Rewrite ConvPerChannelD1OD24_5x5
This new approach utilizes the aconv instruction to perform full matmul.
Creative data shuffling methods allow the implementation to achieve
significantly better useful MAC throughput.
Activation min/max is assumed to be int8 limits to eliminate the clamping
instructions in the postprocessing
ambient_face_2.8.0_504x378_grayscale_per_axis_simulator_iss is used in
testing.
The following items are measured on the simulator:
- Mismatch_count is 0.
- #0 CONV_2D takes 1.52M ticks (down from 4.02M, 2.64x speedup).
- Total cycles is now 52.9M (down from 57.9M, -8.6%).
The implementation currently contains lots of repeated and near-repeated
sections. They can be folded together in the future to improve
readability.
The following improvement items may be doable:
- Further tweak data shuffling sections to take less time
- Rearrange instructions on bottlenecks
Change-Id: Ibfcea94988b9ece4ea98a61c139ce44e967b4fd8
diff --git a/tflm/opt/conv_s8.cc b/tflm/opt/conv_s8.cc
index 7842b0d..10e8f41 100644
--- a/tflm/opt/conv_s8.cc
+++ b/tflm/opt/conv_s8.cc
@@ -184,13 +184,13 @@
const auto dilation_height_factor = params.dilation_height_factor;
const auto pad_width = params.padding_values.width;
const auto pad_height = params.padding_values.height;
- const auto input_batch = input_shape.Dims(0);
const auto input_height = input_shape.Dims(1);
const auto input_width = input_shape.Dims(2);
const auto input_depth = input_shape.Dims(3);
const auto filter_height = filter_shape.Dims(1);
const auto filter_width = filter_shape.Dims(2);
const auto filter_depth = filter_shape.Dims(3);
+ const auto output_height = output_shape.Dims(1);
const auto output_width = output_shape.Dims(2);
const auto output_depth = output_shape.Dims(3);
@@ -207,8 +207,7 @@
stride_width == 1 && dilation_height_factor == 1 &&
dilation_width_factor == 1 && pad_height == 0 && pad_width == 0 &&
(input_depth == filter_depth)) {
-
- if ((input_depth == 32) && (input_batch * input_height * input_width) >= 4) {
+ if ((input_depth == 32) && (batches * input_height * input_width) >= 4) {
RUN_KERNEL(kelvin::opt::ConvS8K1x1D32);
}
@@ -227,8 +226,14 @@
}
if (input_depth == 1 && filter_width == 5 && filter_height == 5 &&
- stride_width == 2 &&
- output_depth == 24) {
+ stride_width == 2 && dilation_width_factor == 1 && output_depth == 24 &&
+ params.quantized_activation_min == -128 &&
+ params.quantized_activation_max == 127 &&
+ // Special limitations of the implementation.
+ pad_height < 5 && pad_width < 5 &&
+ (output_height - 1) * stride_height - pad_height < input_height &&
+ (output_width - 1) * stride_width - pad_width < input_width &&
+ input_width >= 19) {
RUN_KERNEL(kelvin::opt::ConvPerChannelD1OD24_5x5);
}
diff --git a/tflm/opt/conv_s8_d1.cc b/tflm/opt/conv_s8_d1.cc
index c03a7f5..35b90a6 100644
--- a/tflm/opt/conv_s8_d1.cc
+++ b/tflm/opt/conv_s8_d1.cc
@@ -65,394 +65,109 @@
}
}
+// Internal helper function within ConvPerChannelD1OD24_5x5.
+__attribute__((always_inline)) inline void
+ConvPerChannelD1OD24_5x5_inputshuffle() {
+ // IN: v48-v52 (modified)
+ // OUT: v0-v7
+ // CLOBBER: v22-v23, v32-v33
+
+ // Zips row0123 together into v48-v51.
+ vzip_b_vv(v22, v48, v50); // Also writes v23.
+ vzip_b_vv(v32, v49, v51); // Also writes v33.
+ vzip_b_vv(v48, v22, v32); // Also writes v49.
+ vzip_b_vv(v50, v23, v33); // Also writes v51 but it's unused.
+
+ vsliden_h_1_vv(v33, v52, v52);
+ vslidep_w_3_vv(v22, v48, v48);
+ vslidep_w_1_vv(v23, v48, v48);
+ // Patch 0 is row0123[0:19] with row4[0:4].
+ vsliden_w_3_vv(v0, v22, v52); // RHS not yet rotated.
+ // Patch 1 is row0123[8:27] with row4[2:6].
+ vsliden_w_3_vv(v1, v23, v33);
+
+ vsliden_h_2_vv(v32, v52, v52);
+ vsliden_h_3_vv(v33, v52, v52);
+ vsliden_w_1_vv(v22, v48, v49);
+ vsliden_w_3_vv(v23, v48, v49);
+ // Patch 2 is row0123[16:35] with row4[4:8].
+ vsliden_w_3_vv(v2, v22, v32);
+ // Patch 3 is row0123[24:43] with row4[6:10].
+ vsliden_w_3_vv(v3, v23, v33);
+
+ vsliden_h_3_vv(v33, v32, v32);
+ vsliden_h_4_vv(v32, v52, v52);
+ vslidep_w_3_vv(v22, v49, v49);
+ vslidep_w_1_vv(v23, v49, v49);
+ // Patch 4 is row0123[32:51] with row4[8:12].
+ vsliden_w_3_vv(v4, v22, v32);
+ // Patch 5 is row0123[40:59] with row4[10:14].
+ vsliden_w_3_vv(v5, v23, v33);
+
+ vsliden_h_3_vv(v33, v32, v32);
+ vsliden_w_3_vv(v32, v52, v52);
+ vsliden_w_1_vv(v22, v49, v50);
+ vsliden_w_3_vv(v23, v49, v50);
+ // Patch 6 is row0123[48:67] with row4[12:16].
+ vsliden_w_3_vv(v6, v22, v32);
+ // Patch 7 is row0123[56:75] with row4[14:18].
+ vsliden_w_3_vv(v7, v23, v33);
+}
+
+// Internal helper function within ConvPerChannelD1OD24_5x5.
+__attribute__((always_inline)) inline void ConvPerChannelD1OD24_5x5_postproc(
+ const int32_t* output_multiplier, const int32_t* output_shift,
+ int32_t output_offset, int32_t output_activation_min,
+ int32_t output_activation_max, int8_t* out_ptr_col0, int8_t* out_ptr_col4) {
+ // IN: acc and params
+ // OUT: memory, see out_ptr_*
+ // CLOBBER: v22-v23, v32-v33, v48-v55
+
+ // Retrieves results.
+ vcget(v48); // v48-v55 is written.
+
+ // Postprocessing and output.
+ vevnodd_w_vv(v22, v48, v52); // Also writes v23.
+ vevnodd_w_vv(v32, v49, v53); // Also writes v33.
+ vdmulh_w_rn_vx(v22, v22, output_multiplier[0]);
+ vdmulh_w_rn_vx(v23, v23, output_multiplier[4]);
+ vdmulh_w_rn_vx(v32, v32, output_multiplier[2]);
+ vdmulh_w_rn_vx(v33, v33, output_multiplier[6]);
+ vsha_w_r_vx(v22, v22, -output_shift[0]);
+ vsha_w_r_vx(v23, v23, -output_shift[4]);
+ vsha_w_r_vx(v32, v32, -output_shift[2]);
+ vsha_w_r_vx(v33, v33, -output_shift[6]);
+ vzip_w_vv(v22, v22, v23); // Also writes v23.
+ vzip_w_vv(v32, v32, v33); // Also writes v33.
+ vmvp_vv(v48, v22, v32); // Also writes v49.
+ vmvp_vv(v52, v23, v33); // Also writes v53.
+
+ vevnodd_w_vv(v22, v50, v54); // Also writes v23.
+ vevnodd_w_vv(v32, v51, v55); // Also writes v33.
+ vdmulh_w_rn_vx(v22, v22, output_multiplier[1]);
+ vdmulh_w_rn_vx(v23, v23, output_multiplier[5]);
+ vdmulh_w_rn_vx(v32, v32, output_multiplier[3]);
+ vdmulh_w_rn_vx(v33, v33, output_multiplier[7]);
+ vsha_w_r_vx(v22, v22, -output_shift[1]);
+ vsha_w_r_vx(v23, v23, -output_shift[5]);
+ vsha_w_r_vx(v32, v32, -output_shift[3]);
+ vsha_w_r_vx(v33, v33, -output_shift[7]);
+ vzip_w_vv(v22, v22, v23); // Also writes v23.
+ vzip_w_vv(v32, v32, v33); // Also writes v33.
+ vmvp_vv(v50, v22, v32); // Also writes v51.
+ vmvp_vv(v54, v23, v33); // Also writes v55.
