Generate SVE for 80bit stores when possible

Fixes #4126
FEX-Emu · Nov 20, 2024 · 106dc77 · 106dc77
1 parent 4278c48
commit 106dc77
Show file tree

Hide file tree

Showing 4 changed files with 32 additions and 13 deletions.
diff --git a/FEXCore/Source/Interface/Core/JIT/MemoryOps.cpp b/FEXCore/Source/Interface/Core/JIT/MemoryOps.cpp
@@ -1536,6 +1536,17 @@ DEF_OP(StoreMem) {
       str(Src.D(), MemSrc);
       break;
     }
+    case IR::OpSize::f80Bit: {
+      LOGMAN_THROW_A_FMT(HostSupportsSVE128, "80-bit stores are only supported with SVE128 support");
+      const auto MemSrc = GenerateSVEMemOperand(OpSize, MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
+      // We need to generate the predicate register to copy the 10-byte value
+      // FIXME: don't hardcode p5
+      ARMEmitter::PRegister preg = ARMEmitter::PReg::p5;
+      mov(TMP1, 10);
+      whilelt(ARMEmitter::SubRegSize::i8Bit, preg, ARMEmitter::XReg::zr, TMP1);
+      st1b<ARMEmitter::SubRegSize::i8Bit>(Src.Z(), preg, MemSrc);
+      break;
+    }
     case IR::OpSize::i128Bit: {
       str(Src.Q(), MemSrc);
       break;

diff --git a/FEXCore/Source/Interface/IR/PassManager.cpp b/FEXCore/Source/Interface/IR/PassManager.cpp
@@ -70,7 +70,7 @@ void PassManager::AddDefaultPasses(FEXCore::Context::ContextImpl* ctx) {
   FEX_CONFIG_OPT(DisablePasses, O0);
 
   if (!DisablePasses()) {
-    InsertPass(CreateX87StackOptimizationPass());
+    InsertPass(CreateX87StackOptimizationPass(ctx->HostFeatures));
     InsertPass(CreateConstProp(ctx->HostFeatures.SupportsTSOImm9, &ctx->CPUID));
     InsertPass(CreateDeadFlagCalculationEliminination());
   }

diff --git a/FEXCore/Source/Interface/IR/Passes.h b/FEXCore/Source/Interface/IR/Passes.h
@@ -5,6 +5,7 @@
 
 namespace FEXCore {
 class CPUIDEmu;
+struct HostFeatures;
 }
 
 namespace FEXCore::Utils {
@@ -19,7 +20,7 @@ class RegisterAllocationData;
 fextl::unique_ptr<FEXCore::IR::Pass> CreateConstProp(bool SupportsTSOImm9, const FEXCore::CPUIDEmu* CPUID);
 fextl::unique_ptr<FEXCore::IR::Pass> CreateDeadFlagCalculationEliminination();
 fextl::unique_ptr<FEXCore::IR::RegisterAllocationPass> CreateRegisterAllocationPass();
-fextl::unique_ptr<FEXCore::IR::Pass> CreateX87StackOptimizationPass();
+fextl::unique_ptr<FEXCore::IR::Pass> CreateX87StackOptimizationPass(const FEXCore::HostFeatures&);
 
 namespace Validation {
   fextl::unique_ptr<FEXCore::IR::Pass> CreateIRValidation();

diff --git a/FEXCore/Source/Interface/IR/Passes/x87StackOptimizationPass.cpp b/FEXCore/Source/Interface/IR/Passes/x87StackOptimizationPass.cpp
@@ -3,9 +3,10 @@
 #include "Interface/IR/IR.h"
 #include "Interface/IR/IREmitter.h"
 #include "Interface/IR/PassManager.h"
-#include <FEXCore/IR/IR.h>
-#include <FEXCore/Utils/Profiler.h>
-#include <FEXCore/fextl/deque.h>
+#include "FEXCore/IR/IR.h"
+#include "FEXCore/Utils/Profiler.h"
+#include "FEXCore/fextl/deque.h"
+#include "FEXCore/Core/HostFeatures.h"
 
 #include <array>
 #include <cstddef>
@@ -146,13 +147,15 @@ class FixedSizeStack {
 
 class X87StackOptimization final : public Pass {
 public:
-  X87StackOptimization() {
+  X87StackOptimization(const FEXCore::HostFeatures& Features)
+    : Features(Features) {
     FEX_CONFIG_OPT(ReducedPrecision, X87REDUCEDPRECISION);
     ReducedPrecisionMode = ReducedPrecision;
   }
   void Run(IREmitter* Emit) override;
 
 private:
+  const FEXCore::HostFeatures& Features;
   bool ReducedPrecisionMode;
 
   // Helpers
@@ -820,11 +823,15 @@ void X87StackOptimization::Run(IREmitter* Emit) {
               StackNode = IREmit->_F80CVT(Op->StoreSize, StackNode);
             }
             if (Op->StoreSize == OpSize::f80Bit) { // Part of code from StoreResult_WithOpSize()
-              // For X87 extended doubles, split before storing
-              IREmit->_StoreMem(FPRClass, OpSize::i64Bit, AddrNode, StackNode);
-              auto Upper = IREmit->_VExtractToGPR(OpSize::i128Bit, OpSize::i64Bit, StackNode, 1);
-              auto DestAddr = IREmit->_Add(OpSize::i64Bit, AddrNode, GetConstant(8));
-              IREmit->_StoreMem(GPRClass, OpSize::i16Bit, DestAddr, Upper, OpSize::i64Bit);
+              if (Features.SupportsSVE128) {
+                IREmit->_StoreMem(FPRClass, OpSize::f80Bit, AddrNode, StackNode);
+              } else {
+                // For X87 extended doubles, split before storing
+                IREmit->_StoreMem(FPRClass, OpSize::i64Bit, AddrNode, StackNode);
+                auto Upper = IREmit->_VExtractToGPR(OpSize::i128Bit, OpSize::i64Bit, StackNode, 1);
+                auto DestAddr = IREmit->_Add(OpSize::i64Bit, AddrNode, GetConstant(8));
+                IREmit->_StoreMem(GPRClass, OpSize::i16Bit, DestAddr, Upper, OpSize::i64Bit);
+              }
             } else {
               IREmit->_StoreMem(FPRClass, Op->StoreSize, AddrNode, StackNode);
             }
@@ -1025,7 +1032,7 @@ void X87StackOptimization::Run(IREmitter* Emit) {
   return;
 }
 
-fextl::unique_ptr<Pass> CreateX87StackOptimizationPass() {
-  return fextl::make_unique<X87StackOptimization>();
+fextl::unique_ptr<Pass> CreateX87StackOptimizationPass(const FEXCore::HostFeatures& Features) {
+  return fextl::make_unique<X87StackOptimization>(Features);
 }
 } // namespace FEXCore::IR