[llvm-commits] [llvm] r81993 - in /llvm/trunk: lib/Transforms/Scalar/InstructionCombining.cpp test/Transforms/InstCombine/phi-merge-gep.ll
Dan Gohman
gohman at apple.com
Tue Sep 15 19:02:10 PDT 2009
Author: djg
Date: Tue Sep 15 21:01:52 2009
New Revision: 81993
URL: http://llvm.org/viewvc/llvm-project?rev=81993&view=rev
Log:
Don't sink gep operators through phi nodes if the result would require
more than one phi, since that leads to higher register pressure on
entry to the phi. This is especially problematic when the phi is in
a loop header, as it increases register pressure throughout the loop.
Added:
llvm/trunk/test/Transforms/InstCombine/phi-merge-gep.ll
Modified:
llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=81993&r1=81992&r2=81993&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Tue Sep 15 21:01:52 2009
@@ -326,7 +326,7 @@
// instruction. Instead, visit methods should return the value returned by
// this function.
Instruction *EraseInstFromFunction(Instruction &I) {
- DEBUG(errs() << "IC: erase " << I);
+ DEBUG(errs() << "IC: erase " << I << '\n');
assert(I.use_empty() && "Cannot erase instruction that is used!");
// Make sure that we reprocess all operands now that we reduced their
@@ -10401,6 +10401,10 @@
// This is true if all GEP bases are allocas and if all indices into them are
// constants.
bool AllBasePointersAreAllocas = true;
+
+ // We don't want to replace this phi if the replacement would require
+ // more than one phi.
+ bool NeededPhi = false;
// Scan to see if all operands are the same opcode, all have one use, and all
// kill their operands (i.e. the operands have one use).
@@ -10432,7 +10436,16 @@
if (FirstInst->getOperand(op)->getType() !=GEP->getOperand(op)->getType())
return 0;
+
+ // If we already needed a PHI for an earlier operand, and another operand
+ // also requires a PHI, we'd be introducing more PHIs than we're
+ // eliminating, which increases register pressure on entry to the PHI's
+ // block.
+ if (NeededPhi)
+ return 0;
+
FixedOperands[op] = 0; // Needs a PHI.
+ NeededPhi = true;
}
}
Added: llvm/trunk/test/Transforms/InstCombine/phi-merge-gep.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/phi-merge-gep.ll?rev=81993&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/phi-merge-gep.ll (added)
+++ llvm/trunk/test/Transforms/InstCombine/phi-merge-gep.ll Tue Sep 15 21:01:52 2009
@@ -0,0 +1,101 @@
+; RUN: opt < %s -S -instcombine > %t
+; RUN: grep {= getelementptr} %t | count 20
+; RUN: grep {= phi} %t | count 13
+
+; Don't push the geps through these phis; they have multiple uses!
+
+define void @foo(float* %Ar, float* %Ai, i64 %As, float* %Cr, float* %Ci, i64 %Cs, i64 %n) nounwind {
+entry:
+ %0 = getelementptr inbounds float* %Ar, i64 0 ; <float*> [#uses=1]
+ %1 = getelementptr inbounds float* %Ai, i64 0 ; <float*> [#uses=1]
+ %2 = mul i64 %n, %As ; <i64> [#uses=1]
+ %3 = getelementptr inbounds float* %Ar, i64 %2 ; <float*> [#uses=1]
+ %4 = mul i64 %n, %As ; <i64> [#uses=1]
+ %5 = getelementptr inbounds float* %Ai, i64 %4 ; <float*> [#uses=1]
+ %6 = mul i64 %n, 2 ; <i64> [#uses=1]
+ %7 = mul i64 %6, %As ; <i64> [#uses=1]
+ %8 = getelementptr inbounds float* %Ar, i64 %7 ; <float*> [#uses=1]
+ %9 = mul i64 %n, 2 ; <i64> [#uses=1]
+ %10 = mul i64 %9, %As ; <i64> [#uses=1]
+ %11 = getelementptr inbounds float* %Ai, i64 %10 ; <float*> [#uses=1]
+ %12 = getelementptr inbounds float* %Cr, i64 0 ; <float*> [#uses=1]
+ %13 = getelementptr inbounds float* %Ci, i64 0 ; <float*> [#uses=1]
+ %14 = mul i64 %n, %Cs ; <i64> [#uses=1]
+ %15 = getelementptr inbounds float* %Cr, i64 %14 ; <float*> [#uses=1]
+ %16 = mul i64 %n, %Cs ; <i64> [#uses=1]
+ %17 = getelementptr inbounds float* %Ci, i64 %16 ; <float*> [#uses=1]
+ %18 = mul i64 %n, 2 ; <i64> [#uses=1]
+ %19 = mul i64 %18, %Cs ; <i64> [#uses=1]
