commit c0b6f76561580414f08633a804fc548ccad65659
author Georgios Pinitas <georgios.pinitas@arm.com> Mon Nov 02 01:37:17 2020 +0000
committer Georgios Pinitas <georgios.pinitas@arm.com> Thu Nov 12 15:59:25 2020 +0000
tree 4d46b7f479de04f799e29095392948aeb370c029
parent 824061d9910ebb42cbe46b677c0b843db212c9a2

COMPMID-3776: Indirect GEMM

Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I51a1b0f098bc3a8c408c50c92221e4df3061e12c
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4343
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

src/runtime/NEON/functions/NEGEMM.cpp

diff --git a/src/runtime/NEON/functions/NEGEMM.cpp b/src/runtime/NEON/functions/NEGEMM.cpp
index 0215098..9f52e45 100644
--- a/src/runtime/NEON/functions/NEGEMM.cpp
+++ b/src/runtime/NEON/functions/NEGEMM.cpp
@@ -47,7 +47,19 @@
 
 namespace arm_compute
 {
-NEGEMM::~NEGEMM() = default;
+namespace
+{
+AsmGemmInfo init_assembly_metadata(const GEMMInfo &info)
+{
+    AsmGemmInfo asm_info;
+    asm_info.method                  = AsmConvMethod::Im2Col;
+    asm_info.reinterpret_input_as_3d = info.reinterpret_input_as_3d();
+    asm_info.depth_output_gemm3d     = info.depth_output_gemm3d();
+    asm_info.activation_info         = info.activation_info();
+
+    return asm_info;
+}
+} // namespace
 
 NEGEMM::NEGEMM(std::shared_ptr<IMemoryManager> memory_manager, IWeightsManager *weights_manager)
     : _memory_group(memory_manager), _weights_manager(weights_manager), _interleave_kernel(), _transpose_kernel(), _mm_kernel(), _asm_glue(memory_manager, weights_manager), _ma_kernel(),
@@ -56,12 +68,15 @@
 {
 }
 
+NEGEMM::~NEGEMM() = default;
+
 void NEGEMM::configure(const ITensor *a, const ITensor *b, const ITensor *c, ITensor *d, float alpha, float beta, const GEMMInfo &gemm_info)
 {
     ARM_COMPUTE_ERROR_THROW_ON(NEGEMM::validate(a->info(), b->info(), (c != nullptr) ? c->info() : nullptr, d->info(), alpha, beta, gemm_info));
 
-    const bool is_c_bias     = gemm_info.reshape_b_only_on_first_run();
-    bool       run_optimised = bool(NEGEMMAssemblyDispatch::validate(a->info(), b->info(), (is_c_bias && c != nullptr) ? c->info() : nullptr, d->info(), gemm_info));
+    const AsmGemmInfo asm_info      = init_assembly_metadata(gemm_info);
+    const bool        is_c_bias     = gemm_info.reshape_b_only_on_first_run();
+    bool              run_optimised = bool(NEGEMMAssemblyDispatch::validate(a->info(), b->info(), (is_c_bias && c != nullptr) ? c->info() : nullptr, d->info(), asm_info));
 
     // Check if we need to reshape the matrix B only on the first run
     _is_prepared                      = false;
@@ -76,7 +91,7 @@
     if(run_optimised)
     {
         const ITensor *c_to_use = is_c_bias ? c : nullptr;
-        _asm_glue.configure(a, b, c_to_use, d, gemm_info);
+        _asm_glue.configure(a, b, c_to_use, d, asm_info);
         ARM_COMPUTE_ERROR_ON(!_asm_glue.is_configured());
 
         // Scale product by alpha
@@ -221,7 +236,8 @@
     }
 
     // Check if we need to run the optimized assembly kernel
-    const bool run_optimised = bool(NEGEMMAssemblyDispatch::validate(a, b, is_c_bias ? c : nullptr, output, gemm_info));
+    AsmGemmInfo asm_info      = init_assembly_metadata(gemm_info);
+    const bool  run_optimised = bool(NEGEMMAssemblyDispatch::validate(a, b, is_c_bias ? c : nullptr, output, asm_info));
 
     if(!run_optimised)
     {