Refine EarlyClobber assert in register scavenger.

It is legal for an inline asm operand to use an earlyclobber register if the
use operand is tied to the earlyclobber operand. The issue is discussed here:

  http://gcc.gnu.org/ml/gcc/1999-04n/msg00431.html

We should perhaps let only the machine code verifier worry about these finer
details. EarlyClobber operands are not really interesting to the scavenger.

This fixes PR4528 for the third time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79122 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/RegisterScavenging.cpp b/lib/CodeGen/RegisterScavenging.cpp
index 323d3db39a3da90529a2bae19ab867f332272859..a82562838ec99589bd23df1d98e4f4dae306c11d 100644 (file)
--- a/lib/CodeGen/RegisterScavenging.cpp
+++ b/lib/CodeGen/RegisterScavenging.cpp
@@ -236,7 +236,7 @@ void RegScavenger::forward() {
       continue;
     if (MO.isUse()) {
       assert(isUsed(Reg) && "Using an undefined register!");
-      assert(!EarlyClobberRegs.test(Reg) &&
+      assert((!EarlyClobberRegs.test(Reg) || MI->isRegTiedToDefOperand(i)) &&
              "Using an early clobbered register!");
     } else {
       assert(MO.isDef());

diff --git a/test/CodeGen/ARM/2009-08-15-RegScavenger-EarlyClobber.ll b/test/CodeGen/ARM/2009-08-15-RegScavenger-EarlyClobber.ll
new file mode 100644 (file)
index 0000000..15fa145
--- /dev/null
+++ b/test/CodeGen/ARM/2009-08-15-RegScavenger-EarlyClobber.ll
@@ -0,0 +1,42 @@
+; RUN: llvm-as < %s | llc -march=arm
+; PR4528
+
+; Inline asm is allowed to contain operands "=&r", "0".
+
+%struct.device_dma_parameters = type { i32, i32 }
+%struct.iovec = type { i8*, i32 }
+
+define arm_aapcscc i32 @generic_segment_checks(%struct.iovec* nocapture %iov, i32* nocapture %nr_segs, i32* nocapture %count, i32 %access_flags) nounwind optsize {
+entry:
+  br label %bb8
+
+bb:                                               ; preds = %bb8
+  br i1 undef, label %bb10, label %bb2
+
+bb2:                                              ; preds = %bb
+  %asmtmp = tail call %struct.device_dma_parameters asm "adds $1, $2, $3; sbcccs $1, $1, $0; movcc $0, #0", "=&r,=&r,r,Ir,0,~{cc}"(i8* undef, i32 undef, i32 0) nounwind; <%struct.device_dma_parameters> [#uses=1]
+  %asmresult = extractvalue %struct.device_dma_parameters %asmtmp, 0; <i32> [#uses=1]
+  %0 = icmp eq i32 %asmresult, 0                  ; <i1> [#uses=1]
+  br i1 %0, label %bb7, label %bb4
+
+bb4:                                              ; preds = %bb2
+  br i1 undef, label %bb10, label %bb9
+
+bb7:                                              ; preds = %bb2
+  %1 = add i32 %2, 1                              ; <i32> [#uses=1]
+  br label %bb8
+
+bb8:                                              ; preds = %bb7, %entry
+  %2 = phi i32 [ 0, %entry ], [ %1, %bb7 ]        ; <i32> [#uses=3]
+  %scevgep22 = getelementptr %struct.iovec* %iov, i32 %2, i32 0; <i8**> [#uses=0]
+  %3 = load i32* %nr_segs, align 4                ; <i32> [#uses=1]
+  %4 = icmp ult i32 %2, %3                        ; <i1> [#uses=1]
+  br i1 %4, label %bb, label %bb9
+
+bb9:                                              ; preds = %bb8, %bb4
+  store i32 undef, i32* %count, align 4
+  ret i32 0
+
+bb10:                                             ; preds = %bb4, %bb
+  ret i32 0
+}