From 8854509eac45da4fe3f7843599133aa2a6d1beaf Mon Sep 17 00:00:00 2001
From: Kunal Pathak <Kunal.Pathak@microsoft.com>
Date: Fri, 5 May 2023 20:37:52 -0700
Subject: [PATCH] Skip FP registers processing if method don't have floating
 points use (#85744)

* skip processing floating point registers if method doesn't have one

* Rename to needNonIntegerRegisters

* jit format
---
 src/coreclr/jit/importer.cpp  |  4 ++--
 src/coreclr/jit/lsra.cpp      |  8 ++++++--
 src/coreclr/jit/lsra.h        |  4 +++-
 src/coreclr/jit/lsrabuild.cpp | 14 ++++++++++++++
 src/coreclr/jit/target.h      |  5 -----
 5 files changed, 25 insertions(+), 10 deletions(-)

diff --git a/src/coreclr/jit/importer.cpp b/src/coreclr/jit/importer.cpp
index 1c3fb163d1987..e212041d1bba6 100644
--- a/src/coreclr/jit/importer.cpp
+++ b/src/coreclr/jit/importer.cpp
@@ -66,11 +66,11 @@ void Compiler::impPushOnStack(GenTree* tree, typeInfo ti)
     verCurrentState.esStack[verCurrentState.esStackDepth].seTypeInfo = ti;
     verCurrentState.esStack[verCurrentState.esStackDepth++].val      = tree;
 
-    if ((tree->gtType == TYP_LONG) && (compLongUsed == false))
+    if (tree->gtType == TYP_LONG)
     {
         compLongUsed = true;
     }
-    else if (((tree->gtType == TYP_FLOAT) || (tree->gtType == TYP_DOUBLE)) && (compFloatingPointUsed == false))
+    else if ((tree->gtType == TYP_FLOAT) || (tree->gtType == TYP_DOUBLE))
     {
         compFloatingPointUsed = true;
     }
diff --git a/src/coreclr/jit/lsra.cpp b/src/coreclr/jit/lsra.cpp
index d803ae8fcaff5..617c6016bc5a5 100644
--- a/src/coreclr/jit/lsra.cpp
+++ b/src/coreclr/jit/lsra.cpp
@@ -418,6 +418,8 @@ void LinearScan::updateRegsFreeBusyState(RefPosition& refPosition,
 
 regMaskTP LinearScan::internalFloatRegCandidates()
 {
+    needNonIntegerRegisters = true;
+
     if (compiler->compFloatingPointUsed)
     {
         return availableFloatRegs;
@@ -686,13 +688,15 @@ LinearScan::LinearScan(Compiler* theCompiler)
     , refPositions(theCompiler->getAllocator(CMK_LSRA_RefPosition))
     , listNodePool(theCompiler)
 {
+    availableRegCount       = ACTUAL_REG_COUNT;
+    needNonIntegerRegisters = false;
+
 #if defined(TARGET_XARCH)
-    availableRegCount = ACTUAL_REG_COUNT;
 
 #if defined(TARGET_AMD64)
     rbmAllFloat       = compiler->rbmAllFloat;
     rbmFltCalleeTrash = compiler->rbmFltCalleeTrash;
-#endif
+#endif // TARGET_AMD64
 
     if (!compiler->canUseEvexEncoding())
     {
diff --git a/src/coreclr/jit/lsra.h b/src/coreclr/jit/lsra.h
index d869f8f8a126c..a5e79ebfab51c 100644
--- a/src/coreclr/jit/lsra.h
+++ b/src/coreclr/jit/lsra.h
@@ -733,6 +733,7 @@ class LinearScan : public LinearScanInterface
     unsigned int currentSpill[TYP_COUNT];
     bool         needFloatTmpForFPCall;
     bool         needDoubleTmpForFPCall;
+    bool         needNonIntegerRegisters;
 
 #ifdef DEBUG
 private:
@@ -2041,13 +2042,14 @@ class LinearScan : public LinearScanInterface
     }
 #endif // TARGET_AMD64
 
+#endif // TARGET_XARCH
+
     unsigned availableRegCount;
 
     unsigned get_AVAILABLE_REG_COUNT() const
     {
         return this->availableRegCount;
     }
-#endif // TARGET_XARCH
 
     //------------------------------------------------------------------------
     // calleeSaveRegs: Get the set of callee-save registers of the given RegisterType
diff --git a/src/coreclr/jit/lsrabuild.cpp b/src/coreclr/jit/lsrabuild.cpp
index 0f11acdadb701..2e97c8c4bc7da 100644
--- a/src/coreclr/jit/lsrabuild.cpp
+++ b/src/coreclr/jit/lsrabuild.cpp
@@ -1940,6 +1940,14 @@ void LinearScan::buildPhysRegRecords()
         RegRecord* curr = &physRegs[reg];
         curr->regOrder  = (unsigned char)i;
     }
+
+    // TODO-CQ: We build physRegRecords before building intervals
+    // and refpositions. During building intervals/refposition, we
+    // would know if there are floating points used. If we can know
+    // that information before we build intervals, we can skip
+    // initializing the floating registers.
+    // For that `compFloatingPointUsed` should be set accurately
+    // before invoking allocator.
     for (unsigned int i = 0; i < lsraRegOrderFltSize; i++)
     {
         regNumber  reg  = lsraRegOrderFlt[i];
@@ -2727,6 +2735,12 @@ void LinearScan::buildIntervals()
         RefPosition* pos = newRefPosition((Interval*)nullptr, currentLoc, RefTypeBB, nullptr, RBM_NONE);
     }
 
+    needNonIntegerRegisters |= compiler->compFloatingPointUsed;
+    if (!needNonIntegerRegisters)
+    {
+        availableRegCount = REG_INT_COUNT;
+    }
+
 #ifdef DEBUG
     // Make sure we don't have any blocks that were not visited
     for (BasicBlock* const block : compiler->Blocks())
diff --git a/src/coreclr/jit/target.h b/src/coreclr/jit/target.h
index ad7e5be605199..4e385d323a7d8 100644
--- a/src/coreclr/jit/target.h
+++ b/src/coreclr/jit/target.h
@@ -196,12 +196,7 @@ enum _regMask_enum : unsigned
 #error Unsupported target architecture
 #endif
 
-#if defined(TARGET_XARCH)
-// AVAILABLE_REG_COUNT is defined to be dynamic, based on whether AVX-512 high registers are available.
 #define AVAILABLE_REG_COUNT get_AVAILABLE_REG_COUNT()
-#else
-#define AVAILABLE_REG_COUNT ACTUAL_REG_COUNT
-#endif
 
 /*****************************************************************************/