+
+ vadd_w_vx_m(v48, v48, output_offset);
+ vadd_w_vx_m(v52, v52, output_offset);
+
+ vsraqs_b_vx(v48, v48, 0);
+ vsraqs_b_vx(v52, v52, 0);
+ vstq_b_s_xx(v48, out_ptr_col0, /*output_depth=*/24);
+ vstq_b_s_xx(v52, out_ptr_col4, /*output_depth=*/24);
+}
+
} // namespace
-#define FLT_0_0 v0
-#define FLT_0_1 v3
-#define FLT_0_2 v6
-#define FLT_0_3 v9
-#define FLT_0_4 v12
-
-#define FLT_1_0 v1
-#define FLT_1_1 v4
-#define FLT_1_2 v7
-#define FLT_1_3 v10
-#define FLT_1_4 v13
-
-#define FLT_2_0 v2
-#define FLT_2_1 v5
-#define FLT_2_2 v8
-#define FLT_2_3 v11
-#define FLT_2_4 v14
-
-#define FLT_3_0 v15
-#define FLT_3_1 v16
-#define FLT_3_2 v17
-#define FLT_3_3 v18
-#define FLT_3_4 v19
-
-#define FLT_HOLE v20
-#define FLT_4_0 v21
-#define FLT_4_1 v22
-#define FLT_4_2 v23
-#define FLT_4_3 v24
-#define FLT_4_4 v25
-
-#define INPUT_0_0 v26
-#define INPUT_0_1 v29
-#define INPUT_0_2 v32
-#define INPUT_0_3 v35
-#define INPUT_0_4 v38
-
-#define INPUT_1_0 v27
-#define INPUT_1_1 v30
-#define INPUT_1_2 v33
-#define INPUT_1_3 v36
-#define INPUT_1_4 v39
-
-#define INPUT_2_0 v28
-#define INPUT_2_1 v31
-#define INPUT_2_2 v34
-#define INPUT_2_3 v37
-#define INPUT_2_4 v40
-
-#define INPUT_3_0 v41
-#define INPUT_3_1 v42
-#define INPUT_3_2 v43
-#define INPUT_3_3 v44
-#define INPUT_3_4 v45
-
-#define INPUT_4_0 v46
-#define INPUT_4_1 v47
-#define INPUT_4_2 v48
-#define INPUT_4_3 v49
-#define INPUT_4_4 v50
-
-#define CALCULATE_IN_X(in_x_origin) \
- { \
- _Pragma("GCC unroll 5") for (int i = 0; i < 5; ++i) { \
- in_x[i] = in_x_origin + (dilation_width_factor * i); \
- } \
- }
-
-#define CALCULATE_IN_Y(in_y_origin) \
- { \
- _Pragma("GCC unroll 5") for (int i = 0; i < 5; ++i) { \
- in_y[i] = in_y_origin + (dilation_height_factor * i); \
- } \
- }
-
-#define PAD_ROW_0(input_offset) \
- { \
- vdup_b_x(INPUT_0_0, -input_offset); \
- vdup_b_x(INPUT_0_1, -input_offset); \
- vdup_b_x(INPUT_0_2, -input_offset); \
- vdup_b_x(INPUT_0_3, -input_offset); \
- vdup_b_x(INPUT_0_4, -input_offset); \
- }
-#define PAD_ROW_1(input_offset) \
- { \
- vdup_b_x(INPUT_1_0, -input_offset); \
- vdup_b_x(INPUT_1_1, -input_offset); \
- vdup_b_x(INPUT_1_2, -input_offset); \
- vdup_b_x(INPUT_1_3, -input_offset); \
- vdup_b_x(INPUT_1_4, -input_offset); \
- }
-#define PAD_ROW_2(input_offset) \
- { \
- vdup_b_x(INPUT_2_0, -input_offset); \
- vdup_b_x(INPUT_2_1, -input_offset); \
- vdup_b_x(INPUT_2_2, -input_offset); \
- vdup_b_x(INPUT_2_3, -input_offset); \
- vdup_b_x(INPUT_2_4, -input_offset); \
- }
-#define PAD_ROW_3(input_offset) \
- { \
- vdup_b_x(INPUT_3_0, -input_offset); \
- vdup_b_x(INPUT_3_1, -input_offset); \
- vdup_b_x(INPUT_3_2, -input_offset); \
- vdup_b_x(INPUT_3_3, -input_offset); \
- vdup_b_x(INPUT_3_4, -input_offset); \
- }
-#define PAD_ROW_4(input_offset) \
- { \
- vdup_b_x(INPUT_4_0, -input_offset); \
- vdup_b_x(INPUT_4_1, -input_offset); \
- vdup_b_x(INPUT_4_2, -input_offset); \
- vdup_b_x(INPUT_4_3, -input_offset); \
- vdup_b_x(INPUT_4_4, -input_offset); \
- }
-
-#define LOAD_ROW_0(p_input, input_width, in_y, in_x) \
- { \
- const int8_t* p_row = p_input + (in_y[0] * input_width); \
- vdup_b_x(INPUT_0_0, *(p_row + in_x[0])); \
- vdup_b_x(INPUT_0_1, *(p_row + in_x[1])); \
- vdup_b_x(INPUT_0_2, *(p_row + in_x[2])); \
- vdup_b_x(INPUT_0_3, *(p_row + in_x[3])); \
- vdup_b_x(INPUT_0_4, *(p_row + in_x[4])); \
- }
-
-#define LOAD_ROW_1(p_input, input_width, in_y, in_x) \
- { \
- const int8_t* p_row = p_input + (in_y[1] * input_width); \
- vdup_b_x(INPUT_1_0, *(p_row + in_x[0])); \
- vdup_b_x(INPUT_1_1, *(p_row + in_x[1])); \
- vdup_b_x(INPUT_1_2, *(p_row + in_x[2])); \
- vdup_b_x(INPUT_1_3, *(p_row + in_x[3])); \
- vdup_b_x(INPUT_1_4, *(p_row + in_x[4])); \
- }
-
-#define LOAD_ROW_2(p_input, input_width, in_y, in_x) \
- { \
- const int8_t* p_row = p_input + (in_y[2] * input_width); \
- vdup_b_x(INPUT_2_0, *(p_row + in_x[0])); \
- vdup_b_x(INPUT_2_1, *(p_row + in_x[1])); \
- vdup_b_x(INPUT_2_2, *(p_row + in_x[2])); \
- vdup_b_x(INPUT_2_3, *(p_row + in_x[3])); \
- vdup_b_x(INPUT_2_4, *(p_row + in_x[4])); \
- }
-
-#define LOAD_ROW_3(p_input, input_width, in_y, in_x) \
- { \
- const int8_t* p_row = p_input + (in_y[3] * input_width); \
- vdup_b_x(INPUT_3_0, *(p_row + in_x[0])); \
- vdup_b_x(INPUT_3_1, *(p_row + in_x[1])); \
- vdup_b_x(INPUT_3_2, *(p_row + in_x[2])); \
- vdup_b_x(INPUT_3_3, *(p_row + in_x[3])); \
- vdup_b_x(INPUT_3_4, *(p_row + in_x[4])); \
- }
-
-#define LOAD_ROW_4(p_input, input_width, in_y, in_x) \
- { \
- const int8_t* p_row = p_input + (in_y[4] * input_width); \
- vdup_b_x(INPUT_4_0, *(p_row + in_x[0])); \
- vdup_b_x(INPUT_4_1, *(p_row + in_x[1])); \
- vdup_b_x(INPUT_4_2, *(p_row + in_x[2])); \
- vdup_b_x(INPUT_4_3, *(p_row + in_x[3])); \
- vdup_b_x(INPUT_4_4, *(p_row + in_x[4])); \
- }
-
-#define H_PAD_OR_LOAD_ROW_0(p_input, input_width, input_offset, in_y, in_x) \
- if (in_x[0] >= 0 && in_x[4] < input_width) { \
- LOAD_ROW_0(p_input, input_width, in_y, in_x); \
- } else { \
- const int8_t* p_row = p_input + (in_y[0] * input_width); \
- if (in_x[0] < 0 || in_x[0] >= input_width) { \
- vdup_b_x(INPUT_0_0, -input_offset); \
- } else { \
- vdup_b_x(INPUT_0_0, *(p_row + in_x[0])); \
- } \
- if (in_x[1] < 0 || in_x[1] >= input_width) { \
- vdup_b_x(INPUT_0_1, -input_offset); \
- } else { \
- vdup_b_x(INPUT_0_1, *(p_row + in_x[1])); \
- } \
- if (in_x[2] < 0 || in_x[2] >= input_width) { \
- vdup_b_x(INPUT_0_2, -input_offset); \
- } else { \
- vdup_b_x(INPUT_0_2, *(p_row + in_x[2])); \
- } \
- if (in_x[3] < 0 || in_x[3] >= input_width) { \
- vdup_b_x(INPUT_0_3, -input_offset); \
- } else { \
- vdup_b_x(INPUT_0_3, *(p_row + in_x[3])); \
- } \
- if (in_x[4] < 0 || in_x[4] >= input_width) { \
- vdup_b_x(INPUT_0_4, -input_offset); \
- } else { \
- vdup_b_x(INPUT_0_4, *(p_row + in_x[4])); \
- } \
- }
-
-#define H_PAD_OR_LOAD_ROW_1(p_input, input_width, input_offset, in_y, in_x) \
- if (in_x[0] >= 0 && in_x[4] < input_width) { \
- LOAD_ROW_1(p_input, input_width, in_y, in_x); \
- } else { \
- const int8_t* p_row = p_input + (in_y[1] * input_width); \
- if (in_x[0] < 0 || in_x[0] >= input_width) { \
- vdup_b_x(INPUT_1_0, -input_offset); \
- } else { \
- vdup_b_x(INPUT_1_0, *(p_row + in_x[0])); \
- } \
- if (in_x[1] < 0 || in_x[1] >= input_width) { \
- vdup_b_x(INPUT_1_1, -input_offset); \
- } else { \
- vdup_b_x(INPUT_1_1, *(p_row + in_x[1])); \
- } \
- if (in_x[2] < 0 || in_x[2] >= input_width) { \
- vdup_b_x(INPUT_1_2, -input_offset); \
- } else { \
- vdup_b_x(INPUT_1_2, *(p_row + in_x[2])); \
- } \
- if (in_x[3] < 0 || in_x[3] >= input_width) { \
- vdup_b_x(INPUT_1_3, -input_offset); \
- } else { \
- vdup_b_x(INPUT_1_3, *(p_row + in_x[3])); \
- } \
- if (in_x[4] < 0 || in_x[4] >= input_width) { \
- vdup_b_x(INPUT_1_4, -input_offset); \
- } else { \
- vdup_b_x(INPUT_1_4, *(p_row + in_x[4])); \
- } \
- }
-
-#define H_PAD_OR_LOAD_ROW_2(p_input, input_width, input_offset, in_y, in_x) \
- if (in_x[0] >= 0 && in_x[4] < input_width) { \
- LOAD_ROW_2(p_input, input_width, in_y, in_x); \
- } else { \
- const int8_t* p_row = p_input + (in_y[2] * input_width); \
- if (in_x[0] < 0 || in_x[0] >= input_width) { \
- vdup_b_x(INPUT_2_0, -input_offset); \
- } else { \
- vdup_b_x(INPUT_2_0, *(p_row + in_x[0])); \
- } \
- if (in_x[1] < 0 || in_x[1] >= input_width) { \
- vdup_b_x(INPUT_2_1, -input_offset); \
- } else { \
- vdup_b_x(INPUT_2_1, *(p_row + in_x[1])); \
- } \
- if (in_x[2] < 0 || in_x[2] >= input_width) { \
- vdup_b_x(INPUT_2_2, -input_offset); \
- } else { \
- vdup_b_x(INPUT_2_2, *(p_row + in_x[2])); \
- } \
- if (in_x[3] < 0 || in_x[3] >= input_width) { \
- vdup_b_x(INPUT_2_3, -input_offset); \
- } else { \
- vdup_b_x(INPUT_2_3, *(p_row + in_x[3])); \
- } \
- if (in_x[4] < 0 || in_x[4] >= input_width) { \
- vdup_b_x(INPUT_2_4, -input_offset); \
- } else { \
- vdup_b_x(INPUT_2_4, *(p_row + in_x[4])); \
- } \
- }
-
-#define H_PAD_OR_LOAD_ROW_3(p_input, input_width, input_offset, in_y, in_x) \
- if (in_x[0] >= 0 && in_x[4] < input_width) { \
- LOAD_ROW_3(p_input, input_width, in_y, in_x); \
- } else { \
- const int8_t* p_row = p_input + (in_y[3] * input_width); \
- if (in_x[0] < 0 || in_x[0] >= input_width) { \
- vdup_b_x(INPUT_3_0, -input_offset); \
- } else { \
- vdup_b_x(INPUT_3_0, *(p_row + in_x[0])); \
- } \
- if (in_x[1] < 0 || in_x[1] >= input_width) { \
- vdup_b_x(INPUT_3_1, -input_offset); \
- } else { \
- vdup_b_x(INPUT_3_1, *(p_row + in_x[1])); \
- } \
- if (in_x[2] < 0 || in_x[2] >= input_width) { \
- vdup_b_x(INPUT_3_2, -input_offset); \
- } else { \
- vdup_b_x(INPUT_3_2, *(p_row + in_x[2])); \
- } \
- if (in_x[3] < 0 || in_x[3] >= input_width) { \
- vdup_b_x(INPUT_3_3, -input_offset); \
- } else { \
- vdup_b_x(INPUT_3_3, *(p_row + in_x[3])); \
- } \
- if (in_x[4] < 0 || in_x[4] >= input_width) { \
- vdup_b_x(INPUT_3_4, -input_offset); \
