diff options
Diffstat (limited to 'src/core/NEON/kernels/convolution/depthwise/depthwise_3x3_3x3_2x2_fp32_fp32.cpp')
| -rw-r--r-- | src/core/NEON/kernels/convolution/depthwise/depthwise_3x3_3x3_2x2_fp32_fp32.cpp | 70 |
1 files changed, 35 insertions, 35 deletions
diff --git a/src/core/NEON/kernels/convolution/depthwise/depthwise_3x3_3x3_2x2_fp32_fp32.cpp b/src/core/NEON/kernels/convolution/depthwise/depthwise_3x3_3x3_2x2_fp32_fp32.cpp index adc6969fcb..eb2b37a5b0 100644 --- a/src/core/NEON/kernels/convolution/depthwise/depthwise_3x3_3x3_2x2_fp32_fp32.cpp +++ b/src/core/NEON/kernels/convolution/depthwise/depthwise_3x3_3x3_2x2_fp32_fp32.cpp @@ -53,10 +53,10 @@ void Conv::execute_tile<ActivationFunction::None>( "add x17, x16, %[input_row_stride]\n" "add x28, x27, %[input_col_stride1]\n" "add x23, %[output_col_stride1], %[output_col_stride1]\n" - "add x18, x17, %[input_row_stride]\n" + "add x9, x17, %[input_row_stride]\n" "add x13, x28, %[input_col_stride1]\n" "and x24, %[n_channels], #3\n" - "add x19, x18, %[input_row_stride]\n" + "add x19, x9, %[input_row_stride]\n" "add x14, x13, %[input_col_stride1]\n" "lsr x25, %[n_channels], #2\n" "add x20, x19, %[input_row_stride]\n" @@ -99,7 +99,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v15.4s, v19.4s, v6.4s\n" "add %[wbptr], %[wbptr], #160\n" "fmla v17.4s, v28.4s, v12.4s\n" - "ldr q25, [x18]\n" + "ldr q25, [x9]\n" "fmla v16.4s, v30.4s, v12.4s\n" "ldr q24, [x17, %[input_col_stride1]]\n" "fmla v15.4s, v21.4s, v12.4s\n" @@ -117,7 +117,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v15.4s, v23.4s, v9.4s\n" "ldr q30, [x19]\n" "fmla v17.4s, v20.4s, v11.4s\n" - "ldr q29, [x18, %[input_col_stride1]]\n" + "ldr q29, [x9, %[input_col_stride1]]\n" "fmla v16.4s, v24.4s, v11.4s\n" "ldr q28, [x17, x26]\n" "fmla v4.4s, v23.4s, v6.4s\n" @@ -135,7 +135,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v16.4s, v29.4s, v8.4s\n" "ldr q22, [x19, %[input_col_stride1]]\n" "fmla v17.4s, v21.4s, v10.4s\n" - "ldr q26, [x18, x26]\n" + "ldr q26, [x9, x26]\n" "fmla v2.4s, v29.4s, v14.4s\n" "ldr q20, [x17, x27]\n" "fmla v16.4s, v28.4s, v10.4s\n" @@ -153,7 +153,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v2.4s, v18.4s, v9.4s\n" "ldr q17, [x19, x26]\n" "fmla v5.4s, v19.4s, v14.4s\n" - "ldr q18, [x18, x27]\n" + "ldr q18, [x9, x27]\n" "fmla v16.4s, v26.4s, v7.4s\n" "ldr q25, [x17, x28]\n" "fmla v2.4s, v22.4s, v11.4s\n" @@ -175,7 +175,7 @@ void Conv::execute_tile<ActivationFunction::None>( "str q15, [%[outptr0], %[output_col_stride1]]\n" "fmla v2.4s, v28.4s, v8.4s\n" "fmla v5.4s, v30.4s, v11.4s\n" - "ldr q29, [x18, x28]\n" + "ldr q29, [x9, x28]\n" "fmla v1.4s, v17.4s, v12.4s\n" "ldr q27, [x17, x13]\n" "fmla v2.4s, v17.4s, v10.4s\n" @@ -185,7 +185,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v4.4s, v18.4s, v8.4s\n" "ldr q20, [x19, x28]\n" "fmla v1.4s, v18.4s, v14.4s\n" - "ldr q17, [x18, x13]\n" + "ldr q17, [x9, x13]\n" "fmla