_offset, in gemm_q8_strided() argument 58 result_offset + k * lhs_offset * rhs_offset; in gemm_q8_strided() 62 params.right_stream.multiplicative_sum_offset = lhs_offset; in gemm_q8_strided() 76 std::int32_t lhs_offset, std::int32_t rhs_offset, in gemv_q8() argument 102 result_offset + k * lhs_offset * rhs_offset; in gemv_q8() 106 params.right_stream.multiplicative_sum_offset = lhs_offset; in gemv_q8() 124 std::int32_t k, std::int32_t lhs_offset, in gemm_i32_strided() argument 154 params.right_stream.multiplicative_sum_offset = lhs_offset; in gemm_i32_strided() 194 params.right_stream.multiplicative_sum_offset = lhs_offset; in gemv_i32() 209 std::int32_t k, std::int32_t lhs_offset, in gemm_f_strided() argument [all …]
|
/dports/math/gemmlowp/gemmlowp-dc69acd/public/ |
H A D | gemmlowp.h | 38 const LhsOffset& lhs_offset, in GemmWithOutputPipelinePC() argument 42 context, lhs, rhs, result, lhs_offset, rhs_offset, output_pipeline); in GemmWithOutputPipelinePC() 57 int lhs_offset, int rhs_offset, in GemmWithOutputPipeline() argument 61 const OffsetColDup lhs_offset_vector(lhs_offset, lhs.rows()); in GemmWithOutputPipeline() 77 MatrixMap<Scalar, ResultOrder>* result, int lhs_offset, in Gemm() argument 81 context, lhs, rhs, result, lhs_offset, rhs_offset, in Gemm()
|
/dports/math/gemmlowp/gemmlowp-dc69acd/internal/ |
H A D | unpack.h | 115 LoadForBroadcasting<RegisterBlockType>(lhs_offset, src_row); in UnpackResultBlock() 136 const LhsOffset& lhs_offset, const RhsOffset& rhs_offset, in UnpackResult() argument 189 lhs_sums_of_each_slice, rhs_sums_of_each_slice, lhs_offset, in UnpackResult() 202 lhs_sums_of_each_slice, rhs_sums_of_each_slice, lhs_offset, in UnpackResult() 214 lhs_sums_of_each_slice, rhs_sums_of_each_slice, lhs_offset, in UnpackResult() 231 rhs_sums_of_each_slice, lhs_offset, rhs_offset, depth, r, c, in UnpackResult() 238 rhs_sums_of_each_slice, lhs_offset, rhs_offset, depth, r, c, in UnpackResult() 245 rhs_sums_of_each_slice, lhs_offset, rhs_offset, depth, r, c, in UnpackResult() 258 rhs_sums_of_each_slice, lhs_offset, rhs_offset, depth, r, c, in UnpackResult() 265 rhs_sums_of_each_slice, lhs_offset, rhs_offset, depth, r, c, in UnpackResult() [all …]
|
H A D | multi_thread_gemm.h | 484 lhs_offset(_lhs_offset), in GemmWithPackedRhsTask() 518 lhs_offset.block(curr_result_block.start_row, rs), in Run() 532 const LhsOffset& lhs_offset; member 640 const LhsOffset& lhs_offset, const RhsOffset& rhs_offset, in MultiThreadGemm() argument 663 lhs_offset, rhs_offset, in MultiThreadGemm() 710 lhs_offset, rhs_offset, block_params, output_pipeline)); in MultiThreadGemm()
|
H A D | dispatch_gemm_shape.h | 175 const LhsOffset& lhs_offset, const RhsOffset& rhs_offset, 193 Transpose(rhs_offset), Transpose(lhs_offset), 200 lhs_offset, rhs_offset, output_pipeline);
|
H A D | single_thread_gemm.h | 71 const LhsOffset& lhs_offset, const RhsOffset& rhs_offset, in SingleThreadGemm() argument 148 lhs_offset.block(r, rs), rhs_offset.block(c, cs), output_pipeline); in SingleThreadGemm()
|
/dports/math/gemmlowp/gemmlowp-dc69acd/doc/ |
H A D | low-precision.md | 40 - A int32 lhs_offset, that will be added to each entry of the lhs matrix. 46 1. Cast lhs entries from uint8 to int32 and add lhs_offset to each of them. 64 be interpreted during the matrix computation: - lhs_offset - rhs_offset - 70 1. Cast lhs entries from uint8 to int32 and add lhs_offset to each of them. 105 One may then consider adding the lhs_offset and rhs_offset once and for all to 118 Adding lhs_offset to each entry of `lhs`, means adding `lhs_offset * P` to 128 (lhs + lhs_offset * P) * (rhs + rhs_offset * Q) 136 + lhs_offset * P * rhs 138 + lhs_offset * rhs_offset * P * Q 149 Thus, we can compute the second term, `lhs_offset * P * rhs`, by summing each [all …]