- } else { \
- vdup_b_x(INPUT_3_4, *(p_row + in_x[4])); \
- } \
- }
-
-#define H_PAD_OR_LOAD_ROW_4(p_input, input_width, input_offset, in_y, in_x) \
- if (in_x[0] >= 0 && in_x[4] < input_width) { \
- LOAD_ROW_4(p_input, input_width, in_y, in_x); \
- } else { \
- const int8_t* p_row = p_input + (in_y[4] * input_width); \
- if (in_x[0] < 0 || in_x[0] >= input_width) { \
- vdup_b_x(INPUT_4_0, -input_offset); \
- } else { \
- vdup_b_x(INPUT_4_0, *(p_row + in_x[0])); \
- } \
- if (in_x[1] < 0 || in_x[1] >= input_width) { \
- vdup_b_x(INPUT_4_1, -input_offset); \
- } else { \
- vdup_b_x(INPUT_4_1, *(p_row + in_x[1])); \
- } \
- if (in_x[2] < 0 || in_x[2] >= input_width) { \
- vdup_b_x(INPUT_4_2, -input_offset); \
- } else { \
- vdup_b_x(INPUT_4_2, *(p_row + in_x[2])); \
- } \
- if (in_x[3] < 0 || in_x[3] >= input_width) { \
- vdup_b_x(INPUT_4_3, -input_offset); \
- } else { \
- vdup_b_x(INPUT_4_3, *(p_row + in_x[3])); \
- } \
- if (in_x[4] < 0 || in_x[4] >= input_width) { \
- vdup_b_x(INPUT_4_4, -input_offset); \
- } else { \
- vdup_b_x(INPUT_4_4, *(p_row + in_x[4])); \
- } \
- }
-
-#define _H_PAD_OR_LOAD_ROW(row, p_input, input_width, input_offset, in_y, \
- in_x) \
- H_PAD_OR_LOAD_ROW_##row(p_input, input_width, input_offset, in_y, in_x);
-
-#define _PAD_OR_LOAD_ROW(row, p_input, input_height, input_width, in_y, in_x, \
- input_offset) \
- { \
- if (in_y[row] < 0 || in_y[row] >= input_height) { \
- PAD_ROW_##row(input_offset); \
- } else { \
- _H_PAD_OR_LOAD_ROW(row, p_input, input_width, input_offset, in_y, in_x); \
- } \
- }
-
-#define PAD_OR_LOAD_ROW_0(p_input, input_height, input_width, in_y, in_x, \
- input_offset) \
- _PAD_OR_LOAD_ROW(0, p_input, input_height, input_width, in_y, in_x, \
- input_offset);
-#define PAD_OR_LOAD_ROW_1(p_input, input_height, input_width, in_y, in_x, \
- input_offset) \
- _PAD_OR_LOAD_ROW(1, p_input, input_height, input_width, in_y, in_x, \
- input_offset);
-#define PAD_OR_LOAD_ROW_2(p_input, input_height, input_width, in_y, in_x, \
- input_offset) \
- _PAD_OR_LOAD_ROW(2, p_input, input_height, input_width, in_y, in_x, \
- input_offset);
-#define PAD_OR_LOAD_ROW_3(p_input, input_height, input_width, in_y, in_x, \
- input_offset) \
- _PAD_OR_LOAD_ROW(3, p_input, input_height, input_width, in_y, in_x, \
- input_offset);
-#define PAD_OR_LOAD_ROW_4(p_input, input_height, input_width, in_y, in_x, \
- input_offset) \
- _PAD_OR_LOAD_ROW(4, p_input, input_height, input_width, in_y, in_x, \
- input_offset);
-
-#define COMPUTE(cmds, swizzled_bias_data) \
- { \
- vld_w_x_m(v60, swizzled_bias_data); \
- adwinit_v(v60, v60); \
- adwconv_vxv(v60, INPUT_0_0, cmds, FLT_0_0); \
- adwconv_vxv(v60, INPUT_0_1, cmds, FLT_0_1); \
- adwconv_vxv(v60, INPUT_0_2, cmds, FLT_0_2); \
- adwconv_vxv(v60, INPUT_0_3, cmds, FLT_0_3); \
- adwconv_vxv(v60, INPUT_0_4, cmds, FLT_0_4); \
- adwconv_vxv(v60, INPUT_3_0, cmds, FLT_3_0); \
- adwconv_vxv(v60, INPUT_3_3, cmds, FLT_3_3); \
- adwconv_vxv(v60, INPUT_3_4, cmds, FLT_HOLE); \
- vdwconv_vxv(v60, INPUT_4_2, cmds, FLT_4_2); \
- }
-
-#define OUTPUT(output_activation_min, output_activation_max, output_offset, \
- local_output_data, n_channels) \
- { \
- INT32_TO_INT8_OUTPUT_PIPELINE_INPLACE( \
- v60, v52, v56, output_activation_min, output_activation_max, \
- output_offset); \
- vsraqs_b_vx(v60, v60, 0); \
- vst_b_l_xx(v60, local_output_data, n_channels); \
- }
-
// Estimated count of arithmetic ops: 58.297 M ops, equivalently 29.148 M MACs
void ConvPerChannelD1OD24_5x5(
const tflite::ConvParams& params, const int32_t* output_multiplier,
@@ -463,27 +178,33 @@
int8_t* output_data) {
// Get parameters.
const int32_t input_offset = params.input_offset; // r = s(q - Z)
- const int stride_width = params.stride_width;
+ constexpr int stride_width = 2;
+ TFLITE_DCHECK_EQ(params.stride_width, stride_width);
const int stride_height = params.stride_height;
- const int dilation_width_factor = params.dilation_width_factor;
+ constexpr int dilation_width_factor = 1;
+ TFLITE_DCHECK_EQ(params.dilation_width_factor, dilation_width_factor);
const int dilation_height_factor = params.dilation_height_factor;
const int pad_width = params.padding_values.width;
const int pad_height = params.padding_values.height;
const int32_t output_offset = params.output_offset;
// Set min and max value of the output.
- const int32_t output_activation_min = params.quantized_activation_min;
- const int32_t output_activation_max = params.quantized_activation_max;
+ constexpr int32_t output_activation_min = -128;
+ TFLITE_DCHECK_EQ(params.quantized_activation_min, output_activation_min);
+ constexpr int32_t output_activation_max = 127;
+ TFLITE_DCHECK_EQ(params.quantized_activation_max, output_activation_max);
// Consistency check.
TFLITE_DCHECK_LE(output_activation_min, output_activation_max);
TFLITE_DCHECK_EQ(input_shape.DimensionsCount(), 4);
TFLITE_DCHECK_EQ(filter_shape.DimensionsCount(), 4);
TFLITE_DCHECK_EQ(output_shape.DimensionsCount(), 4);
- // const int batches = tflite::MatchingDim(input_shape, 0, output_shape, 0);
- const int input_depth = input_shape.Dims(3);
- const int output_depth =
- tflite::MatchingDim(filter_shape, 0, output_shape, 3);
+ const int batches = tflite::MatchingDim(input_shape, 0, output_shape, 0);
+ constexpr int input_depth = 1;
+ TFLITE_DCHECK_EQ(input_shape.Dims(3), input_depth);
+ constexpr const int output_depth = 24;
+ TFLITE_DCHECK_EQ(tflite::MatchingDim(filter_shape, 0, output_shape, 3),
+ output_depth);
if (bias_data) {
TFLITE_DCHECK_EQ(bias_shape.FlatSize(), output_depth);
}
@@ -491,472 +212,1242 @@
// Check dimensions of the tensors.
const int input_height = input_shape.Dims(1);
const int input_width = input_shape.Dims(2);
- const int filter_height = filter_shape.Dims(1);
- const int filter_width = filter_shape.Dims(2);
- const int filter_input_depth = filter_shape.Dims(3);
- const int groups = input_depth / filter_input_depth;
- TFLITE_DCHECK_NE(groups, 0);
- TFLITE_DCHECK_EQ(input_depth % filter_input_depth, 0);
- const int filters_per_group = output_depth / groups;
- TFLITE_DCHECK_NE(filters_per_group, 0);
+ constexpr int filter_height = 5;
+ TFLITE_DCHECK_EQ(filter_shape.Dims(1), filter_height);
+ constexpr int filter_width = 5;
+ TFLITE_DCHECK_EQ(filter_shape.Dims(2), filter_width);
+ // Input depth is 1 so filter input depth must be 1.
+ TFLITE_DCHECK_EQ(filter_shape.Dims(3), 1);
const int output_height = output_shape.Dims(1);
const int output_width = output_shape.Dims(2);
- // Scratch pads to juggle data
- const size_t swizzled_filter_data_size = 24 * filter_height * filter_width;
- std::unique_ptr<int8_t> swizzled_filter_data(reinterpret_cast<int8_t*>(
- ::aligned_alloc(32, swizzled_filter_data_size)));
- int32_t swizzled_bias_data[32];
- int32_t swizzled_output_multiplier[32];
- int32_t swizzled_output_shift[32];
- // Transpose filter for easy loading
- for (int filter_y = 0; filter_y < filter_height; ++filter_y) {
- for (int filter_x = 0; filter_x < filter_width; ++filter_x) {
- for (int i = 0; i < 24; i++) {
- int filter_location =
- (filter_y * filter_width * 24) + (filter_x * 24) + i;
- swizzled_filter_data.get()[filter_location] =
- filter_data[tflite::Offset(filter_shape, i, filter_y, filter_x, 0)];
- }
- }
- }
+ constexpr int patches_per_iteration = 8;
+ constexpr int load_width =
+ 1 + (patches_per_iteration /*=8*/ - 1) * stride_width /*=2*/ +
+ (filter_width /*=5*/ - 1) * dilation_width_factor /*=1*/;
- const int8_t* p_flt_0 = swizzled_filter_data.get() + (0 * filter_width * 24);
- const int8_t* p_flt_1 = swizzled_filter_data.get() + (1 * filter_width * 24);
- const int8_t* p_flt_2 = swizzled_filter_data.get() + (2 * filter_width * 24);
- const int8_t* p_flt_3 = swizzled_filter_data.get() + (3 * filter_width * 24);
- const int8_t* p_flt_4 = swizzled_filter_data.get() + (4 * filter_width * 24);
- vld_b_lp_xx(FLT_0_0, p_flt_0, 24);
- vld_b_lp_xx(FLT_0_1, p_flt_0, 24);
- vld_b_lp_xx(FLT_0_2, p_flt_0, 24);
- vld_b_lp_xx(FLT_0_3, p_flt_0, 24);
- vld_b_lp_xx(FLT_0_4, p_flt_0, 24);
+ // This optimized path requires all output pixels must read at least 1
+ // input pixel.
+ TFLITE_DCHECK_LT(pad_height, 5);
+ TFLITE_DCHECK_LT(pad_width, 5);
+ TFLITE_DCHECK_LT((output_height - 1) * stride_height - pad_height,
+ input_height);
+ TFLITE_DCHECK_LT((output_width - 1) * stride_width /*=2*/ - pad_width,
+ input_width);
+ // This optimized path is complex enough and we don't wish to handle a load
+ // with padding on both sides.
+ TFLITE_DCHECK_GE(input_width * input_depth /*=1*/, load_width);
- vld_b_lp_xx(FLT_1_0, p_flt_1, 24);
- vld_b_lp_xx(FLT_1_1, p_flt_1, 24);
- vld_b_lp_xx(FLT_1_2, p_flt_1, 24);
- vld_b_lp_xx(FLT_1_3, p_flt_1, 24);
- vld_b_lp_xx(FLT_1_4, p_flt_1, 24);
+ // Hopefully this lambda helps the compiler get as much done statically
+ // as possible.
