diff options
author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2018-12-07 11:18:09 +0000 |
---|---|---|
committer | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2018-12-11 14:55:03 +0000 |
commit | 3b0a2654034714c16f5930d2b24936d8be7b18a6 (patch) | |
tree | 2a3ed3703a4f454ef42084363049aba3bf54ebd6 /arm_compute/core/utils | |
parent | ff0bccfb4697c591d569db9c2dc223f2e311a7d3 (diff) | |
download | ComputeLibrary-3b0a2654034714c16f5930d2b24936d8be7b18a6.tar.gz |
COMPMID-1775: Implement CLGEMMReshapeRHSMatrixKernel to reshape the RHS matrix of GEMM/GEMMLowp
Change-Id: I77f2bfcc5d170bcc2428a2f27104942c1ec877d7
Reviewed-on: https://review.mlplatform.org/375
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'arm_compute/core/utils')
-rw-r--r-- | arm_compute/core/utils/misc/ShapeCalculator.h | 28 |
1 files changed, 28 insertions, 0 deletions
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h index 88ce8d9e7b..33893ad877 100644 --- a/arm_compute/core/utils/misc/ShapeCalculator.h +++ b/arm_compute/core/utils/misc/ShapeCalculator.h @@ -138,6 +138,34 @@ inline TensorShape compute_lhs_reshaped_shape(const ITensorInfo &a, const GEMMLH return lhs_shape; } +inline TensorShape compute_rhs_reshaped_shape(const ITensorInfo &a, const GEMMRHSMatrixInfo &rhs_info) +{ + ARM_COMPUTE_ERROR_ON(rhs_info.n0 == 0); + ARM_COMPUTE_ERROR_ON(rhs_info.k0 == 0); + ARM_COMPUTE_ERROR_ON(rhs_info.h0 == 0); + + // Input width/height + const unsigned int input_width = a.dimension(0); + const unsigned int input_height = a.dimension(1); + + // Number of horizontal/vertical blocks in the input tensor + const unsigned int num_horiz_blocks = std::ceil(input_width / static_cast<float>(rhs_info.n0)); + const unsigned int num_vert_blocks = std::ceil(input_height / static_cast<float>(rhs_info.k0)); + + // Block size + const unsigned int block_size = rhs_info.n0 * rhs_info.k0; + + // Output width/height + const unsigned int output_width = block_size * num_vert_blocks * rhs_info.h0; + const unsigned int output_height = std::ceil(num_horiz_blocks / static_cast<float>(rhs_info.h0)); + + TensorShape rhs_shape{ a.tensor_shape() }; + rhs_shape.set(0, output_width); + rhs_shape.set(1, output_height); + + return rhs_shape; +} + inline TensorShape compute_interleaved_shape(const ITensorInfo &a, int mult_interleave4x4_height = 1, bool reinterpret_input_as_3d = false) { // The interleaved output matrix will have the following shape: [ a_height * W, ceil(a_width / W) ] where W = 4 * mult_interleave4x4_height |