|
H A D | public.md | 21 `lhs_offset`, `rhs_offset` to them, is explained in 39 int lhs_offset, int rhs_offset, 49 &uint8_result_matrix, lhs_offset, rhs_offset, output_pipeline); 94 * `lhs_offset`, `rhs_offset` are constants added to each matrix entry in the 142 This is a variant where `lhs_offset` and `rhs_offset` may be vectors instead of
|
H A D | quantization_example.cc | 332 const int lhs_offset = -lhs_qparams.zero_point; in main() local 364 &actual_uint8_result_map, lhs_offset, rhs_offset, output_pipeline); in main()
|
/dports/math/gemmlowp/gemmlowp-dc69acd/test/ |
H A D | test.cc | 144 const OffsetColDup lhs_offset_vector(lhs_offset, rows); in Gemm() 185 const OffsetColDup lhs_offset_vector(lhs_offset, rows); in Gemm() 745 const OffsetColMap lhs_offset(a_offset, m); in TestWithSmallDataPerChannelQuantization() local 855 const OffsetColMap lhs_offset(a_offset, m); in TestWithLargeDataPerChannelQuantization() local 1042 const OffsetColMap lhs_offset(&a_offset[0], m); in TestMultithreadedPerChannelQuantization() local 1208 const int lhs_offset = 12; in TestOutputStages() local 1223 static_cast<std::int32_t>(lhs(r, d)) + lhs_offset; in TestOutputStages() 1242 lhs_offset, rhs_offset, quantize_down_pipeline); in TestOutputStages() 1336 lhs_offset, rhs_offset, row_bias_addition_pipeline); in TestOutputStages() 1357 lhs_offset, rhs_offset, col_bias_addition_pipeline); in TestOutputStages() [all …]
|
H A D | correctness_meta_gemm.cc | 64 std::int32_t lhs_offset, std::int32_t rhs_offset, in check_result() argument 74 (static_cast<std::int32_t>(left[depth * i + k]) + lhs_offset) * in check_result() 107 std::int32_t lhs_offset, std::int32_t rhs_offset, in check_result_f() argument 115 (static_cast<std::int32_t>(left[depth * i + k]) + lhs_offset) * in check_result_f() 141 std::int32_t lhs_offset, std::int32_t rhs_offset) { in check_result_i32() argument 148 (static_cast<std::int32_t>(left[depth * i + k]) + lhs_offset) * in check_result_i32()
|
/dports/math/gemmlowp/gemmlowp-dc69acd/eight_bit_int_gemm/ |
H A D | eight_bit_int_gemm.cc | 75 const int lhs_offset = a_offset; in EightBitIntGemmImpl() local 96 context, lhs, rhs, &result, lhs_offset, rhs_offset, result_offset, \ in EightBitIntGemmImpl() 113 const int lhs_offset = a_offset; in EightBitIntGemmInt32Impl() local 133 context, lhs, rhs, &result, lhs_offset, rhs_offset, empty_pipeline); \ in EightBitIntGemmInt32Impl() 251 std::int32_t lhs_offset, std::int32_t rhs_offset, in MetaGemmQuantized8Bit() argument 261 scratch->buffer(), lhs, rhs, m, n, k, lhs_offset, in MetaGemmQuantized8Bit() 268 lhs_offset, sum_offset, multiplicative_offset, in MetaGemmQuantized8Bit() 277 std::int32_t lhs_offset, std::int32_t rhs_offset, in MetaGemmFloat() argument 285 scratch->buffer(), lhs, rhs, m, n, k, lhs_offset, in MetaGemmFloat() 291 lhs_offset, result_offset, result); in MetaGemmFloat()
|
/dports/math/gemmlowp/gemmlowp-dc69acd/meta/generators/ |
H A D | quantized_mul_kernels_common.py | 58 lhs_offset = _ReadParams(emitter, registers, lhs, kernel_m, 4) 71 self.lhs_offsets = _Duplicate(emitter, registers, kernel_m, lhs_offset) 79 for (row, lhs_offset) in zip(data, self.lhs_offsets): 81 emitter.EmitVAdd('s32', row_register, row_register, lhs_offset) 134 lhs_offset = _ReadParams(emitter, registers, lhs, kernel_m, 4) 136 self.lhs_offsets = _Duplicate(emitter, registers, kernel_m, lhs_offset) 144 for (row, lhs_offset) in zip(data, self.lhs_offsets): 146 emitter.EmitVAdd('s32', row_register, row_register, lhs_offset) 166 lhs_offset = _ReadParams(emitter, registers, lhs, kernel_m, 4) 179 for (row, lhs_offset) in zip(data, self.lhs_offsets): [all …]
