[llvm] 529ad40 - [PowerPC] Fix missing kill flag update for XVCVDPSP transformations (#67997)

via llvm-commits llvm-commits at lists.llvm.org
Fri Oct 6 07:24:59 PDT 2023


Author: Lei
Date: 2023-10-06T10:24:54-04:00
New Revision: 529ad40e058b8c4014e9bc27dc2e7e3766acb245

URL: https://github.com/llvm/llvm-project/commit/529ad40e058b8c4014e9bc27dc2e7e3766acb245
DIFF: https://github.com/llvm/llvm-project/commit/529ad40e058b8c4014e9bc27dc2e7e3766acb245.diff

LOG: [PowerPC] Fix missing kill flag update for XVCVDPSP transformations (#67997)

Add transformed register to kill flag work list for XVCVDPSP tranformations.

Ref: reviews.llvm.org/D133103

Added: 
    llvm/test/CodeGen/PowerPC/kill_flag_verification.ll

Modified: 
    llvm/lib/Target/PowerPC/PPCMIPeephole.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/PowerPC/PPCMIPeephole.cpp b/llvm/lib/Target/PowerPC/PPCMIPeephole.cpp
index d7a51891087e413..494e4b52a5b5eb7 100644
--- a/llvm/lib/Target/PowerPC/PPCMIPeephole.cpp
+++ b/llvm/lib/Target/PowerPC/PPCMIPeephole.cpp
@@ -895,8 +895,9 @@ bool PPCMIPeephole::simplifyCode() {
               LLVM_DEBUG(MI.dump());
               LLVM_DEBUG(dbgs() << "Through instruction:\n");
               LLVM_DEBUG(DefMI->dump());
-              RoundInstr->eraseFromParent();
               addRegToUpdate(ConvReg1);
+              addRegToUpdate(FRSPDefines);
+              ToErase = RoundInstr;
             }
           };
 

diff  --git a/llvm/test/CodeGen/PowerPC/kill_flag_verification.ll b/llvm/test/CodeGen/PowerPC/kill_flag_verification.ll
new file mode 100644
index 000000000000000..79cc407fc9be97c
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/kill_flag_verification.ll
@@ -0,0 +1,45 @@
+; RUN: llc < %s -mtriple powerpc64le-unknown-linux-gnu
+; RUN: llc < %s -mtriple powerpc64-unknown-linux-gnu
+; RUN: llc < %s -mtriple powerpc-ibm-aix
+; RUN: llc < %s -mtriple powerpc64-ibm-aix
+
+define void @xvcvdpsp_kill_flag() {
+entry:
+  %call49 = tail call double @sin()
+  %0 = insertelement <2 x double> poison, double %call49, i64 1
+  %1 = fmul <2 x double> %0, zeroinitializer
+  %2 = shufflevector <2 x double> %1, <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
+  %3 = insertelement <4 x double> %2, double 0.000000e+00, i64 2
+  %4 = insertelement <4 x double> %3, double poison, i64 3
+  %5 = fptrunc <4 x double> %4 to <4 x float>
+  %6 = shufflevector <4 x float> %5, <4 x float> poison, <4 x i32> <i32 0, i32 0, i32 1, i32 1>
+  %7 = shufflevector <4 x float> %5, <4 x float> poison, <4 x i32> <i32 2, i32 2, i32 3, i32 3>
+  %8 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %7, <4 x float> <float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00>, <4 x float> zeroinitializer)
+  br label %if.end1
+
+if.end1:                                         ; preds = %entry
+  br i1 poison, label %for.cond1.preheader, label %if.then2
+
+for.cond1.preheader:                            ; preds = %if.end1
+  br label %for.body2.preheader
+
+for.body2.preheader:                            ; preds = %for.cond1.preheader
+  br i1 poison, label %for.loopexit, label %for.body3
+
+for.body3:                                      ; preds = %for.body2.preheader
+  %9 = tail call <4 x float> @llvm.ppc.fnmsub.v4f32(<4 x float> zeroinitializer, <4 x float> %6, <4 x float> zeroinitializer)
+  %10 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>, <4 x float> %8, <4 x float> %9)
+  %11 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %10, <4 x float> zeroinitializer, <4 x float> zeroinitializer)
+  store <4 x float> %11, ptr poison, align 16
+  unreachable
+
+for.loopexit:                    ; preds = %for.body2.preheader
+  unreachable
+
+if.then2:                                       ; preds = %if.end1
+  ret void
+}
+
+declare double @sin() local_unnamed_addr #0
+declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>)
+declare <4 x float> @llvm.ppc.fnmsub.v4f32(<4 x float>, <4 x float>, <4 x float>)


        


More information about the llvm-commits mailing list