v3.4s, v25.4s, v12.4s\n" "ldr q18, [x17, x14]\n" "fmla v4.4s, v25.4s, v10.4s\n" @@ -197,7 +197,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v2.4s, v23.4s, v7.4s\n" "add x17, x17, #16\n" "fmla v5.4s, v21.4s, v10.4s\n" - "ldr q21, [x18, x14]\n" + "ldr q21, [x9, x14]\n" "fmla v1.4s, v23.4s, v9.4s\n" "ldr q23, [x20, x13]\n" "str q2, [x22]\n" @@ -241,7 +241,7 @@ void Conv::execute_tile<ActivationFunction::None>( "str q3, [x21, x23]\n" "mov v16.16b, v27.16b\n" "mov v15.16b, v27.16b\n" - "add x18, x18, #16\n" + "add x9, x9, #16\n" "fmla v0.4s, v21.4s, v13.4s\n" "ldr q11, [%[wbptr], #80]\n" "mov v2.16b, v27.16b\n" @@ -272,7 +272,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v15.4s, v19.4s, v6.4s\n" "add %[wbptr], %[wbptr], #160\n" "fmla v17.4s, v28.4s, v12.4s\n" - "ldr q25, [x18]\n" + "ldr q25, [x9]\n" "fmla v16.4s, v30.4s, v12.4s\n" "ldr q24, [x17, %[input_col_stride1]]\n" "fmla v15.4s, v21.4s, v12.4s\n" @@ -290,7 +290,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v15.4s, v23.4s, v9.4s\n" "fmla v4.4s, v23.4s, v6.4s\n" "fmla v17.4s, v20.4s, v11.4s\n" - "ldr q29, [x18, %[input_col_stride1]]\n" + "ldr q29, [x9, %[input_col_stride1]]\n" "fmla v16.4s, v24.4s, v11.4s\n" "ldr q28, [x17, x26]\n" "fmla v15.4s, v26.4s, v11.4s\n" @@ -308,7 +308,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v16.4s, v29.4s, v8.4s\n" "fmla v4.4s, v28.4s, v12.4s\n" "fmla v17.4s, v21.4s, v10.4s\n" - "ldr q26, [x18, x26]\n" + "ldr q26, [x9, x26]\n" "fmla v2.4s, v29.4s, v14.4s\n" "ldr q20, [x17, x27]\n" "fmla v16.4s, v28.4s, v10.4s\n" @@ -326,7 +326,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v5.4s, v19.4s, v14.4s\n" "ldr q17, [x19, x26]\n" "fmla v2.4s, v22.4s, v11.4s\n" - "ldr q18, [x18, x27]\n" + "ldr q18, [x9, x27]\n" "fmla v16.4s, v26.4s, v7.4s\n" "ldr q25, [x17, x28]\n" "fmla v4.4s, v26.4s, v9.4s\n" @@ -346,7 +346,7 @@ void Conv::execute_tile<ActivationFunction::None>( "str q15, [%[outptr0], %[output_col_stride1]]\n" "fmla v3.4s, v27.4s, v6.4s\n" "fmla v5.4s, v30.4s, v11.4s\n" - "ldr q29, [x18, x28]\n" + "ldr q29, [x9, x28]\n" "fmla v2.4s, v28.4s, v8.4s\n" "ldr q27, [x17, x13]\n" "fmla v1.4s, v17.4s, v12.4s\n" @@ -356,7 +356,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v2.4s, v17.4s, v10.4s\n" "ldr q20, [x19, x28]\n" "fmla v4.4s, v18.4s, v8.4s\n" - "ldr q17, [x18, x13]\n" + "ldr q17, [x9, x13]\n" "fmla v1.4s, v18.4s, v14.4s\n" "ldr q18, [x17, x14]\n" "fmla v3.4s, v25.4s, v12.4s\n" @@ -370,7 +370,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v2.4s, v23.4s, v7.4s\n" "fmla v1.4s, v23.4s, v9.4s\n" "fmla v5.4s, v21.4s, v10.4s\n" - "ldr q21, [x18, x14]\n" + "ldr q21, [x9, x14]\n" "fmla v4.4s, v29.4s, v7.4s\n" "ldr q23, [x20, x13]\n" "str q2, [x22]\n" @@ -382,7 +382,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v1.4s, v29.4s, v13.4s\n" "ldr q25, [x20, x14]\n" "fmla v3.4s, v27.4s, v11.4s\n" - "add x18, x18, #16\n" + "add x9, x9, #16\n" "fmla