|
/dports/lang/gcc48/gcc-4.8.5/gcc/ |
H A D | ipa-prop.c | 1277 HOST_WIDE_INT lhs_offset, lhs_size, lhs_max_size; in determine_known_aggregate_parts() local 1297 || (lhs_offset < arg_offset in determine_known_aggregate_parts() 1298 && lhs_offset + lhs_size > arg_offset) in determine_known_aggregate_parts() 1299 || (lhs_offset < arg_offset + arg_size in determine_known_aggregate_parts() 1300 && lhs_offset + lhs_size > arg_offset + arg_size)) in determine_known_aggregate_parts() 1318 if (lhs_offset + lhs_size < arg_offset in determine_known_aggregate_parts() 1319 || lhs_offset >= (arg_offset + arg_size)) in determine_known_aggregate_parts() 1324 while (*p && (*p)->offset < lhs_offset) in determine_known_aggregate_parts() 1326 if ((*p)->offset + (*p)->size > lhs_offset) in determine_known_aggregate_parts() 1335 if (*p && (*p)->offset < lhs_offset + lhs_size) in determine_known_aggregate_parts() [all …]
|
/dports/devel/mingw32-gcc/gcc-4.8.1/gcc/ |
H A D | ipa-prop.c | 1260 HOST_WIDE_INT lhs_offset, lhs_size, lhs_max_size; in determine_known_aggregate_parts() local 1280 || (lhs_offset < arg_offset in determine_known_aggregate_parts() 1281 && lhs_offset + lhs_size > arg_offset) in determine_known_aggregate_parts() 1282 || (lhs_offset < arg_offset + arg_size in determine_known_aggregate_parts() 1283 && lhs_offset + lhs_size > arg_offset + arg_size)) in determine_known_aggregate_parts() 1301 if (lhs_offset + lhs_size < arg_offset in determine_known_aggregate_parts() 1302 || lhs_offset >= (arg_offset + arg_size)) in determine_known_aggregate_parts() 1307 while (*p && (*p)->offset < lhs_offset) in determine_known_aggregate_parts() 1309 if ((*p)->offset + (*p)->size > lhs_offset) in determine_known_aggregate_parts() 1318 if (*p && (*p)->offset < lhs_offset + lhs_size) in determine_known_aggregate_parts() [all …]
|
/dports/devel/arm-none-eabi-gcc492/gcc-4.9.2/gcc/ |
H A D | ipa-prop.c | 1448 HOST_WIDE_INT lhs_offset, lhs_size, lhs_max_size; in determine_known_aggregate_parts() local 1468 || (lhs_offset < arg_offset in determine_known_aggregate_parts() 1469 && lhs_offset + lhs_size > arg_offset) in determine_known_aggregate_parts() 1470 || (lhs_offset < arg_offset + arg_size in determine_known_aggregate_parts() 1471 && lhs_offset + lhs_size > arg_offset + arg_size)) in determine_known_aggregate_parts() 1489 if (lhs_offset + lhs_size < arg_offset in determine_known_aggregate_parts() 1490 || lhs_offset >= (arg_offset + arg_size)) in determine_known_aggregate_parts() 1495 while (*p && (*p)->offset < lhs_offset) in determine_known_aggregate_parts() 1497 if ((*p)->offset + (*p)->size > lhs_offset) in determine_known_aggregate_parts() 1506 if (*p && (*p)->offset < lhs_offset + lhs_size) in determine_known_aggregate_parts() [all …]
|
/dports/lang/gcc6-aux/gcc-6-20180516/gcc/ |
H A D | tree-ssa-sccvn.c | 2058 HOST_WIDE_INT rhs_offset, copy_size, lhs_offset; in vn_reference_lookup_3() local 2070 lhs_offset = 0; in vn_reference_lookup_3() 2085 &lhs_offset); in vn_reference_lookup_3() 2094 lhs_offset += tree_to_uhwi (TREE_OPERAND (tem, 1)); in vn_reference_lookup_3() 2149 if (lhs_offset >= at + maxsize / BITS_PER_UNIT in vn_reference_lookup_3() 2150 || lhs_offset + copy_size <= at) in vn_reference_lookup_3() 2153 if (lhs_offset > at in vn_reference_lookup_3() 2154 || lhs_offset + copy_size < at + maxsize / BITS_PER_UNIT) in vn_reference_lookup_3() 2172 op.op0 = build_int_cst (ptr_type_node, at - lhs_offset + rhs_offset); in vn_reference_lookup_3() 2173 op.off = at - lhs_offset + rhs_offset; in vn_reference_lookup_3()