+ auto make_aconv_cmd = [](int in_off) {
+ union {
+ vconv_u8_t cmd;
+ uint32_t raw;
+ } cmd;
+ vconv_u8_t conv_cmd = {
+ .mode = 0,
+ .start = 0,
+ .stop = 6, // We're doing [8,25]x[25,8] which is 7 ticks.
+ .sbias1 = in_off,
+ .sdata1 = true,
+ .sbias2 = 0,
+ .sdata2 = true,
+ };
+ cmd.cmd = conv_cmd;
+ return cmd.raw;
+ };
+ auto batch_start_offset = [](int b, int height, int width, int depth) {
+ return b * height * width * depth;
+ };
+ auto row_start_offset = [](int y, int width, int depth) {
+ return y * width * depth;
+ };
+ const uint32_t aconv_cmd = make_aconv_cmd(input_offset);
- vld_b_lp_xx(FLT_2_0, p_flt_2, 24);
- vld_b_lp_xx(FLT_2_1, p_flt_2, 24);
- vld_b_lp_xx(FLT_2_2, p_flt_2, 24);
- vld_b_lp_xx(FLT_2_3, p_flt_2, 24);
- vld_b_lp_xx(FLT_2_4, p_flt_2, 24);
-
- vld_b_lp_xx(FLT_3_0, p_flt_3, 24);
- vld_b_lp_xx(FLT_3_1, p_flt_3, 24);
- vld_b_lp_xx(FLT_3_2, p_flt_3, 24);
- vld_b_lp_xx(FLT_3_3, p_flt_3, 24);
- vld_b_lp_xx(FLT_3_4, p_flt_3, 24);
-
- vdup_b_x(FLT_HOLE, 0);
- vld_b_lp_xx(FLT_4_0, p_flt_4, 24);
- vld_b_lp_xx(FLT_4_1, p_flt_4, 24);
- vld_b_lp_xx(FLT_4_2, p_flt_4, 24);
- vld_b_lp_xx(FLT_4_3, p_flt_4, 24);
- vld_b_lp_xx(FLT_4_4, p_flt_4, 24);
-
- union {
- vdwconv_u8_t dwconv;
- uint32_t raw;
- } cmds;
- cmds.raw = 0;
- cmds.dwconv.sdata1 = true;
- cmds.dwconv.sbias1 = input_offset;
- cmds.dwconv.sdata2 = true;
- cmds.dwconv.sbias2 = 0;
- cmds.dwconv.mode = 0;
- cmds.dwconv.sparsity = 0;
- cmds.dwconv.regbase = 0;
- int out_channel = 0;
- int n_channels = 24;
-
- memset(swizzled_bias_data, 0, 32 * sizeof(uint32_t));
- JumptableSwizzle(bias_data + out_channel, swizzled_bias_data, n_channels);
- memset(swizzled_output_multiplier, 0, 32 * sizeof(uint32_t));
- JumptableSwizzle(output_multiplier + out_channel, swizzled_output_multiplier,
- n_channels);
- JumptableSwizzle(output_shift + out_channel, swizzled_output_shift,
- n_channels);
- vld_w_x_m(v52, swizzled_output_multiplier);
- vld_w_x_m(v56, swizzled_output_shift);
- vrsub_w_vx_m(v56, v56, 0);
-
- int8_t* local_output_data = output_data + out_channel;
- int in_y[5];
- int in_x[7];
int out_y = 0;
- const int8_t* p_input = input_data;
- // Handle top row padding
- for (; out_y < pad_height; ++out_y) {
- int out_x = 0;
- const int in_y_origin = (out_y * stride_height) - pad_height;
- CALCULATE_IN_Y(in_y_origin);
- // Left padding required
- for (; out_x < pad_width; ++out_x) {
- const int in_x_origin = (out_x * stride_width) - pad_width;
- CALCULATE_IN_X(in_x_origin);
- PAD_OR_LOAD_ROW_0(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_1(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_2(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_3(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_4(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- COMPUTE(cmds, swizzled_bias_data);
- OUTPUT(output_activation_min, output_activation_max, output_offset,
- local_output_data, n_channels);
- local_output_data += output_depth;
+ int in_y_min = -pad_height;
+ int in_y_max = in_y_min + (filter_height - 1) * dilation_height_factor;
+
+ // Reg Map:
+ // v0-v7 : input patches[0..7]
+ // v8-v15 : weight for OD[0..7]
+ // v16-v21: bias for OD[0..23], each reg is a int32x4 stored twice
+ // v22-v23: temp
+ // v24-v31: weight for OD[8..15]
+ // v32-v34: temp
+ // v35-v39: unused
+ // v40-v47: weight for OD[16..23]
+ // v48-v55: temp
+ // v56-v63: unused
+
+ // Temp reg usage:
+ // - input loading
+ // - v22-v23: additional loading buffer.
+ // - v32-v34: additional loading buffer.
+ // - v48-v52: input for use in next step.
+ // - v53: scratch pad.
+ // - input handling
+ // - v22-v23: scratch pad.
+ // - v32-v33: scratch pad.
+ // - v48-v52: ready-to-use input, with horizontal padding applied where
+ // needed.
+ // - aconv
+ // - v48-v55: prepare and set acc.
+ // - output
+ // - v22-v23: scratch pad.
+ // - v32-v33: scratch pad.
+ // - v48-v55: get acc and run postproc in place.
+
+ if (bias_data) {
+ int32_t od_param_buffer[48] __attribute__((aligned(32)));
+ // Preloads bias.
+ for (int i = 0; i < 6; i++) {
+ od_param_buffer[(8 * i) + 0] = bias_data[(4 * i) + 0];
+ od_param_buffer[(8 * i) + 1] = bias_data[(4 * i) + 2];
+ od_param_buffer[(8 * i) + 2] = bias_data[(4 * i) + 1];
+ od_param_buffer[(8 * i) + 3] = bias_data[(4 * i) + 3];
+ od_param_buffer[(8 * i) + 4] = bias_data[(4 * i) + 0];
+ od_param_buffer[(8 * i) + 5] = bias_data[(4 * i) + 2];
+ od_param_buffer[(8 * i) + 6] = bias_data[(4 * i) + 1];
+ od_param_buffer[(8 * i) + 7] = bias_data[(4 * i) + 3];
}
- // No side padding
- for (; out_x < (output_width - pad_width); ++out_x) {
- const int in_x_origin = (out_x * stride_width) - pad_width;
- CALCULATE_IN_X(in_x_origin);
- PAD_OR_LOAD_ROW_0(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_1(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_2(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_3(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_4(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- COMPUTE(cmds, swizzled_bias_data);
- OUTPUT(output_activation_min, output_activation_max, output_offset,
- local_output_data, n_channels);
- local_output_data += output_depth;
- }
- // Right padding required
- for (; out_x < output_width; ++out_x) {
- const int in_x_origin = (out_x * stride_width) - pad_width;
- CALCULATE_IN_X(in_x_origin);
- PAD_OR_LOAD_ROW_0(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_1(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_2(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_3(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_4(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- COMPUTE(cmds, swizzled_bias_data);
- OUTPUT(output_activation_min, output_activation_max, output_offset,
- local_output_data, n_channels);
- local_output_data += output_depth;
- }
+ vld_b_x_m(v16, &od_param_buffer[0]);
+ vld_b_x(v20, &od_param_buffer[32]);
+ vld_b_x(v21, &od_param_buffer[40]);
+ } else {
+ vdup_b_x_m(v16, 0);
+ vdup_b_x(v20, 0);
+ vdup_b_x(v21, 0);
}
- // No height padding
- for (; out_y < (output_height - pad_height); ++out_y) {
- const int in_y_origin = (out_y * stride_height) - pad_height;
- CALCULATE_IN_Y(in_y_origin);
- // Left padding
- int out_x = 0;
- for (; out_x < pad_width; ++out_x) {
- const int in_x_origin = (out_x * stride_width) - pad_width;
- CALCULATE_IN_X(in_x_origin);
- PAD_OR_LOAD_ROW_0(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_1(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_2(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_3(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_4(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- COMPUTE(cmds, swizzled_bias_data);
- OUTPUT(output_activation_min, output_activation_max, output_offset,
- local_output_data, n_channels);
- local_output_data += output_depth;
- }
- for (; out_x + 2 <= (output_width - pad_width); out_x += 2) {
- const int in_x_origin = (out_x * stride_width) - pad_width;
-
- #pragma GCC unroll 7
- for (int i = 0; i < 7; ++i) {
- in_x[i] = in_x_origin + (dilation_width_factor * i);
+ {
+ // Prepares weight data for preloading. This arrangement is irregular.
+ // Logically, this means the following representation for each OD:
+ // A0 B0 C0 D0 A1 B1 C1 D1 ... A4 B4 C4 D4 E0 E1 E2 E3 E4 0 (x7)
+ // Every 8 OD must then be zipped together into 8 regs (A0*8 B0*8 ......)
+ int8_t filter_regs[output_depth /*=24*/ * 32] __attribute__((aligned(32)));
+ ::memset(filter_regs, 0, output_depth /*=24*/ * 32);
+ for (int ch = 0; ch < output_depth /*=24*/; ++ch) {
+ const int regbank = ch / 8;
+ const int ch_tail = ch & 0x7;
+ for (int y = 0; y < 4; ++y) { // Row 4 will be handled separately.
+ for (int x = 0; x < 5; ++x) {
+ filter_regs[regbank * 256 + x * 32 + ch_tail * 4 + y] =
+ filter_data[tflite::Offset(filter_shape, ch, y, x, 0)];
+ }
}
- const int8_t* p_rows[5];
- #pragma GCC unroll 5
- for (int i = 0; i < 5; ++i) {
- p_rows[i] = p_input + (in_y[i] * input_width);
+ // Reg 5 in each bank is 8ch x E0-E3.
