commit 5aa1a0b7ca5eed010e4b297a95b1c4851f741328
author Georgios Pinitas <georgios.pinitas@arm.com> Thu Jul 02 20:02:20 2020 +0100
committer Georgios Pinitas <georgios.pinitas@arm.com> Mon Jul 06 16:51:32 2020 +0000
tree ba882de9e86589dfdd33937d538a89bbdf01c40e
parent 42550c039105597ff6acd4e5efc0ee3c7c20b08e

COMPID-3324: Clean GEMM kernels

Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I170de1671e061a78740caee31fb4a1b8642c1369
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3505
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>

src/runtime/CPUUtils.cpp

diff --git a/src/runtime/CPUUtils.cpp b/src/runtime/CPUUtils.cpp
index d8f01a9..e874f0f 100644
--- a/src/runtime/CPUUtils.cpp
+++ b/src/runtime/CPUUtils.cpp
@@ -73,6 +73,7 @@
     {
         case CPUModel::GENERIC_FP16_DOT:
         case CPUModel::A55r1:
+        case CPUModel::X1:
             return true;
         default:
             return false;
@@ -86,6 +87,7 @@
         case CPUModel::GENERIC_FP16:
         case CPUModel::GENERIC_FP16_DOT:
         case CPUModel::A55r1:
+        case CPUModel::X1:
             return true;
         default:
             return false;
@@ -121,6 +123,9 @@
                     model = CPUModel::A55r0;
                 }
                 break;
+            case 0xd44: // X1
+                model = CPUModel::X1;
+                break;
             case 0xd0a: // A75
                 if(variant != 0)
                 {

src/runtime/NEON/functions/NEGEMMAssemblyDispatch.cpp

diff --git a/src/runtime/NEON/functions/NEGEMMAssemblyDispatch.cpp b/src/runtime/NEON/functions/NEGEMMAssemblyDispatch.cpp
index 1fcac58..8a2506f 100644
--- a/src/runtime/NEON/functions/NEGEMMAssemblyDispatch.cpp
+++ b/src/runtime/NEON/functions/NEGEMMAssemblyDispatch.cpp
@@ -284,7 +284,7 @@
     //if we disable this code below in brackets then ConvLayer deadlocks when threads > 1 and
     //the shapes are In=1x1x1024 Weights=1x1x1024x1001 Biases=1001 Out=1x1x1001
     {
-        const unsigned int window_size = get_total_window_size(*_gemm_kernel_asm);
+        const unsigned int window_size = _gemm_kernel_asm->get_window_size().total_size();
         if(window_size < static_cast<unsigned int>(args._maxthreads))
         {
             _gemm_kernel_asm->set_nthreads(window_size);
@@ -408,7 +408,7 @@
     if(_workspace.buffer() != nullptr)
     {
         _gemm_kernel_asm->set_working_space(reinterpret_cast<void *>(_workspace.buffer()));
-        const unsigned int window_size = get_total_window_size(*_gemm_kernel_asm);
+        const unsigned int window_size = _gemm_kernel_asm->get_window_size().total_size();
         unsigned int       num_threads = NEScheduler::get().num_threads();
         if(window_size < num_threads)
         {