+ %20 = getelementptr inbounds float* %Cr, i64 %19 ; <float*> [#uses=1]
+ %21 = mul i64 %n, 2 ; <i64> [#uses=1]
+ %22 = mul i64 %21, %Cs ; <i64> [#uses=1]
+ %23 = getelementptr inbounds float* %Ci, i64 %22 ; <float*> [#uses=1]
+ br label %bb13
+
+bb: ; preds = %bb13
+ %24 = load float* %A0r.0, align 4 ; <float> [#uses=1]
+ %25 = load float* %A0i.0, align 4 ; <float> [#uses=1]
+ %26 = load float* %A1r.0, align 4 ; <float> [#uses=2]
+ %27 = load float* %A1i.0, align 4 ; <float> [#uses=2]
+ %28 = load float* %A2r.0, align 4 ; <float> [#uses=2]
+ %29 = load float* %A2i.0, align 4 ; <float> [#uses=2]
+ %30 = fadd float %26, %28 ; <float> [#uses=2]
+ %31 = fadd float %27, %29 ; <float> [#uses=2]
+ %32 = fsub float %26, %28 ; <float> [#uses=1]
+ %33 = fsub float %27, %29 ; <float> [#uses=1]
+ %34 = fadd float %24, %30 ; <float> [#uses=2]
+ %35 = fadd float %25, %31 ; <float> [#uses=2]
+ %36 = fmul float %30, -1.500000e+00 ; <float> [#uses=1]
+ %37 = fmul float %31, -1.500000e+00 ; <float> [#uses=1]
+ %38 = fadd float %34, %36 ; <float> [#uses=2]
+ %39 = fadd float %35, %37 ; <float> [#uses=2]
+ %40 = fmul float %32, 0x3FEBB67AE0000000 ; <float> [#uses=2]
+ %41 = fmul float %33, 0x3FEBB67AE0000000 ; <float> [#uses=2]
+ %42 = fadd float %38, %41 ; <float> [#uses=1]
+ %43 = fsub float %39, %40 ; <float> [#uses=1]
+ %44 = fsub float %38, %41 ; <float> [#uses=1]
+ %45 = fadd float %39, %40 ; <float> [#uses=1]
+ store float %34, float* %C0r.0, align 4
+ store float %35, float* %C0i.0, align 4
+ store float %42, float* %C1r.0, align 4
+ store float %43, float* %C1i.0, align 4
+ store float %44, float* %C2r.0, align 4
+ store float %45, float* %C2i.0, align 4
+ %46 = getelementptr inbounds float* %A0r.0, i64 %As ; <float*> [#uses=1]
+ %47 = getelementptr inbounds float* %A0i.0, i64 %As ; <float*> [#uses=1]
+ %48 = getelementptr inbounds float* %A1r.0, i64 %As ; <float*> [#uses=1]
+ %49 = getelementptr inbounds float* %A1i.0, i64 %As ; <float*> [#uses=1]
+ %50 = getelementptr inbounds float* %A2r.0, i64 %As ; <float*> [#uses=1]
+ %51 = getelementptr inbounds float* %A2i.0, i64 %As ; <float*> [#uses=1]
+ %52 = getelementptr inbounds float* %C0r.0, i64 %Cs ; <float*> [#uses=1]
+ %53 = getelementptr inbounds float* %C0i.0, i64 %Cs ; <float*> [#uses=1]
+ %54 = getelementptr inbounds float* %C1r.0, i64 %Cs ; <float*> [#uses=1]
+ %55 = getelementptr inbounds float* %C1i.0, i64 %Cs ; <float*> [#uses=1]
+ %56 = getelementptr inbounds float* %C2r.0, i64 %Cs ; <float*> [#uses=1]
+ %57 = getelementptr inbounds float* %C2i.0, i64 %Cs ; <float*> [#uses=1]
+ %58 = add nsw i64 %i.0, 1 ; <i64> [#uses=1]
+ br label %bb13
+
+bb13: ; preds = %bb, %entry
+ %i.0 = phi i64 [ 0, %entry ], [ %58, %bb ] ; <i64> [#uses=2]
+ %C2i.0 = phi float* [ %23, %entry ], [ %57, %bb ] ; <float*> [#uses=2]
+ %C2r.0 = phi float* [ %20, %entry ], [ %56, %bb ] ; <float*> [#uses=2]
+ %C1i.0 = phi float* [ %17, %entry ], [ %55, %bb ] ; <float*> [#uses=2]
+ %C1r.0 = phi float* [ %15, %entry ], [ %54, %bb ] ; <float*> [#uses=2]
+ %C0i.0 = phi float* [ %13, %entry ], [ %53, %bb ] ; <float*> [#uses=2]
+ %C0r.0 = phi float* [ %12, %entry ], [ %52, %bb ] ; <float*> [#uses=2]
+ %A2i.0 = phi float* [ %11, %entry ], [ %51, %bb ] ; <float*> [#uses=2]
+ %A2r.0 = phi float* [ %8, %entry ], [ %50, %bb ] ; <float*> [#uses=2]
+ %A1i.0 = phi float* [ %5, %entry ], [ %49, %bb ] ; <float*> [#uses=2]
+ %A1r.0 = phi float* [ %3, %entry ], [ %48, %bb ] ; <float*> [#uses=2]
+ %A0i.0 = phi float* [ %1, %entry ], [ %47, %bb ] ; <float*> [#uses=2]
+ %A0r.0 = phi float* [ %0, %entry ], [ %46, %bb ] ; <float*> [#uses=2]
+ %59 = icmp slt i64 %i.0, %n ; <i1> [#uses=1]
+ br i1 %59, label %bb, label %bb14
+
+bb14: ; preds = %bb13
+ br label %return
+
+return: ; preds = %bb14
+ ret void
+}
More information about the llvm-commits
mailing list