+ for (int x = 0; x < 4; ++x) {
+ filter_regs[regbank * 256 + 5 * 32 + ch_tail * 4 + x] =
+ filter_data[tflite::Offset(filter_shape, ch, 4, x, 0)];
}
-
- vdup_b_x(INPUT_0_0, *(p_rows[0] + in_x[0]));
- vdup_b_x(INPUT_0_1, *(p_rows[0] + in_x[1]));
- vdup_b_x(INPUT_0_2, *(p_rows[0] + in_x[2]));
- vdup_b_x(INPUT_0_3, *(p_rows[0] + in_x[3]));
- vdup_b_x(INPUT_0_4, *(p_rows[0] + in_x[4]));
-
- vdup_b_x(INPUT_1_0, *(p_rows[1] + in_x[0]));
- vdup_b_x(INPUT_1_1, *(p_rows[1] + in_x[1]));
- vdup_b_x(INPUT_1_2, *(p_rows[1] + in_x[2]));
- vdup_b_x(INPUT_1_3, *(p_rows[1] + in_x[3]));
- vdup_b_x(INPUT_1_4, *(p_rows[1] + in_x[4]));
-
- vdup_b_x(INPUT_2_0, *(p_rows[2] + in_x[0]));
- vdup_b_x(INPUT_2_1, *(p_rows[2] + in_x[1]));
- vdup_b_x(INPUT_2_2, *(p_rows[2] + in_x[2]));
- vdup_b_x(INPUT_2_3, *(p_rows[2] + in_x[3]));
- vdup_b_x(INPUT_2_4, *(p_rows[2] + in_x[4]));
-
- vdup_b_x(INPUT_3_0, *(p_rows[3] + in_x[0]));
- vdup_b_x(INPUT_3_1, *(p_rows[3] + in_x[1]));
- vdup_b_x(INPUT_3_2, *(p_rows[3] + in_x[2]));
- vdup_b_x(INPUT_3_3, *(p_rows[3] + in_x[3]));
- vdup_b_x(INPUT_3_4, *(p_rows[3] + in_x[4]));
-
- vdup_b_x(INPUT_4_0, *(p_rows[4] + in_x[0]));
- vdup_b_x(INPUT_4_1, *(p_rows[4] + in_x[1]));
- vdup_b_x(INPUT_4_2, *(p_rows[4] + in_x[2]));
- vdup_b_x(INPUT_4_3, *(p_rows[4] + in_x[3]));
- vdup_b_x(INPUT_4_4, *(p_rows[4] + in_x[4]));
-
- vld_w_x_m(v60, swizzled_bias_data);
- adwinit_v(v60, v60);
- adwconv_vxv(v60, INPUT_0_0, cmds, FLT_0_0);
- adwconv_vxv(v60, INPUT_0_1, cmds, FLT_0_1);
- adwconv_vxv(v60, INPUT_0_2, cmds, FLT_0_2);
- adwconv_vxv(v60, INPUT_0_3, cmds, FLT_0_3);
- adwconv_vxv(v60, INPUT_0_4, cmds, FLT_0_4);
- adwconv_vxv(v60, INPUT_3_0, cmds, FLT_3_0);
- adwconv_vxv(v60, INPUT_3_3, cmds, FLT_3_3);
- adwconv_vxv(v60, INPUT_3_4, cmds, FLT_HOLE);
- vdwconv_vxv(v60, INPUT_4_2, cmds, FLT_4_2);
- vmv_v(INPUT_0_0, v60);
- vmv_v(INPUT_1_0, v61);
- vmv_v(INPUT_2_0, v62);
- vmv_v(INPUT_0_1, v63);
-
- vdup_b_x(INPUT_3_0, *(p_rows[3] + in_x[5]));
- vdup_b_x(INPUT_3_1, *(p_rows[3] + in_x[6]));
-
- vmv_v(INPUT_4_0, INPUT_4_2);
- vmv_v(INPUT_4_1, INPUT_4_3);
- vmv_v(INPUT_4_2, INPUT_4_4);
- vdup_b_x(INPUT_4_3, *(p_rows[4] + in_x[5]));
- vdup_b_x(INPUT_4_4, *(p_rows[4] + in_x[6]));
-
- vld_w_x_m(v60, swizzled_bias_data);
- adwinit_v(v60, v60);
- adwconv_vxv(v60, INPUT_0_2, cmds, FLT_0_0);
- adwconv_vxv(v60, INPUT_0_3, cmds, FLT_0_1);
-
- vmv_v(INPUT_0_2, INPUT_0_0);
- vmv_v(INPUT_1_2, INPUT_1_0);
- vmv_v(INPUT_2_2, INPUT_2_0);
- vmv_v(INPUT_0_3, INPUT_0_1);
-
- vdup_b_x(INPUT_0_0, *(p_rows[0] + in_x[5]));
- vdup_b_x(INPUT_0_1, *(p_rows[0] + in_x[6]));
- vdup_b_x(INPUT_1_0, *(p_rows[1] + in_x[5]));
- vdup_b_x(INPUT_1_1, *(p_rows[1] + in_x[6]));
- vdup_b_x(INPUT_2_0, *(p_rows[2] + in_x[5]));
- vdup_b_x(INPUT_2_1, *(p_rows[2] + in_x[6]));
-
- adwconv_vxv(v60, INPUT_0_4, cmds, FLT_0_2);
- adwconv_vxv(v60, INPUT_0_0, cmds, FLT_0_3);
- adwconv_vxv(v60, INPUT_0_1, cmds, FLT_0_4);
- adwconv_vxv(v60, INPUT_3_2, cmds, FLT_3_0);
- adwconv_vxv(v60, INPUT_3_0, cmds, FLT_3_3);
- adwconv_vxv(v60, INPUT_3_4, cmds, FLT_HOLE);
- vdwconv_vxv(v60, INPUT_4_2, cmds, FLT_4_2);
- INT32_TO_INT8_OUTPUT_PIPELINE_INPLACE2(
- v60, INPUT_0_2, v52, v56, output_activation_min,
- output_activation_max, output_offset
- );
- vsraqs_b_vx(INPUT_0_2, INPUT_0_2, 0);
- vst_b_l_xx(INPUT_0_2, local_output_data, n_channels);
- local_output_data += output_depth;
- vsraqs_b_vx(v60, v60, 0);
- vst_b_l_xx(v60, local_output_data, n_channels);
- local_output_data += output_depth;
+ // Reg 6 in each bank is 8ch x E4.
+ filter_regs[regbank * 256 + 6 * 32 + ch_tail * 4] =
+ filter_data[tflite::Offset(filter_shape, ch, 4, 4, 0)];
+ // Reg 7 is unused and SBZ.
}
- for (; out_x < (output_width - pad_width); ++out_x) {
- const int in_x_origin = (out_x * stride_width) - pad_width;
-
- CALCULATE_IN_X(in_x_origin);
- LOAD_ROW_0(p_input, input_width, in_y, in_x);
- LOAD_ROW_1(p_input, input_width, in_y, in_x);
- LOAD_ROW_2(p_input, input_width, in_y, in_x);
- LOAD_ROW_3(p_input, input_width, in_y, in_x);
- LOAD_ROW_4(p_input, input_width, in_y, in_x);
- COMPUTE(cmds, swizzled_bias_data);
- OUTPUT(output_activation_min, output_activation_max, output_offset,
- local_output_data, n_channels);
- local_output_data += output_depth;
- }
- // Right padding
- for (; out_x < output_width; ++out_x) {
- const int in_x_origin = (out_x * stride_width) - pad_width;
- CALCULATE_IN_X(in_x_origin);
- PAD_OR_LOAD_ROW_0(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_1(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_2(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_3(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_4(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- COMPUTE(cmds, swizzled_bias_data);
- OUTPUT(output_activation_min, output_activation_max, output_offset,
- local_output_data, n_channels);
- local_output_data += output_depth;
- }
+ vld_b_x_m(v8, &filter_regs[0 * 32]);
+ vld_b_x_m(v12, &filter_regs[4 * 32]);
+ vld_b_x_m(v24, &filter_regs[8 * 32]);
+ vld_b_x_m(v28, &filter_regs[12 * 32]);
+ vld_b_x_m(v40, &filter_regs[16 * 32]);
+ vld_b_x_m(v44, &filter_regs[20 * 32]);
}
- // Handle bottom row padding
- for (; out_y < output_height; ++out_y) {
- const int in_y_origin = (out_y * stride_height) - pad_height;
- CALCULATE_IN_Y(in_y_origin);
- int out_x = 0;
- for (; out_x < pad_width; ++out_x) {
- const int in_x_origin = (out_x * stride_width) - pad_width;
- CALCULATE_IN_X(in_x_origin);
- PAD_OR_LOAD_ROW_0(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_1(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_2(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_3(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_4(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- COMPUTE(cmds, swizzled_bias_data);
- OUTPUT(output_activation_min, output_activation_max, output_offset,
- local_output_data, n_channels);
- local_output_data += output_depth;
+ for (int batch = 0; batch < batches; ++batch) {
+ const int8_t* const p_input = &input_data[batch_start_offset(
+ batch, input_height, input_width, input_depth /*=1*/)];
+ int8_t* const p_output = &output_data[batch_start_offset(
+ batch, output_height, output_width, output_depth /*=24*/)];
+
+ // Top loop.
+ for (; in_y_min < 0;
+ ++out_y, in_y_min += stride_height, in_y_max += stride_height) {
+ int out_x_min = 0;
+ int out_x_max = patches_per_iteration /*=8*/ - 1;
+ int in_x_min = -pad_width;
+ int in_x_max = in_x_min + load_width - 1;
+ // Top padding is active so we're not going to read the first row.
+ const int8_t* in_ptr_row1 =
+ &p_input[row_start_offset(in_y_min + 1 * dilation_height_factor,
+ input_width /*=5*/, input_depth /*=1*/)];
+ const int8_t* in_ptr_row2 =
+ &p_input[row_start_offset(in_y_min + 2 * dilation_height_factor,
+ input_width /*=5*/, input_depth /*=1*/)];
+ const int8_t* in_ptr_row3 =
+ &p_input[row_start_offset(in_y_min + 3 * dilation_height_factor,
+ input_width /*=5*/, input_depth /*=1*/)];
+ const int8_t* in_ptr_row4 = &p_input[row_start_offset(
+ in_y_max, input_width /*=5*/, input_depth /*=1*/)];
+ int8_t* out_ptr_col0 = &p_output[row_start_offset(out_y, output_width,
+ output_depth /*=24*/)];
+ int8_t* out_ptr_col4 = out_ptr_col0 + 4 * output_depth /*=24*/;
+
+ // Top left corner.
+ // This could only happen [0:1] times.
+ if (pad_width > 0) {
+ // Same as in_x_max + 1.
+ const int true_load_width = load_width - pad_width;
+
+ // Loads all needed rows.
+ switch (-in_y_min) {
+ case 1:
+ vld_b_l_xx(v23, in_ptr_row1, true_load_width);
+ [[fallthrough]];
+ case 2:
+ vld_b_l_xx(v32, in_ptr_row2, true_load_width);
+ [[fallthrough]];
+ case 3:
+ vld_b_l_xx(v33, in_ptr_row3, true_load_width);
+ [[fallthrough]];
+ case 4:
+ vld_b_l_xx(v34, in_ptr_row4, true_load_width);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+ // Fills padding rows after loading, to improve parallelism.
+ // An extra padding-only row to use in horizontal padding.
+ vdup_b_x(v53, -input_offset);
+ switch (in_y_min) {
+ case -4:
+ vmv_v(v33, v53);
+ [[fallthrough]];
+ case -3:
+ vmv_v(v32, v53);
+ [[fallthrough]];
+ case -2:
+ vmv_v(v23, v53);
+ [[fallthrough]];
+ case -1:
+ // The first row is vertical padding, no need to pad horizontally.
+ vmv_v(v48, v53);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+
+ // Applies left padding as needed.
+ // Can't pass pad_width because the vslide* wants imm.
+ // Can't use vx encoding because scalar is only accepted on the RHS.
+ // v53 is -input_offset broadcasted to all lanes.
