Fix dynamic weights for CPU connected layer
Resolves: COMPMID-5995
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Change-Id: I707b8918bebee7e70d4de5207ef555c806e7a305
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9405
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/runtime/NEON/functions/NEGEMM.cpp b/src/runtime/NEON/functions/NEGEMM.cpp
index 0266c48..e51f2f9 100644
--- a/src/runtime/NEON/functions/NEGEMM.cpp
+++ b/src/runtime/NEON/functions/NEGEMM.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2022 Arm Limited.
+ * Copyright (c) 2017-2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -71,7 +71,14 @@
_impl->original_b = b;
_impl->op = std::make_unique<cpu::CpuGemm>();
- _impl->op->configure(a->info(), b->info(), (c != nullptr) ? c->info() : nullptr, d->info(), alpha, beta, gemm_info);
+ // Make the B matrix dynamic values.
+ auto b_info_to_use = b->info()->clone();
+ if(!gemm_info.reshape_b_only_on_first_run())
+ {
+ b_info_to_use->set_are_values_constant(false);
+ }
+
+ _impl->op->configure(a->info(), b_info_to_use.get(), (c != nullptr) ? c->info() : nullptr, d->info(), alpha, beta, gemm_info);
_impl->aux_mem_req = _impl->op->workspace();
_impl->run_pack = { { ACL_SRC_0, a }, { ACL_SRC_1, b }, { ACL_SRC_2, c }, { ACL_DST, d } };
@@ -81,7 +88,14 @@
Status NEGEMM::validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, const ITensorInfo *output, float alpha, float beta, const GEMMInfo &gemm_info)
{
- return cpu::CpuGemm::validate(a, b, c, output, alpha, beta, gemm_info);
+ // Make the B matrix dynamic values.
+ auto b_to_use = b->clone();
+ if(!gemm_info.reshape_b_only_on_first_run())
+ {
+ b_to_use->set_are_values_constant(false);
+ }
+
+ return cpu::CpuGemm::validate(a, b_to_use.get(), c, output, alpha, beta, gemm_info);
}
Status NEGEMM::has_opt_impl(arm_compute::WeightFormat &expected_weight_format, const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, const ITensorInfo *output,
diff --git a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
index 6c179f8..453d3ced 100644
--- a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
+++ b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2021 Arm Limited.
+ * Copyright (c) 2017-2021, 2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -61,9 +61,17 @@
void NEGEMMLowpMatrixMultiplyCore::configure(const ITensor *a, const ITensor *b, const ITensor *c, ITensor *output, const GEMMInfo &gemm_info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(a, b, output);
+
+ // Make the B matrix dynamic values.
+ auto b_info_to_use = b->info()->clone();
+ if(!gemm_info.reshape_b_only_on_first_run())
+ {
+ b_info_to_use->set_are_values_constant(false);
+ }
+
_impl->b = b;
_impl->op = std::make_unique<cpu::CpuGemmLowpMatrixMultiplyCore>();
- _impl->op->configure(a->info(), b->info(), (c != nullptr ? c->info() : nullptr), output->info(), gemm_info);
+ _impl->op->configure(a->info(), b_info_to_use.get(), (c != nullptr ? c->info() : nullptr), output->info(), gemm_info);
_impl->run_pack =
{
{ TensorType::ACL_SRC_0, a },
@@ -82,7 +90,14 @@
Status NEGEMMLowpMatrixMultiplyCore::validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, const ITensorInfo *output, const GEMMInfo &gemm_info)
{
- return cpu::CpuGemmLowpMatrixMultiplyCore::validate(a, b, c, output, gemm_info);
+ // Make the B matrix dynamic values.
+ auto b_info_to_use = b->clone();
+ if(!gemm_info.reshape_b_only_on_first_run())
+ {
+ b_info_to_use->set_are_values_constant(false);
+ }
+
+ return cpu::CpuGemmLowpMatrixMultiplyCore::validate(a, b_info_to_use.get(), c, output, gemm_info);
}
void NEGEMMLowpMatrixMultiplyCore::run()