|
/dports/lang/gnat_util/gcc-6-20180516/gcc/ |
H A D | tree-ssa-sccvn.c | 2058 HOST_WIDE_INT rhs_offset, copy_size, lhs_offset; in vn_reference_lookup_3() local 2070 lhs_offset = 0; in vn_reference_lookup_3() 2085 &lhs_offset); in vn_reference_lookup_3() 2094 lhs_offset += tree_to_uhwi (TREE_OPERAND (tem, 1)); in vn_reference_lookup_3() 2149 if (lhs_offset >= at + maxsize / BITS_PER_UNIT in vn_reference_lookup_3() 2150 || lhs_offset + copy_size <= at) in vn_reference_lookup_3() 2153 if (lhs_offset > at in vn_reference_lookup_3() 2154 || lhs_offset + copy_size < at + maxsize / BITS_PER_UNIT) in vn_reference_lookup_3() 2172 op.op0 = build_int_cst (ptr_type_node, at - lhs_offset + rhs_offset); in vn_reference_lookup_3() 2173 op.off = at - lhs_offset + rhs_offset; in vn_reference_lookup_3()
|
/dports/math/py-z3-solver/z3-z3-4.8.10/src/ast/rewriter/ |
H A D | poly_rewriter_def.h | 877 const unsigned lhs_offset = insert_c_lhs ? 0 : 1; in cancel_monomials() local 881 …lhs_result = mk_add_app(new_lhs_monomials.size() - lhs_offset, new_lhs_monomials.c_ptr() + lhs_off… in cancel_monomials()
|
/dports/math/z3/z3-z3-4.8.13/src/ast/rewriter/ |
H A D | poly_rewriter_def.h | 877 const unsigned lhs_offset = insert_c_lhs ? 0 : 1; in cancel_monomials() local 881 …lhs_result = mk_add_app(new_lhs_monomials.size() - lhs_offset, new_lhs_monomials.data() + lhs_offs… in cancel_monomials()
|
/dports/devel/aarch64-none-elf-gcc/gcc-8.4.0/gcc/ |
H A D | ipa-prop.c | 1490 HOST_WIDE_INT lhs_offset, in get_place_in_agg_contents_list() argument 1495 while (*p && (*p)->offset < lhs_offset) in get_place_in_agg_contents_list() 1497 if ((*p)->offset + (*p)->size > lhs_offset) in get_place_in_agg_contents_list() 1502 if (*p && (*p)->offset < lhs_offset + lhs_size) in get_place_in_agg_contents_list() 1504 if ((*p)->offset == lhs_offset && (*p)->size == lhs_size) in get_place_in_agg_contents_list() 1628 HOST_WIDE_INT lhs_offset, lhs_size; in determine_locally_known_aggregate_parts() local 1644 lhs_base = get_ref_base_and_extent_hwi (lhs, &lhs_offset, in determine_locally_known_aggregate_parts() 1665 p = get_place_in_agg_contents_list (&list, lhs_offset, lhs_size, in determine_locally_known_aggregate_parts() 1675 n->offset = lhs_offset; in determine_locally_known_aggregate_parts()
|
/dports/lang/gcc9/gcc-9.4.0/gcc/ |
H A D | ipa-prop.c | 1482 HOST_WIDE_INT lhs_offset, in get_place_in_agg_contents_list() argument 1487 while (*p && (*p)->offset < lhs_offset) in get_place_in_agg_contents_list() 1489 if ((*p)->offset + (*p)->size > lhs_offset) in get_place_in_agg_contents_list() 1494 if (*p && (*p)->offset < lhs_offset + lhs_size) in get_place_in_agg_contents_list() 1496 if ((*p)->offset == lhs_offset && (*p)->size == lhs_size) in get_place_in_agg_contents_list() 1620 HOST_WIDE_INT lhs_offset, lhs_size; in determine_locally_known_aggregate_parts() local 1636 lhs_base = get_ref_base_and_extent_hwi (lhs, &lhs_offset, in determine_locally_known_aggregate_parts() 1657 p = get_place_in_agg_contents_list (&list, lhs_offset, lhs_size, in determine_locally_known_aggregate_parts() 1667 n->offset = lhs_offset; in determine_locally_known_aggregate_parts()
|