Add a GR32_NOREX_NOSP register class and fix a bug where getMatchingSuperRegClass()

was saying that the matching superregister class of GR32_NOREX in GR64_NOREX_NOSP
is GR64_NOREX, which drops the NOSP constraint. This fixes PR10032.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132225 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86RegisterInfo.cpp b/lib/Target/X86/X86RegisterInfo.cpp
index fc1b8cf08c1dcf5e6a171f6305adce0669e6c4c7..ebd0d055c65ef5e6844a49d06149a15f08e8eb38 100644 (file)
--- a/lib/Target/X86/X86RegisterInfo.cpp
+++ b/lib/Target/X86/X86RegisterInfo.cpp
@@ -301,9 +301,16 @@ X86RegisterInfo::getMatchingSuperRegClass(const TargetRegisterClass *A,
           A == &X86::GR64_NOREX_NOSPRegClass)
         return &X86::GR64_ABCDRegClass;
     } else if (B == &X86::GR32_NOREXRegClass) {
+      if (A == &X86::GR64RegClass || A == &X86::GR64_NOREXRegClass)
+        return &X86::GR64_NOREXRegClass;
+      else if (A == &X86::GR64_NOSPRegClass || A == &X86::GR64_NOREX_NOSPRegClass)
+        return &X86::GR64_NOREX_NOSPRegClass;
+      else if (A == &X86::GR64_ABCDRegClass)
+        return &X86::GR64_ABCDRegClass;
+    } else if (B == &X86::GR32_NOREX_NOSPRegClass) {
       if (A == &X86::GR64RegClass || A == &X86::GR64_NOREXRegClass ||
           A == &X86::GR64_NOSPRegClass || A == &X86::GR64_NOREX_NOSPRegClass)
-        return &X86::GR64_NOREXRegClass;
+        return &X86::GR64_NOREX_NOSPRegClass;
       else if (A == &X86::GR64_ABCDRegClass)
         return &X86::GR64_ABCDRegClass;
     }

diff --git a/lib/Target/X86/X86RegisterInfo.td b/lib/Target/X86/X86RegisterInfo.td
index d2dce1fe676042f877841a6f8bb0ec70a2025eba..a383d3db662dffb44d149f577496db8c1b4f7972 100644 (file)
--- a/lib/Target/X86/X86RegisterInfo.td
+++ b/lib/Target/X86/X86RegisterInfo.td
@@ -601,12 +601,20 @@ def GR64_NOSP : RegisterClass<"X86", [i64], 64,
                        (GR32_NOSP sub_32bit)];
 }
 
+// GR32_NOREX_NOSP - GR32 registers which do not require a REX prefix except
+// ESP.
+def GR32_NOREX_NOSP : RegisterClass<"X86", [i32], 32,
+                               [EAX, ECX, EDX, ESI, EDI, EBX, EBP]> {
+  let SubRegClasses = [(GR8_NOREX sub_8bit, sub_8bit_hi),
+                       (GR16_NOREX sub_16bit)];
+}
+
 // GR64_NOREX_NOSP - GR64_NOREX registers except RSP.
 def GR64_NOREX_NOSP : RegisterClass<"X86", [i64], 64,
                                     [RAX, RCX, RDX, RSI, RDI, RBX, RBP]> {
   let SubRegClasses = [(GR8_NOREX sub_8bit, sub_8bit_hi),
                        (GR16_NOREX sub_16bit),
-                       (GR32_NOREX sub_32bit)];
+                       (GR32_NOREX_NOSP sub_32bit)];
 }
 
 // A class to support the 'A' assembler constraint: EAX then EDX.

diff --git a/test/CodeGen/X86/2011-05-27-CrossClassCoalescing.ll b/test/CodeGen/X86/2011-05-27-CrossClassCoalescing.ll
new file mode 100644 (file)
index 0000000..c595bba
--- /dev/null
+++ b/test/CodeGen/X86/2011-05-27-CrossClassCoalescing.ll
@@ -0,0 +1,41 @@
+; RUN: llc < %s -verify-coalescing
+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-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+target triple = "x86_64-apple-macosx10.6.0"
+
+@bit_count = external constant [256 x i32], align 16
+
+define fastcc void @unate_intersect() nounwind uwtable ssp {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.inc.i
+  br label %do.body.i
+
+do.body.i:                                        ; preds = %do.body.i, %for.body
+  %exitcond149 = icmp eq i64 undef, undef
+  br i1 %exitcond149, label %land.lhs.true, label %do.body.i
+
+land.lhs.true:                                    ; preds = %do.body.i
+  br label %for.body.i
+
+for.body.i:                                       ; preds = %for.inc.i, %if.then
+  %tmp3524.i = phi i32 [ 0, %land.lhs.true ], [ %tmp351.i, %for.inc.i ]
+  %tmp6.i12 = load i32* undef, align 4
+  br i1 undef, label %for.inc.i, label %if.then.i17
+
+if.then.i17:                                      ; preds = %for.body.i
+  %shr.i14 = lshr i32 %tmp6.i12, 8
+  %and14.i = and i32 %shr.i14, 255
+  %idxprom15.i = zext i32 %and14.i to i64
+  %arrayidx16.i = getelementptr inbounds [256 x i32]* @bit_count, i64 0, i64 %idxprom15.i
+  %tmp17.i15 = load i32* %arrayidx16.i, align 4
+  %add.i = add i32 0, %tmp3524.i
+  %add24.i = add i32 %add.i, %tmp17.i15
+  %add31.i = add i32 %add24.i, 0
+  %add33.i = add i32 %add31.i, 0
+  br label %for.inc.i
+
+for.inc.i:                                        ; preds = %if.then.i17, %for.body.i
+  %tmp351.i = phi i32 [ %add33.i, %if.then.i17 ], [ %tmp3524.i, %for.body.i ]
+  br label %for.body.i
+}