v5.4s, v28.4s, v7.4s\n" "add x19, x19, #16\n" "fmla v1.4s, v26.4s, v8.4s\n" @@ -447,7 +447,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v15.4s, v19.4s, v6.4s\n" "add %[wbptr], %[wbptr], #40\n" "fmla v17.4s, v28.4s, v12.4s\n" - "ldr s25, [x18]\n" + "ldr s25, [x9]\n" "fmla v16.4s, v30.4s, v12.4s\n" "ldr s24, [x17, %[input_col_stride1]]\n" "fmla v15.4s, v21.4s, v12.4s\n" @@ -465,7 +465,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v15.4s, v23.4s, v9.4s\n" "ldr s30, [x19]\n" "fmla v17.4s, v20.4s, v11.4s\n" - "ldr s29, [x18, %[input_col_stride1]]\n" + "ldr s29, [x9, %[input_col_stride1]]\n" "fmla v16.4s, v24.4s, v11.4s\n" "ldr s28, [x17, x26]\n" "fmla v4.4s, v23.4s, v6.4s\n" @@ -483,7 +483,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v16.4s, v29.4s, v8.4s\n" "ldr s22, [x19, %[input_col_stride1]]\n" "fmla v17.4s, v21.4s, v10.4s\n" - "ldr s26, [x18, x26]\n" + "ldr s26, [x9, x26]\n" "fmla v2.4s, v29.4s, v14.4s\n" "ldr s20, [x17, x27]\n" "fmla v16.4s, v28.4s, v10.4s\n" @@ -501,7 +501,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v2.4s, v18.4s, v9.4s\n" "ldr s17, [x19, x26]\n" "fmla v5.4s, v19.4s, v14.4s\n" - "ldr s18, [x18, x27]\n" + "ldr s18, [x9, x27]\n" "fmla v16.4s, v26.4s, v7.4s\n" "ldr s25, [x17, x28]\n" "fmla v2.4s, v22.4s, v11.4s\n" @@ -523,7 +523,7 @@ void Conv::execute_tile<ActivationFunction::None>( "str s15, [%[outptr0], %[output_col_stride1]]\n" "fmla v2.4s, v28.4s, v8.4s\n" "fmla v5.4s, v30.4s, v11.4s\n" - "ldr s29, [x18, x28]\n" + "ldr s29, [x9, x28]\n" "fmla v1.4s, v17.4s, v12.4s\n" "ldr s27, [x17, x13]\n" "fmla v2.4s, v17.4s, v10.4s\n" @@ -533,7 +533,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v4.4s, v18.4s, v8.4s\n" "ldr s20, [x19, x28]\n" "fmla v1.4s, v18.4s, v14.4s\n" - "ldr s17, [x18, x13]\n" + "ldr s17, [x9, x13]\n" "fmla v3.4s, v25.4s, v12.4s\n" "ldr s18, [x17, x14]\n" "fmla v4.4s, v25.4s, v10.4s\n" @@ -545,7 +545,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v2.4s, v23.4s, v7.4s\n" "add x17, x17, #4\n" "fmla v5.4s, v21.4s, v10.4s\n" - "ldr s21, [x18, x14]\n" + "ldr s21, [x9, x14]\n" "fmla v1.4s, v23.4s, v9.4s\n" "ldr s23, [x20, x13]\n" "str s2, [x22]\n" @@ -589,7 +589,7 @@ void Conv::execute_tile<ActivationFunction::None>( "str s3, [x21, x23]\n" "mov v16.16b, v27.16b\n" "mov v15.16b, v27.16b\n" - "add x18, x18, #4\n" + "add x9, x9, #4\n" "fmla v0.4s, v21.4s, v13.4s\n" "ldr s11, [%[wbptr], #20]\n" "mov v2.16b, v27.16b\n" @@ -620,7 +620,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v15.4s, v19.4s, v6.4s\n" "add %[wbptr], %[wbptr], #40\n" "fmla v17.4s, v28.4s, v12.4s\n" - "ldr s25, [x18]\n" + "ldr s25, [x9]\n" "fmla v16.4s, v30.4s, v12.4s\n" "ldr s24, [x17, %[input_col_stride1]]\n" "fmla v15.4s, v21.4s, v12.4s\n" @@ -638,7 +638,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v15.4s, v23.4s, v9.4s\n" "fmla v4.4s, v23.4s, v6.4s\n" "fmla v17.4s, v20.4s, v11.4s\n" - "ldr