+ switch (pad_width) {
+ case 1:
+ vslidep_b_1_vv(v49, v53, v23);
+ vslidep_b_1_vv(v50, v53, v32);
+ vslidep_b_1_vv(v51, v53, v33);
+ vslidep_b_1_vv(v52, v53, v34);
+ break;
+ case 2:
+ vslidep_b_2_vv(v49, v53, v23);
+ vslidep_b_2_vv(v50, v53, v32);
+ vslidep_b_2_vv(v51, v53, v33);
+ vslidep_b_2_vv(v52, v53, v34);
+ break;
+ case 3:
+ vslidep_b_3_vv(v49, v53, v23);
+ vslidep_b_3_vv(v50, v53, v32);
+ vslidep_b_3_vv(v51, v53, v33);
+ vslidep_b_3_vv(v52, v53, v34);
+ break;
+ case 4:
+ vslidep_b_4_vv(v49, v53, v23);
+ vslidep_b_4_vv(v50, v53, v32);
+ vslidep_b_4_vv(v51, v53, v33);
+ vslidep_b_4_vv(v52, v53, v34);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+
+ // Rearranges input data into place.
+ ConvPerChannelD1OD24_5x5_inputshuffle();
+
+ // Computes OD[0..7]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v16, v17); // Also writes v49.
+ vmvp_vv(v50, v16, v17); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v8); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(output_multiplier, output_shift,
+ output_offset, output_activation_min,
+ output_activation_max, out_ptr_col0,
+ out_ptr_col4);
+
+ // Computes OD[8..15]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v18, v19); // Also writes v49.
+ vmvp_vv(v50, v18, v19); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v24); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 8, output_shift + 8, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 8,
+ out_ptr_col4 + 8);
+
+ // Computes OD[16..23]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v20, v21); // Also writes v49.
+ vmvp_vv(v50, v20, v21); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v40); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 16, output_shift + 16, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 16,
+ out_ptr_col4 + 16);
+
+ // Proceed to next block.
+ out_x_min += patches_per_iteration /*=8*/;
+ out_x_max += patches_per_iteration /*=8*/;
+ in_x_min += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_x_max += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_ptr_row1 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ in_ptr_row2 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ in_ptr_row3 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ in_ptr_row4 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ out_ptr_col0 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ out_ptr_col4 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ }
+
+ // Top edge.
+ while (out_x_max < output_width && in_x_max < input_width) {
+ // Loads all rows that we're not padding.
+ switch (-in_y_min) {
+ case 1:
+ vld_b_l_xx(v49, in_ptr_row1, load_width);
+ [[fallthrough]];
+ case 2:
+ vld_b_l_xx(v50, in_ptr_row2, load_width);
+ [[fallthrough]];
+ case 3:
+ vld_b_l_xx(v51, in_ptr_row3, load_width);
+ [[fallthrough]];
+ case 4:
+ vld_b_l_xx(v52, in_ptr_row4, load_width);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+ // Fills padding rows.
+ switch (in_y_min) {
+ case -4:
+ vdup_b_x(v51, -input_offset);
+ [[fallthrough]];
+ case -3:
+ vdup_b_x(v50, -input_offset);
+ [[fallthrough]];
+ case -2:
+ vdup_b_x(v49, -input_offset);
+ [[fallthrough]];
+ case -1:
+ vdup_b_x(v48, -input_offset);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+
+ // Rearranges input data into place.
+ ConvPerChannelD1OD24_5x5_inputshuffle();
+
+ // Computes OD[0..7]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v16, v17); // Also writes v49.
+ vmvp_vv(v50, v16, v17); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v8); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(output_multiplier, output_shift,
+ output_offset, output_activation_min,
+ output_activation_max, out_ptr_col0,
+ out_ptr_col4);
+
+ // Computes OD[8..15]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v18, v19); // Also writes v49.
+ vmvp_vv(v50, v18, v19); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v24); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 8, output_shift + 8, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 8,
+ out_ptr_col4 + 8);
+
+ // Computes OD[16..23]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v20, v21); // Also writes v49.
+ vmvp_vv(v50, v20, v21); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v40); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 16, output_shift + 16, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 16,
+ out_ptr_col4 + 16);
+
+ // Proceed to next block.
+ out_x_min += patches_per_iteration /*=8*/;
+ out_x_max += patches_per_iteration /*=8*/;
+ in_x_min += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_x_max += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_ptr_row1 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row2 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row3 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row4 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ out_ptr_col0 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ out_ptr_col4 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ }
+
+ // Top right corner.
+ while (out_x_min < output_width) {
+ const int true_patches =
+ std::min(output_width - out_x_min, patches_per_iteration);
+ const int true_load_width = std::max(input_width - in_x_min, 0);
+
+ // Prepare the selector vector for right padding.
+ {
+ int8_t selector[32];
+ memset(selector, 1, true_load_width);
+ memset(selector + true_load_width, 0, 32 - true_load_width);
+ vld_b_x(v53, selector);
+ }
+ // Loads all needed rows and applies right padding.
+ switch (-in_y_min) {
+ case 1:
+ vld_b_l_xx(v49, in_ptr_row1, true_load_width);
+ vsel_b_vx(v49, v53, -input_offset);
+ [[fallthrough]];
+ case 2:
+ vld_b_l_xx(v50, in_ptr_row2, true_load_width);
+ vsel_b_vx(v50, v53, -input_offset);
+ [[fallthrough]];
+ case 3:
+ vld_b_l_xx(v51, in_ptr_row3, true_load_width);
+ vsel_b_vx(v51, v53, -input_offset);
+ [[fallthrough]];
+ case 4:
+ vld_b_l_xx(v52, in_ptr_row4, true_load_width);
+ vsel_b_vx(v52, v53, -input_offset);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+ // Fills padding rows.
+ switch (in_y_min) {
+ case -4:
+ vdup_b_x(v51, -input_offset);
+ [[fallthrough]];
+ case -3:
+ vdup_b_x(v50, -input_offset);
+ [[fallthrough]];
+ case -2:
+ vdup_b_x(v49, -input_offset);
+ [[fallthrough]];
+ case -1:
+ vdup_b_x(v48, -input_offset);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+
+ // Rearranges input data into place.
+ ConvPerChannelD1OD24_5x5_inputshuffle();
+
+ // We added enough padding to complete a full block, but VSTQ
+ // cannot have a write-limiter attached. To workaround this we
+ // store to a large-enough buffer and then copy as needed.
+ int8_t vstq_buffer[patches_per_iteration /*=8*/ * output_depth /*=24*/]
+ __attribute__((aligned(32)));
+ int8_t* temp_out_ptr_col0 = &vstq_buffer[0];
+ int8_t* temp_out_ptr_col4 = &vstq_buffer[4 * output_depth /*=24*/];
+
+ // Computes OD[0..7]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v16, v17); // Also writes v49.
+ vmvp_vv(v50, v16, v17); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v8); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(output_multiplier, output_shift,
+ output_offset, output_activation_min,
+ output_activation_max,
+ temp_out_ptr_col0, temp_out_ptr_col4);
+
+ // Computes OD[8..15]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v18, v19); // Also writes v49.
+ vmvp_vv(v50, v18, v19); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v24); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 8, output_shift + 8, output_offset,
+ output_activation_min, output_activation_max, temp_out_ptr_col0 + 8,
+ temp_out_ptr_col4 + 8);
+
+ // Computes OD[16..23]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v20, v21); // Also writes v49.
+ vmvp_vv(v50, v20, v21); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v40); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 16, output_shift + 16, output_offset,
+ output_activation_min, output_activation_max,
+ temp_out_ptr_col0 + 16, temp_out_ptr_col4 + 16);
+
+ // Copies useful results back.
+ ::memcpy(out_ptr_col0, temp_out_ptr_col0,
+ true_patches * output_depth /*=24*/);
+
+ // Proceed to next block.
+ // It is easier to use patches_per_iteration instead of true_patches
+ // here because the former one is constexpr. These two values will
+ // differ iff we've just finished the last block on the row.
+ out_x_min += patches_per_iteration /*=8*/;
+ out_x_max += patches_per_iteration /*=8*/;
+ in_x_min += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_x_max += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_ptr_row1 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row2 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row3 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row4 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ out_ptr_col0 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ out_ptr_col4 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ }
}
- for (; out_x < (output_width - pad_width); ++out_x) {
- const int in_x_origin = (out_x * stride_width) - pad_width;
- CALCULATE_IN_X(in_x_origin);
- PAD_OR_LOAD_ROW_0(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_1(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_2(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_3(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_4(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- COMPUTE(cmds, swizzled_bias_data);
- OUTPUT(output_activation_min, output_activation_max, output_offset,
- local_output_data, n_channels);
- local_output_data += output_depth;
+
+ // Main loop (no vertical padding).
+ for (; out_y < output_height && in_y_max < input_height;
+ ++out_y, in_y_min += stride_height, in_y_max += stride_height) {
+ int out_x_min = 0;
+ int out_x_max = patches_per_iteration /*=8*/ - 1;
+ int in_x_min = -pad_width;
+ int in_x_max = in_x_min + load_width - 1;
+ const int8_t* in_ptr_row0 = &p_input[row_start_offset(
+ in_y_min, input_width /*=5*/, input_depth /*=1*/)];
+ const int8_t* in_ptr_row1 =
+ &p_input[row_start_offset(in_y_min + 1 * dilation_height_factor,
+ input_width /*=5*/, input_depth /*=1*/)];
+ const int8_t* in_ptr_row2 =
+ &p_input[row_start_offset(in_y_min + 2 * dilation_height_factor,
+ input_width /*=5*/, input_depth /*=1*/)];
+ const int8_t* in_ptr_row3 =
+ &p_input[row_start_offset(in_y_min + 3 * dilation_height_factor,
+ input_width /*=5*/, input_depth /*=1*/)];
+ const int8_t* in_ptr_row4 = &p_input[row_start_offset(
+ in_y_max, input_width /*=5*/, input_depth /*=1*/)];
+ int8_t* out_ptr_col0 = &p_output[row_start_offset(out_y, output_width,
+ output_depth /*=24*/)];
+ int8_t* out_ptr_col4 = out_ptr_col0 + 4 * output_depth /*=24*/;
+
+ // Left edge.
+ // This could only happen [0:1] times.
+ if (pad_width > 0) {
+ // Same as in_x_max + 1.
+ const int true_load_width = load_width - pad_width;
+
+ // Loads all needed rows.
+ vld_b_l_xx(v22, in_ptr_row0, true_load_width);
+ vld_b_l_xx(v23, in_ptr_row1, true_load_width);
+ vld_b_l_xx(v32, in_ptr_row2, true_load_width);
+ vld_b_l_xx(v33, in_ptr_row3, true_load_width);
+ vld_b_l_xx(v34, in_ptr_row4, true_load_width);
+
+ // Applies left padding as needed.
+ vdup_b_x(v53, -input_offset);
+ // Can't pass pad_width because the vslide* wants imm.
+ // Can't use vx encoding because scalar has to be on the RHS.
+ // v48 (padded input row 0) is -input_offset broadcasted.
