Searched refs:MS_FLOAT16X8 (Results 1 – 7 of 7) sorted by relevance
/third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp16/ |
D | conv_depthwise_fp16.c | 35 MS_FLOAT16X8 v0, v1, v2, v3; in ConvDw3x3RowLeftFp16() 42 MS_FLOAT16X8 b0 = MS_SUBQ_F16(v0, v2); in ConvDw3x3RowLeftFp16() 43 MS_FLOAT16X8 b1 = MS_ADDQ_F16(v1, v2); in ConvDw3x3RowLeftFp16() 44 MS_FLOAT16X8 b2 = MS_SUBQ_F16(v2, v1); in ConvDw3x3RowLeftFp16() 45 MS_FLOAT16X8 b3 = MS_SUBQ_F16(v3, v1); in ConvDw3x3RowLeftFp16() 70 MS_FLOAT16X8 v0, v1, v2, v3; in ConvDw3x3RowMiddleFp16() 77 MS_FLOAT16X8 b0 = MS_SUBQ_F16(v0, v2); in ConvDw3x3RowMiddleFp16() 78 MS_FLOAT16X8 b1 = MS_ADDQ_F16(v1, v2); in ConvDw3x3RowMiddleFp16() 79 MS_FLOAT16X8 b2 = MS_SUBQ_F16(v2, v1); in ConvDw3x3RowMiddleFp16() 80 MS_FLOAT16X8 b3 = MS_SUBQ_F16(v3, v1); in ConvDw3x3RowMiddleFp16() [all …]
|
D | activation_fp16.c | 144 const MS_FLOAT16X8 zero_data = vdupq_n_f16(0); in HSwishFp16() 145 const MS_FLOAT16X8 three_data = vdupq_n_f16(3); in HSwishFp16() 146 const MS_FLOAT16X8 six_data = vdupq_n_f16(6); in HSwishFp16() 148 MS_FLOAT16X8 in_data = MS_LDQ_F16(src + i); in HSwishFp16() 149 MS_FLOAT16X8 tmp = MS_MAXQ_F16(in_data + three_data, zero_data); in HSwishFp16() 177 const MS_FLOAT16X8 zero_data = vdupq_n_f16(0); in HSigmoidFp16() 178 const MS_FLOAT16X8 three_data = vdupq_n_f16(3); in HSigmoidFp16() 179 const MS_FLOAT16X8 six_data = vdupq_n_f16(6); in HSigmoidFp16() 181 MS_FLOAT16X8 relu6_data = MS_LDQ_F16(src + offset) + three_data; in HSigmoidFp16()
|
D | exp_fp16.c | 55 MS_FLOAT16X8 scale = MS_MOVQ_F16(param->in_scale_); in ExpFusionFp16() 68 MS_FLOAT16X8 scale = MS_MOVQ_F16(param->out_scale_); in ExpFusionFp16()
|
D | pooling_fp16.c | 34 MS_FLOAT16X8 min_value = MS_MOVQ_F16(min); in AvgPoolingFp16() 35 MS_FLOAT16X8 max_value = MS_MOVQ_F16(max); in AvgPoolingFp16() 64 MS_FLOAT16X8 tmp_avg = MS_MOVQ_F16(0); in AvgPoolingFp16()
|
D | matmul_fp16.c | 829 MS_FLOAT16X8 src_data = MS_LDQ_F16(src + r * col + c); in RowMajor2RowNMajorFp16() 830 MS_FLOAT16X8 src_data1 = MS_LDQ_F16(src + r * col + c + C8NUM); in RowMajor2RowNMajorFp16() 835 MS_FLOAT16X8 src_data = MS_LDQ_F16(src + r * col + c); in RowMajor2RowNMajorFp16() 886 MS_FLOAT16X8 src_data = MS_LDQ_F16((const float16_t *)src + r * col + c * C8NUM); in RowMajor2Row8MajorFp16()
|
/third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp16_grad/ |
D | activation_grad.c | 70 const MS_FLOAT16X8 one_8 = vdupq_n_f16(1); in LReluFp16Grad() 72 MS_FLOAT16X8 src0_8 = MS_LDQ_F16(src0 + i); in LReluFp16Grad() 73 MS_FLOAT16X8 src1_8 = MS_LDQ_F16(src1 + i); in LReluFp16Grad()
|
/third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/intrinsics/ |
D | ms_simd_instructions_fp16.h | 95 #define MS_FLOAT16X8 float16x8_t macro
|