s29, [x18, %[input_col_stride1]]\n" + "ldr s29, [x9, %[input_col_stride1]]\n" "fmla v16.4s, v24.4s, v11.4s\n" "ldr s28, [x17, x26]\n" "fmla v15.4s, v26.4s, v11.4s\n" @@ -656,7 +656,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v16.4s, v29.4s, v8.4s\n" "fmla v4.4s, v28.4s, v12.4s\n" "fmla v17.4s, v21.4s, v10.4s\n" - "ldr s26, [x18, x26]\n" + "ldr s26, [x9, x26]\n" "fmla v2.4s, v29.4s, v14.4s\n" "ldr s20, [x17, x27]\n" "fmla v16.4s, v28.4s, v10.4s\n" @@ -674,7 +674,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v5.4s, v19.4s, v14.4s\n" "ldr s17, [x19, x26]\n" "fmla v2.4s, v22.4s, v11.4s\n" - "ldr s18, [x18, x27]\n" + "ldr s18, [x9, x27]\n" "fmla v16.4s, v26.4s, v7.4s\n" "ldr s25, [x17, x28]\n" "fmla v4.4s, v26.4s, v9.4s\n" @@ -694,7 +694,7 @@ void Conv::execute_tile<ActivationFunction::None>( "str s15, [%[outptr0], %[output_col_stride1]]\n" "fmla v3.4s, v27.4s, v6.4s\n" "fmla v5.4s, v30.4s, v11.4s\n" - "ldr s29, [x18, x28]\n" + "ldr s29, [x9, x28]\n" "fmla v2.4s, v28.4s, v8.4s\n" "ldr s27, [x17, x13]\n" "fmla v1.4s, v17.4s, v12.4s\n" @@ -704,7 +704,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v2.4s, v17.4s, v10.4s\n" "ldr s20, [x19, x28]\n" "fmla v4.4s, v18.4s, v8.4s\n" - "ldr s17, [x18, x13]\n" + "ldr s17, [x9, x13]\n" "fmla v1.4s, v18.4s, v14.4s\n" "ldr s18, [x17, x14]\n" "fmla v3.4s, v25.4s, v12.4s\n" @@ -718,7 +718,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v2.4s, v23.4s, v7.4s\n" "fmla v1.4s, v23.4s, v9.4s\n" "fmla v5.4s, v21.4s, v10.4s\n" - "ldr s21, [x18, x14]\n" + "ldr s21, [x9, x14]\n" "fmla v4.4s, v29.4s, v7.4s\n" "ldr s23, [x20, x13]\n" "str s2, [x22]\n" @@ -730,7 +730,7 @@ void Conv::execute_tile<ActivationFunction::None>( "fmla v1.4s, v29.4s, v13.4s\n" "ldr s25, [x20, x14]\n" "fmla v3.4s, v27.4s, v11.4s\n" - "add x18, x18, #4\n" + "add x9, x9, #4\n" "fmla v5.4s, v28.4s, v7.4s\n" "add x19, x19, #4\n" "fmla v1.4s, v26.4s, v8.4s\n" @@ -759,7 +759,7 @@ void Conv::execute_tile<ActivationFunction::None>( "7:\n" : [wbptr] "+r" (weight_bias_ptr), [inptr0] "+r" (input), [outptr0] "+r" (output) : [n_channels] "r" ((long long) n_channels), [output_col_stride1] "r" (output_col_stride * sizeof(float)), [input_row_stride] "r" (input_row_stride * sizeof(float)), [input_col_stride1] "r" (input_col_stride * sizeof(float)), [output_row_stride] "r" (output_row_stride * sizeof(float)) - : "cc", "v0", "v1", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v2", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v3", "v30", "v4", "v5", "v6", "v7", "v8", "v9", "x15", "x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "x13", "x14", "memory" + : "cc", "v0", "v1", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v2", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v3", "v30", "v4", "v5", "v6", "v7", "v8", "v9", "x15", "x16", "x17", "x9", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "x13", "x14", "memory" ); } #endif // __aarch64__ |