+ switch (pad_width) {
+ case 1:
+ vslidep_b_1_vv(v48, v53, v22);
+ vslidep_b_1_vv(v49, v53, v23);
+ vslidep_b_1_vv(v50, v53, v32);
+ vslidep_b_1_vv(v51, v53, v33);
+ vslidep_b_1_vv(v52, v53, v34);
+ break;
+ case 2:
+ vslidep_b_2_vv(v48, v53, v22);
+ vslidep_b_2_vv(v49, v53, v23);
+ vslidep_b_2_vv(v50, v53, v32);
+ vslidep_b_2_vv(v51, v53, v33);
+ vslidep_b_2_vv(v52, v53, v34);
+ break;
+ case 3:
+ vslidep_b_3_vv(v48, v53, v22);
+ vslidep_b_3_vv(v49, v53, v23);
+ vslidep_b_3_vv(v50, v53, v32);
+ vslidep_b_3_vv(v51, v53, v33);
+ vslidep_b_3_vv(v52, v53, v34);
+ break;
+ case 4:
+ vslidep_b_4_vv(v48, v53, v22);
+ vslidep_b_4_vv(v49, v53, v23);
+ vslidep_b_4_vv(v50, v53, v32);
+ vslidep_b_4_vv(v51, v53, v33);
+ vslidep_b_4_vv(v52, v53, v34);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+
+ // Rearranges input data into place.
+ ConvPerChannelD1OD24_5x5_inputshuffle();
+
+ // Computes OD[0..7]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v16, v17); // Also writes v49.
+ vmvp_vv(v50, v16, v17); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v8); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(output_multiplier, output_shift,
+ output_offset, output_activation_min,
+ output_activation_max, out_ptr_col0,
+ out_ptr_col4);
+
+ // Computes OD[8..15]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v18, v19); // Also writes v49.
+ vmvp_vv(v50, v18, v19); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v24); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 8, output_shift + 8, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 8,
+ out_ptr_col4 + 8);
+
+ // Computes OD[16..23]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v20, v21); // Also writes v49.
+ vmvp_vv(v50, v20, v21); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v40); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 16, output_shift + 16, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 16,
+ out_ptr_col4 + 16);
+
+ // Proceed to next block.
+ out_x_min += patches_per_iteration /*=8*/;
+ out_x_max += patches_per_iteration /*=8*/;
+ in_x_min += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_x_max += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_ptr_row0 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ in_ptr_row1 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ in_ptr_row2 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ in_ptr_row3 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ in_ptr_row4 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ out_ptr_col0 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ out_ptr_col4 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ }
+
+ // Center.
+ while (out_x_max < output_width && in_x_max < input_width) {
+ // Loads all needed rows.
+ // TODO(davidgao): all these reads are misaligned.
+ // Pad the input image may give further speedup.
+ vld_b_l_xx(v48, in_ptr_row0, load_width);
+ vld_b_l_xx(v49, in_ptr_row1, load_width);
+ vld_b_l_xx(v50, in_ptr_row2, load_width);
+ vld_b_l_xx(v51, in_ptr_row3, load_width);
+ vld_b_l_xx(v52, in_ptr_row4, load_width);
+
+ // Rearranges input data into place.
+ ConvPerChannelD1OD24_5x5_inputshuffle();
+
+ // Computes OD[0..7]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v16, v17); // Also writes v49.
+ vmvp_vv(v50, v16, v17); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v8); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(output_multiplier, output_shift,
+ output_offset, output_activation_min,
+ output_activation_max, out_ptr_col0,
+ out_ptr_col4);
+
+ // Computes OD[8..15]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v18, v19); // Also writes v49.
+ vmvp_vv(v50, v18, v19); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v24); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 8, output_shift + 8, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 8,
+ out_ptr_col4 + 8);
+
+ // Computes OD[16..23]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v20, v21); // Also writes v49.
+ vmvp_vv(v50, v20, v21); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v40); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 16, output_shift + 16, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 16,
+ out_ptr_col4 + 16);
+
+ // Proceed to next block.
+ out_x_min += patches_per_iteration /*=8*/;
+ out_x_max += patches_per_iteration /*=8*/;
+ in_x_min += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_x_max += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_ptr_row0 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row1 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row2 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row3 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row4 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ out_ptr_col0 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ out_ptr_col4 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ }
+
+ // Right edge.
+ while (out_x_min < output_width) {
+ const int true_patches =
+ std::min(output_width - out_x_min, patches_per_iteration);
+ const int true_load_width = std::max(input_width - in_x_min, 0);
+
+ // Prepare the selector vector for right padding.
+ {
+ int8_t selector[32];
+ memset(selector, 1, true_load_width);
+ memset(selector + true_load_width, 0, 32 - true_load_width);
+ vld_b_x(v53, selector);
+ }
+ // Loads all needed rows and applies right padding.
+ // Loads all needed rows.
+ vld_b_l_xx(v48, in_ptr_row0, true_load_width);
+ vld_b_l_xx(v49, in_ptr_row1, true_load_width);
+ vld_b_l_xx(v50, in_ptr_row2, true_load_width);
+ vld_b_l_xx(v51, in_ptr_row3, true_load_width);
+ vld_b_l_xx(v52, in_ptr_row4, true_load_width);
+ vsel_b_vx(v48, v53, -input_offset);
+ vsel_b_vx(v49, v53, -input_offset);
+ vsel_b_vx(v50, v53, -input_offset);
+ vsel_b_vx(v51, v53, -input_offset);
+ vsel_b_vx(v52, v53, -input_offset);
+
+ // Rearranges input data into place.
+ ConvPerChannelD1OD24_5x5_inputshuffle();
+
+ // We added enough padding to complete a full block, but VSTQ
+ // cannot have a write-limiter attached. To workaround this we
+ // store to a large-enough buffer and then copy as needed.
+ int8_t vstq_buffer[patches_per_iteration /*=8*/ * output_depth /*=24*/]
+ __attribute__((aligned(32)));
+ int8_t* temp_out_ptr_col0 = &vstq_buffer[0];
+ int8_t* temp_out_ptr_col4 = &vstq_buffer[4 * output_depth /*=24*/];
+
+ // Computes OD[0..7]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v16, v17); // Also writes v49.
+ vmvp_vv(v50, v16, v17); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v8); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(output_multiplier, output_shift,
+ output_offset, output_activation_min,
+ output_activation_max,
+ temp_out_ptr_col0, temp_out_ptr_col4);
+
+ // Computes OD[8..15]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v18, v19); // Also writes v49.
+ vmvp_vv(v50, v18, v19); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v24); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 8, output_shift + 8, output_offset,
+ output_activation_min, output_activation_max, temp_out_ptr_col0 + 8,
+ temp_out_ptr_col4 + 8);
+
+ // Computes OD[16..23]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v20, v21); // Also writes v49.
+ vmvp_vv(v50, v20, v21); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v40); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 16, output_shift + 16, output_offset,
+ output_activation_min, output_activation_max,
+ temp_out_ptr_col0 + 16, temp_out_ptr_col4 + 16);
+
+ // Copies useful results back.
+ // TODO(davidgao): this could use some vector copying.
+ ::memcpy(out_ptr_col0, temp_out_ptr_col0,
+ true_patches * output_depth /*=24*/);
+
+ // Proceed to next block.
+ // It is easier to use patches_per_iteration instead of true_patches
+ // here because the former one is constexpr. These two values will
+ // differ iff we've just finished the last block on the row.
+ out_x_min += patches_per_iteration /*=8*/;
+ out_x_max += patches_per_iteration /*=8*/;
+ in_x_min += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_x_max += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_ptr_row0 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row1 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row2 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row3 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row4 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ out_ptr_col0 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ out_ptr_col4 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ }
}
- for (; out_x < output_width; ++out_x) {
- const int in_x_origin = (out_x * stride_width) - pad_width;
- CALCULATE_IN_X(in_x_origin);
- PAD_OR_LOAD_ROW_0(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_1(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_2(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_3(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- PAD_OR_LOAD_ROW_4(p_input, input_height, input_width, in_y, in_x,
- input_offset);
- COMPUTE(cmds, swizzled_bias_data);
- OUTPUT(output_activation_min, output_activation_max, output_offset,
- local_output_data, n_channels);
- local_output_data += output_depth;
+
+ // Bottom loop.
+ for (; out_y < output_height;
+ ++out_y, in_y_min += stride_height, in_y_max += stride_height) {
+ int out_x_min = 0;
+ int out_x_max = patches_per_iteration - 1;
+ int in_x_min = -pad_width;
+ int in_x_max = in_x_min + load_width - 1;
+ // Bottom padding is active so we're not going to read the last row.
+ const int8_t* in_ptr_row0 = &p_input[row_start_offset(
+ in_y_min, input_width /*=5*/, input_depth /*=1*/)];
+ const int8_t* in_ptr_row1 = &p_input[row_start_offset(
+ in_y_min + 1, input_width /*=5*/, input_depth /*=1*/)];
+ const int8_t* in_ptr_row2 = &p_input[row_start_offset(
+ in_y_min + 2, input_width /*=5*/, input_depth /*=1*/)];
+ const int8_t* in_ptr_row3 = &p_input[row_start_offset(
+ in_y_min + 3, input_width /*=5*/, input_depth /*=1*/)];
+ int8_t* out_ptr_col0 = &p_output[row_start_offset(out_y, output_width,
+ output_depth /*=24*/)];
+ int8_t* out_ptr_col4 = out_ptr_col0 + 4 * output_depth /*=24*/;
+
+ // Bottom left corner.
+ // This could only happen [0:1] times.
+ if (pad_width > 0) {
+ // Same as in_x_max + 1.
+ const int true_load_width = load_width - pad_width;
+
+ // Loads all needed rows.
+ switch (in_y_max - input_height) {
+ case 0:
+ vld_b_l_xx(v33, in_ptr_row3, true_load_width);
+ [[fallthrough]];
+ case 1:
+ vld_b_l_xx(v32, in_ptr_row2, true_load_width);
+ [[fallthrough]];
+ case 2:
+ vld_b_l_xx(v23, in_ptr_row1, true_load_width);
+ [[fallthrough]];
+ case 3:
+ vld_b_l_xx(v22, in_ptr_row0, true_load_width);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+ // Fills padding rows.
+ vdup_b_x(v53, -input_offset);
+ switch (input_height - in_y_max) {
+ case -3:
+ vmv_v(v23, v53);
+ [[fallthrough]];
+ case -2:
+ vmv_v(v32, v53);
+ [[fallthrough]];
+ case -1:
+ vmv_v(v33, v53);
+ [[fallthrough]];
+ case 0:
+ // The last row always padding-only.
+ vmv_v(v52, v53);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+
+ // Applies left padding as needed.
+ // Can't pass pad_width because the vslide* wants imm.
+ // Can't use vx encoding because scalar has to be on the RHS.
+ // v53 is -input_offset broadcasted to all lanes.
+ switch (pad_width) {
+ case 1:
+ vslidep_b_1_vv(v48, v53, v22);
+ vslidep_b_1_vv(v49, v53, v23);
+ vslidep_b_1_vv(v50, v53, v32);
+ vslidep_b_1_vv(v51, v53, v33);
+ break;
+ case 2:
+ vslidep_b_2_vv(v48, v53, v22);
+ vslidep_b_2_vv(v49, v53, v23);
+ vslidep_b_2_vv(v50, v53, v32);
+ vslidep_b_2_vv(v51, v53, v33);
+ break;
+ case 3:
+ vslidep_b_3_vv(v48, v53, v22);
+ vslidep_b_3_vv(v49, v53, v23);
+ vslidep_b_3_vv(v50, v53, v32);
+ vslidep_b_3_vv(v51, v53, v33);
+ break;
+ case 4:
+ vslidep_b_4_vv(v48, v53, v22);
+ vslidep_b_4_vv(v49, v53, v23);
+ vslidep_b_4_vv(v50, v53, v32);
+ vslidep_b_4_vv(v51, v53, v33);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+
+ // Rearranges input data into place.
