From bc672082ae31778164ed3ec23b7a4a8f1a8dc454 Mon Sep 17 00:00:00 2001 From: Gian Marco Iodice Date: Tue, 10 Jan 2023 12:46:29 +0000 Subject: Update the heuristic for CLDepthwiseConvolutionNative kernel MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - Use T_LOAD2D_INDIRECT macro instead of T_LOAD_NHWC_WITH_DILATION in the depthwise convolution opencl kernels - Update the heuristic for Arm® Mali™-G77 Resolves COMPMID-5716 Signed-off-by: Gian Marco Iodice Change-Id: I32d375b220e04bf05f5d8f0af2231bde600f0665 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8930 Benchmark: Arm Jenkins Tested-by: Arm Jenkins Reviewed-by: Jakub Sujak Reviewed-by: Viet-Hoa Do Comments-Addressed: Arm Jenkins --- src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl | 17 +++++++++++++++-- .../CL/cl_kernels/nhwc/dwc_native_quantized_nhwc.cl | 18 ++++++++++++++++-- .../dwc_native/ClDWCNativeDefaultConfigValhall.cpp | 19 +++++++++++++++++-- 3 files changed, 48 insertions(+), 6 deletions(-) diff --git a/src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl b/src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl index dcbae220b6..6d64e270ef 100644 --- a/src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl +++ b/src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl @@ -108,7 +108,6 @@ __kernel void dwc_native_fp_nhwc( #define _IN0_A N0_A // Cols tile A. It can be either 1 (for DEPTH_MULTIPLIER > 1) or N0 (for DEPTH_MULTIPLIER == 1) #define _IM0_B _IWEI_WIDTH // Rows tile B #define _IN0_B N0 // Cols tile B -#define _IBOUNDARY_CHECK (!((WEI_WIDTH == 1 && WEI_HEIGHT == 1 && PAD_LEFT == 0 && PAD_TOP == 0 && M0 == 1))) const int cout = GET_SPATIAL_IDX(0, N0, PARTIAL_N0); // OFM const int xo = GET_SPATIAL_IDX(1, M0, 0); // WIDTH @@ -146,8 +145,22 @@ __kernel void dwc_native_fp_nhwc( a[i].v = 0; }) + TILE(int, 1, _IM0_A, mi); + + LOOP_UNROLLING(int, xk_i, 0, 1, _IM0_A, + { + int x_s = xi + xk_i * (DILATION_X); + int y_s = yi + yk * (DILATION_Y); + mi[0].s[xk_i] = x_s + y_s * SRC_WIDTH; + mi[0].s[xk_i] = mi[0].s[xk_i] + bout * (int)(SRC_WIDTH * SRC_HEIGHT); + mi[0].s[xk_i] = select(-1, mi[0].s[xk_i], x_s >= 0); + mi[0].s[xk_i] = select(-1, mi[0].s[xk_i], x_s < SRC_WIDTH); + mi[0].s[xk_i] = select(-1, mi[0].s[xk_i], y_s >= 0); + mi[0].s[xk_i] = select(-1, mi[0].s[xk_i], y_s < SRC_HEIGHT); + }) + // Load tile from the src tensor (TILE A) - T_LOAD_NHWC_WITH_DILATION(SRC_DATA_TYPE, 1, _IM0_A, _IN0_A, SRC_TENSOR_TYPE, src, bout, yi + yk * DILATION_Y, xi, (cout / DEPTH_MULTIPLIER), SRC_WIDTH, SRC_HEIGHT, DILATION_X, 1, _IBOUNDARY_CHECK, a); + T_LOAD2D_INDIRECT(SRC_DATA_TYPE, _IM0_A, _IN0_A, SRC_TENSOR_TYPE, src, (cout / DEPTH_MULTIPLIER), src_stride_y, mi, a); TILE(WEI_DATA_TYPE, _IM0_B, _IN0_B, b); diff --git a/src/core/CL/cl_kernels/nhwc/dwc_native_quantized_nhwc.cl b/src/core/CL/cl_kernels/nhwc/dwc_native_quantized_nhwc.cl index 2d255e5b61..e502d721d5 100644 --- a/src/core/CL/cl_kernels/nhwc/dwc_native_quantized_nhwc.cl +++ b/src/core/CL/cl_kernels/nhwc/dwc_native_quantized_nhwc.cl @@ -180,8 +180,22 @@ __kernel void dwc_native_quantized_nhwc( a[i].v = ZERO_VALUE; }) - // Load tile from the src tensor (TILE A) - T_LOAD_NHWC_WITH_DILATION(SRC_DATA_TYPE, 1, _IM0_A, _IN0_A, SRC_TENSOR_TYPE, src, bout, yi + yk * DILATION_Y, xi, (cout / DEPTH_MULTIPLIER), src_w, src_h, DILATION_X, 1, _IBOUNDARY_CHECK, a); + TILE(int, 1, _IM0_A, my); + + LOOP_UNROLLING(int, xk_i, 0, 1, _IM0_A, + { + int x_s = xi + xk_i * (DILATION_X); + int y_s = yi + yk * (DILATION_Y); + my[0].s[xk_i] = x_s + y_s * SRC_WIDTH; + my[0].s[xk_i] = my[0].s[xk_i] + bout * (int)(SRC_WIDTH * SRC_HEIGHT); + my[0].s[xk_i] = select(-1, my[0].s[xk_i], x_s >= 0); + my[0].s[xk_i] = select(-1, my[0].s[xk_i], x_s < SRC_WIDTH); + my[0].s[xk_i] = select(-1, my[0].s[xk_i], y_s >= 0); + my[0].s[xk_i] = select(-1, my[0].s[xk_i], y_s < SRC_HEIGHT); + }) + + // Load tile from the src tensor + T_LOAD2D_INDIRECT(SRC_DATA_TYPE, _IM0_A, _IN0_A, SRC_TENSOR_TYPE, src, (cout / DEPTH_MULTIPLIER), src_stride_y, my, a); TILE(WEI_DATA_TYPE, _IM0_B, _IN0_B, b); diff --git a/src/runtime/heuristics/dwc_native/ClDWCNativeDefaultConfigValhall.cpp b/src/runtime/heuristics/dwc_native/ClDWCNativeDefaultConfigValhall.cpp index 49485c83a9..d0ade1bdd7 100644 --- a/src/runtime/heuristics/dwc_native/ClDWCNativeDefaultConfigValhall.cpp +++ b/src/runtime/heuristics/dwc_native/ClDWCNativeDefaultConfigValhall.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2022 Arm Limited. + * Copyright (c) 2022-2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -283,15 +283,30 @@ DWCComputeKernelInfo ClDWCNativeDefaultConfigValhall::configure_G77_f16(const IT desc.n0 = adjust_vec_size(desc.n0, kernel_c); // Set m0 only if stride_x == 1 and dilation_x == 1 + // m0 affects the number of rows to load from the input tensor. In fact, when depth_multiplier = 1, the number of rows + // loaded from the input tensors are -> kernel_width - (M0 - 1) + // The bigger the kernel_width, the smaller the M0 to avoid register spilling. if(conv_info.stride().first == 1 && dilation.x() == 1) { + // When the kernel width and kernel height are unit, it means that we have a pointwise multiplication. Therefore, M0 can be 1 if((kernel_w >= 9) || (kernel_w == 1)) { desc.m0 = 1; } else { - desc.m0 = 2; + switch(kernel_w) + { + case 3: + desc.m0 = 4; + break; + case 5: + desc.m0 = 3; + break; + default: + desc.m0 = 2; + break; + } } } else -- cgit v1.2.1