From: Hal Finkel <hfinkel@anl.gov>
Date: Sat, 3 Aug 2013 12:25:10 +0000 (+0000)
Subject: Fix PPC64 64-bit GPR inline asm constraint matching
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=b176acb6b72e12f301e0ea9c9e8664f5ba6de610;p=platform%2Fupstream%2Fllvm.git

Fix PPC64 64-bit GPR inline asm constraint matching

Internally, the PowerPC backend names the 32-bit GPRs R[0-9]+, and names the
64-bit parent GPRs X[0-9]+. When matching inline assembly constraints with
explicit register names, on PPC64 when an i64 MVT has been requested, we need
to follow gcc's convention of using r[0-9]+ to refer to the 64-bit (parent)
registers.

At some point, we'll probably want to arrange things so that the generic code
in TargetLowering uses the AsmName fields declared in *RegisterInfo.td in order
to match these inline asm register constraints. If we do that, this change can
be reverted.

llvm-svn: 187693
---

diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index aadab82..664dd12 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -7632,7 +7632,24 @@ PPCTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
     }
   }
 
-  return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
+  std::pair<unsigned, const TargetRegisterClass*> R =
+    TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
+
+  // r[0-9]+ are used, on PPC64, to refer to the corresponding 64-bit registers
+  // (which we call X[0-9]+). If a 64-bit value has been requested, and a
+  // 32-bit GPR has been selected, then 'upgrade' it to the 64-bit parent
+  // register.
+  // FIXME: If TargetLowering::getRegForInlineAsmConstraint could somehow use
+  // the AsmName field from *RegisterInfo.td, then this would not be necessary.
+  if (R.first && VT == MVT::i64 && PPCSubTarget.isPPC64() &&
+      PPC::GPRCRegClass.contains(R.first)) {
+    const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
+    return std::make_pair(TRI->getMatchingSuperReg(R.first,
+                            PPC::sub_32, &PPC::GPRCRegClass),
+                          &PPC::G8RCRegClass);
+  }
+
+  return R;
 }
 
 
diff --git a/llvm/test/CodeGen/PowerPC/inlineasm-i64-reg.ll b/llvm/test/CodeGen/PowerPC/inlineasm-i64-reg.ll
new file mode 100644
index 0000000..fa9aa45
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/inlineasm-i64-reg.ll
@@ -0,0 +1,65 @@
+; RUN: llc -mtriple=powerpc64-bgq-linux -mcpu=a2 < %s | FileCheck %s
+target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
+target triple = "powerpc64-bgq-linux"
+
+%struct.BG_CoordinateMapping_t = type { [4 x i8] }
+
+; Function Attrs: alwaysinline inlinehint nounwind
+define zeroext i32 @Kernel_RanksToCoords(i64 %mapsize, %struct.BG_CoordinateMapping_t* %map, i64* %numentries) #0 {
+entry:
+  %mapsize.addr = alloca i64, align 8
+  %map.addr = alloca %struct.BG_CoordinateMapping_t*, align 8
+  %numentries.addr = alloca i64*, align 8
+  %r0 = alloca i64, align 8
+  %r3 = alloca i64, align 8
+  %r4 = alloca i64, align 8
+  %r5 = alloca i64, align 8
+  %tmp = alloca i64, align 8
+  store i64 %mapsize, i64* %mapsize.addr, align 8
+  store %struct.BG_CoordinateMapping_t* %map, %struct.BG_CoordinateMapping_t** %map.addr, align 8
+  store i64* %numentries, i64** %numentries.addr, align 8
+  store i64 1055, i64* %r0, align 8
+  %0 = load i64* %mapsize.addr, align 8
+  store i64 %0, i64* %r3, align 8
+  %1 = load %struct.BG_CoordinateMapping_t** %map.addr, align 8
+  %2 = ptrtoint %struct.BG_CoordinateMapping_t* %1 to i64
+  store i64 %2, i64* %r4, align 8
+  %3 = load i64** %numentries.addr, align 8
+  %4 = ptrtoint i64* %3 to i64
+  store i64 %4, i64* %r5, align 8
+  %5 = load i64* %r0, align 8
+  %6 = load i64* %r3, align 8
+  %7 = load i64* %r4, align 8
+  %8 = load i64* %r5, align 8
+  %9 = call { i64, i64, i64, i64 } asm sideeffect "sc", "={r0},={r3},={r4},={r5},{r0},{r3},{r4},{r5},~{r6},~{r7},~{r8},~{r9},~{r10},~{r11},~{r12},~{cr0},~{memory}"(i64 %5, i64 %6, i64 %7, i64 %8) #1, !srcloc !0
+
+; CHECK-LABEL: @Kernel_RanksToCoords
+
+; These need to be 64-bit loads, not 32-bit loads (not lwz).
+; CHECK-NOT: lwz
+
+; CHECK: #APP
+; CHECK: sc
+; CHECK: #NO_APP
+
+; CHECK: blr
+
+  %asmresult = extractvalue { i64, i64, i64, i64 } %9, 0
+  %asmresult1 = extractvalue { i64, i64, i64, i64 } %9, 1
+  %asmresult2 = extractvalue { i64, i64, i64, i64 } %9, 2
+  %asmresult3 = extractvalue { i64, i64, i64, i64 } %9, 3
+  store i64 %asmresult, i64* %r0, align 8
+  store i64 %asmresult1, i64* %r3, align 8
+  store i64 %asmresult2, i64* %r4, align 8
+  store i64 %asmresult3, i64* %r5, align 8
+  %10 = load i64* %r3, align 8
+  store i64 %10, i64* %tmp
+  %11 = load i64* %tmp
+  %conv = trunc i64 %11 to i32
+  ret i32 %conv
+}
+
+attributes #0 = { alwaysinline inlinehint nounwind }
+attributes #1 = { nounwind }
+
+!0 = metadata !{i32 -2146895770}