+ ConvPerChannelD1OD24_5x5_inputshuffle();
+
+ // Computes OD[0..7]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v16, v17); // Also writes v49.
+ vmvp_vv(v50, v16, v17); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v8); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(output_multiplier, output_shift,
+ output_offset, output_activation_min,
+ output_activation_max, out_ptr_col0,
+ out_ptr_col4);
+
+ // Computes OD[8..15]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v18, v19); // Also writes v49.
+ vmvp_vv(v50, v18, v19); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v24); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 8, output_shift + 8, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 8,
+ out_ptr_col4 + 8);
+
+ // Computes OD[16..23]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v20, v21); // Also writes v49.
+ vmvp_vv(v50, v20, v21); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v40); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 16, output_shift + 16, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 16,
+ out_ptr_col4 + 16);
+
+ // Proceed to next block.
+ out_x_min += patches_per_iteration /*=8*/;
+ out_x_max += patches_per_iteration /*=8*/;
+ in_x_min += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_x_max += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_ptr_row0 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ in_ptr_row1 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ in_ptr_row2 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ in_ptr_row3 +=
+ (patches_per_iteration /*=8*/ * stride_width /*=2*/ - pad_width) *
+ input_depth /*=1*/;
+ out_ptr_col0 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ out_ptr_col4 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ }
+
+ // Bottom edge.
+ while (out_x_max < output_width && in_x_max < input_width) {
+ // Loads all needed rows.
+ switch (in_y_max - input_height) {
+ case 0:
+ vld_b_l_xx(v51, in_ptr_row3, load_width);
+ [[fallthrough]];
+ case 1:
+ vld_b_l_xx(v50, in_ptr_row2, load_width);
+ [[fallthrough]];
+ case 2:
+ vld_b_l_xx(v49, in_ptr_row1, load_width);
+ [[fallthrough]];
+ case 3:
+ vld_b_l_xx(v48, in_ptr_row0, load_width);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+ // Fills padding rows.
+ switch (input_height - in_y_max) {
+ case -3:
+ vdup_b_x(v49, -input_offset);
+ [[fallthrough]];
+ case -2:
+ vdup_b_x(v50, -input_offset);
+ [[fallthrough]];
+ case -1:
+ vdup_b_x(v51, -input_offset);
+ [[fallthrough]];
+ case 0:
+ vdup_b_x(v52, -input_offset);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+
+ // Rearranges input data into place.
+ ConvPerChannelD1OD24_5x5_inputshuffle();
+
+ // Computes OD[0..7]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v16, v17); // Also writes v49.
+ vmvp_vv(v50, v16, v17); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v8); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(output_multiplier, output_shift,
+ output_offset, output_activation_min,
+ output_activation_max, out_ptr_col0,
+ out_ptr_col4);
+
+ // Computes OD[8..15]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v18, v19); // Also writes v49.
+ vmvp_vv(v50, v18, v19); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v24); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 8, output_shift + 8, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 8,
+ out_ptr_col4 + 8);
+
+ // Computes OD[16..23]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v20, v21); // Also writes v49.
+ vmvp_vv(v50, v20, v21); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v40); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 16, output_shift + 16, output_offset,
+ output_activation_min, output_activation_max, out_ptr_col0 + 16,
+ out_ptr_col4 + 16);
+
+ // Proceed to next block.
+ out_x_min += patches_per_iteration /*=8*/;
+ out_x_max += patches_per_iteration /*=8*/;
+ in_x_min += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_x_max += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_ptr_row0 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row1 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row2 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row3 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ out_ptr_col0 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ out_ptr_col4 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ }
+
+ // Bottom right corner.
+ while (out_x_min < output_width) {
+ const int true_patches =
+ std::min(output_width - out_x_min, patches_per_iteration);
+ const int true_load_width = std::max(input_width - in_x_min, 0);
+
+ // Prepare the selector vector for right padding.
+ {
+ int8_t selector[32];
+ memset(selector, 1, true_load_width);
+ memset(selector + true_load_width, 0, 32 - true_load_width);
+ vld_b_x(v53, selector);
+ }
+ // Loads all needed rows and applies right padding.
+ switch (in_y_max - input_height) {
+ case 0:
+ vld_b_l_xx(v51, in_ptr_row3, true_load_width);
+ vsel_b_vx(v51, v53, -input_offset);
+ [[fallthrough]];
+ case 1:
+ vld_b_l_xx(v50, in_ptr_row2, true_load_width);
+ vsel_b_vx(v50, v53, -input_offset);
+ [[fallthrough]];
+ case 2:
+ vld_b_l_xx(v49, in_ptr_row1, true_load_width);
+ vsel_b_vx(v49, v53, -input_offset);
+ [[fallthrough]];
+ case 3:
+ vld_b_l_xx(v48, in_ptr_row0, true_load_width);
+ vsel_b_vx(v48, v53, -input_offset);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+ // Fills padding rows.
+ switch (input_height - in_y_max) {
+ case -3:
+ vdup_b_x(v49, -input_offset);
+ [[fallthrough]];
+ case -2:
+ vdup_b_x(v50, -input_offset);
+ [[fallthrough]];
+ case -1:
+ vdup_b_x(v51, -input_offset);
+ [[fallthrough]];
+ case 0:
+ vdup_b_x(v52, -input_offset);
+ break;
+ default:
+ __builtin_unreachable();
+ }
+
+ // Rearranges input data into place.
+ ConvPerChannelD1OD24_5x5_inputshuffle();
+
+ // We added enough padding to complete a full block, but VSTQ
+ // cannot have a write-limiter attached. To workaround this we
+ // store to a large-enough buffer and then copy as needed.
+ int8_t vstq_buffer[patches_per_iteration /*=8*/ * output_depth /*=24*/]
+ __attribute__((aligned(32)));
+ int8_t* temp_out_ptr_col0 = &vstq_buffer[0];
+ int8_t* temp_out_ptr_col4 = &vstq_buffer[4 * output_depth /*=24*/];
+
+ // Computes OD[0..7]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v16, v17); // Also writes v49.
+ vmvp_vv(v50, v16, v17); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v8); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(output_multiplier, output_shift,
+ output_offset, output_activation_min,
+ output_activation_max,
+ temp_out_ptr_col0, temp_out_ptr_col4);
+
+ // Computes OD[8..15]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v18, v19); // Also writes v49.
+ vmvp_vv(v50, v18, v19); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v24); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 8, output_shift + 8, output_offset,
+ output_activation_min, output_activation_max, temp_out_ptr_col0 + 8,
+ temp_out_ptr_col4 + 8);
+
+ // Computes OD[16..23]
+ // Initializes the accumulators from bias.
+ vmvp_vv(v48, v20, v21); // Also writes v49.
+ vmvp_vv(v50, v20, v21); // Also writes v51.
+ vmv_v_m(v52, v48);
+ actr_v(v48, v48); // v48 is read but not written.
+ // Performs matmul.
+ aconv_vxv(v48, v0, aconv_cmd, v40); // v48 is not actually written.
+ ConvPerChannelD1OD24_5x5_postproc(
+ output_multiplier + 16, output_shift + 16, output_offset,
+ output_activation_min, output_activation_max,
+ temp_out_ptr_col0 + 16, temp_out_ptr_col4 + 16);
+
+ // Copies useful results back.
+ ::memcpy(out_ptr_col0, temp_out_ptr_col0,
+ true_patches * output_depth /*=24*/);
+
+ // Proceed to next block.
+ // It is easier to use patches_per_iteration instead of true_patches
+ // here because the former one is constexpr. These two values will
+ // differ iff we've just finished the last block on the row.
+ out_x_min += patches_per_iteration /*=8*/;
+ out_x_max += patches_per_iteration /*=8*/;
+ in_x_min += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_x_max += patches_per_iteration /*=8*/ * stride_width /*=2*/;
+ in_ptr_row0 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row1 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row2 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ in_ptr_row3 += (patches_per_iteration /*=8*/ * stride_width /*=2*/) *
+ input_depth /*=1*/;
+ out_ptr_col0 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ out_ptr_col4 += patches_per_iteration /*=8*/ * output_depth /*=24*/;
+ }
}
}
}
-#undef PAD_OR_LOAD_ROW_0
-#undef PAD_OR_LOAD_ROW_1
-#undef PAD_OR_LOAD_ROW_2
-#undef PAD_OR_LOAD_ROW_3
-#undef PAD_OR_LOAD_ROW_4
-#undef _PAD_OR_LOAD_ROW
-#undef _H_PAD_OR_LOAD_ROW
-#undef H_PAD_OR_LOAD_ROW_0
-#undef H_PAD_OR_LOAD_ROW_1
-#undef H_PAD_OR_LOAD_ROW_2
-#undef H_PAD_OR_LOAD_ROW_3
-#undef H_PAD_OR_LOAD_ROW_4
-#undef LOAD_ROW_0
-#undef LOAD_ROW_1
-#undef LOAD_ROW_2
-#undef LOAD_ROW_3
-#undef LOAD_ROW_4
-#undef PAD_ROW_0
-#undef PAD_ROW_1
-#undef PAD_ROW_2
-#undef PAD_ROW_3
-#undef PAD_ROW_4
-#undef CALCULATE_IN_X
-#undef CALCULATE_IN_Y
-#undef INPUT_0_0
-#undef INPUT_0_1
-#undef INPUT_0_2
-#undef INPUT_0_3
-#undef INPUT_0_4
-#undef INPUT_1_0
-#undef INPUT_1_1
-#undef INPUT_1_2
-#undef INPUT_1_3
-#undef INPUT_1_4
-#undef INPUT_2_0
-#undef INPUT_2_1
-#undef INPUT_2_2
-#undef INPUT_2_3
-#undef INPUT_2_4
-#undef INPUT_3_0
-#undef INPUT_3_1
-#undef INPUT_3_2
-#undef INPUT_3_3
-#undef INPUT_3_4
-#undef INPUT_4_0
-#undef INPUT_4_1
-#undef INPUT_4_2
-#undef INPUT_4_3
-#undef INPUT_4_4
-#undef INPUT_0_5
-#undef INPUT_1_5
-#undef INPUT_2_5
-#undef INPUT_3_5
-#undef INPUT_4_5
-#undef FLT_0_0
-#undef FLT_0_1
-#undef FLT_0_2
-#undef FLT_0_3
-#undef FLT_0_4
-#undef FLT_1_0
-#undef FLT_1_1
-#undef FLT_1_2
-#undef FLT_1_3
-#undef FLT_1_4
-#undef FLT_2_0
-#undef FLT_2_1
-#undef FLT_2_2
-#undef FLT_2_3
-#undef FLT_2_4
-#undef FLT_3_0
-#undef FLT_3_1
-#undef FLT_3_2
-#undef FLT_3_3
-#undef FLT_3_4
-#undef FLT_HOLE
-#undef FLT_4_0
-#undef FLT_4_1
-#undef FLT_4_2
-#undef FLT_4_3
-#undef FLT_4_4
-
void ConvPerChannelD1(
const tflite::ConvParams& params, const int32_t* output_multiplier,
const int32_t* output_shift, const tflite::RuntimeShape& input_shape,
@@ -1106,4 +1597,4 @@
}
}
-} // namespace kelvin::opt
\ No newline at end of file
+} // namespace kelvin::opt
