commit 81d15be76923ec23b86e72bbc2d3a2a6542a325e
author Shalini Salomi Bodapati <shalini.salomi.bodapati@intel.com> Thu May 30 11:00:42 2019 +0530
committer Vladimir Marko <vmarko@google.com> Wed Jun 05 16:06:49 2019 +0000
tree 57bf31c2951cbeebddd72c94a415711423c88f88
parent 5fd673488c25991ecc85d26c2dc54415773eb407

Add AVX support for packed add/sub instructions on x86

Test: ./test.py --host, test-art-host-gtest

Change-Id: I48d05e6f6befd54657d962119a543b27a8a51d71
Signed-off-by: Shalini Salomi Bodapati <shalini.salomi.bodapati@intel.com>

compiler/optimizing/loop_optimization.cc

diff --git a/compiler/optimizing/loop_optimization.cc b/compiler/optimizing/loop_optimization.cc
index 6c76ab8..c6e7560 100644
--- a/compiler/optimizing/loop_optimization.cc
+++ b/compiler/optimizing/loop_optimization.cc
@@ -351,8 +351,11 @@
 
 // Translates vector operation to reduction kind.
 static HVecReduce::ReductionKind GetReductionKind(HVecOperation* reduction) {
-  if (reduction->IsVecAdd() ||
+  if (reduction->IsVecAdd()  ||
       reduction->IsVecSub() ||
+      #if defined(ART_ENABLE_CODEGEN_x86) || defined(ART_ENABLE_CODEGEN_x86_64)
+      reduction->IsVecAvxSub() || reduction->IsVecAvxAdd() ||
+      #endif
       reduction->IsVecSADAccumulate() ||
       reduction->IsVecDotProd()) {
     return HVecReduce::kSum;
@@ -1940,10 +1943,34 @@
         new (global_allocator_) HVecCnv(global_allocator_, opa, type, vector_length_, dex_pc),
         new (global_allocator_) HTypeConversion(org_type, opa, dex_pc));
     case HInstruction::kAdd:
+      #if defined(ART_ENABLE_CODEGEN_x86) || defined(ART_ENABLE_CODEGEN_x86_64)
+      if ((compiler_options_->GetInstructionSet() == InstructionSet::kX86 ||
+           compiler_options_->GetInstructionSet() == InstructionSet::kX86_64) &&
+           compiler_options_->GetInstructionSetFeatures()->AsX86InstructionSetFeatures()
+               ->HasAVX2()) {
+        GENERATE_VEC(
+          new (global_allocator_) HVecAvxAdd(
+                                      global_allocator_, opa, opb, type, vector_length_, dex_pc),
+          new (global_allocator_) HAdd(org_type, opa, opb, dex_pc));
+        UNREACHABLE();  // GENERATE_VEC ends with a "break".
+      }
+      #endif
       GENERATE_VEC(
         new (global_allocator_) HVecAdd(global_allocator_, opa, opb, type, vector_length_, dex_pc),
         new (global_allocator_) HAdd(org_type, opa, opb, dex_pc));
     case HInstruction::kSub:
+      #if defined(ART_ENABLE_CODEGEN_x86) || defined(ART_ENABLE_CODEGEN_x86_64)
+      if ((compiler_options_->GetInstructionSet() == InstructionSet::kX86 ||
+           compiler_options_->GetInstructionSet() == InstructionSet::kX86_64) &&
+           compiler_options_->GetInstructionSetFeatures()->AsX86InstructionSetFeatures()
+               ->HasAVX2()) {
+        GENERATE_VEC(
+          new (global_allocator_) HVecAvxSub(
+                                      global_allocator_, opa, opb, type, vector_length_, dex_pc),
+          new (global_allocator_) HSub(org_type, opa, opb, dex_pc));
+        UNREACHABLE();  // GENERATE_VEC ends with a "break".
+      }
+      #endif
       GENERATE_VEC(
         new (global_allocator_) HVecSub(global_allocator_, opa, opb, type, vector_length_, dex_pc),
         new (global_allocator_) HSub(org_type, opa, opb, dex_pc));