From 41f6259a4b54d2dc58a4040e942b37f00d306b7b Mon Sep 17 00:00:00 2001
From: Chris Lattner <sabre@nondot.org>
Date: Tue, 29 Apr 2008 04:29:54 +0000
Subject: [PATCH] add support for multiple return values in inline asm. This is
 a step towards PR2094.  It now compiles the attached .ll file to:

_sad16_sse2:
	movslq	%ecx, %rax
	## InlineAsm Start
	%ecx %rdx %rax %rax %r8d %rdx %rsi
	## InlineAsm End
	## InlineAsm Start
	set %eax
	## InlineAsm End
	ret

which is pretty decent for a 3 output, 4 input asm.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@50386 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp | 31 ++++++++++++-------
 test/CodeGen/X86/inline-asm-mrv.ll            | 17 ++++++++++
 2 files changed, 36 insertions(+), 12 deletions(-)
 create mode 100644 test/CodeGen/X86/inline-asm-mrv.ll

diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
index 2d65a331457..a0deb8151e6 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
@@ -176,6 +176,15 @@ namespace {
       }
     }
     
+    /// append - Add the specified values to this one.
+    void append(const RegsForValue &RHS) {
+      TLI = RHS.TLI;
+      ValueVTs.append(RHS.ValueVTs.begin(), RHS.ValueVTs.end());
+      RegVTs.append(RHS.RegVTs.begin(), RHS.RegVTs.end());
+      Regs.append(RHS.Regs.begin(), RHS.Regs.end());
+    }
+    
+    
     /// getCopyFromRegs - Emit a series of CopyFromReg nodes that copies from
     /// this value and returns the result as a ValueVTs value.  This uses 
     /// Chain/Flag as the input and updates them for the output Chain/Flag.
@@ -3754,7 +3763,6 @@ GetRegistersForValue(SDISelAsmOperandInfo &OpInfo, bool HasEarlyClobber,
   }
   
   // Otherwise, we couldn't allocate enough registers for this.
-  return;
 }
 
 
@@ -3938,7 +3946,7 @@ void SelectionDAGLowering::visitInlineAsm(CallSite CS) {
   // Loop over all of the inputs, copying the operand values into the
   // appropriate registers and processing the output regs.
   RegsForValue RetValRegs;
-  
+ 
   // IndirectStoresToEmit - The set of stores to emit after the inline asm node.
   std::vector<std::pair<RegsForValue, Value*> > IndirectStoresToEmit;
   
@@ -3970,15 +3978,16 @@ void SelectionDAGLowering::visitInlineAsm(CallSite CS) {
         exit(1);
       }
 
-      if (!OpInfo.isIndirect) {
-        // This is the result value of the call.
-        assert(RetValRegs.Regs.empty() &&
-               "Cannot have multiple output constraints yet!");
-        assert(CS.getType() != Type::VoidTy && "Bad inline asm!");
-        RetValRegs = OpInfo.AssignedRegs;
-      } else {
+      // If this is an indirect operand, store through the pointer after the
+      // asm.
+      if (OpInfo.isIndirect) {
         IndirectStoresToEmit.push_back(std::make_pair(OpInfo.AssignedRegs,
                                                       OpInfo.CallOperandVal));
+      } else {
+        // This is the result value of the call.
+        assert(CS.getType() != Type::VoidTy && "Bad inline asm!");
+        // Concatenate this output onto the outputs list.
+        RetValRegs.append(OpInfo.AssignedRegs);
       }
       
       // Add information to the INLINEASM node to know that this register is
@@ -4115,9 +4124,7 @@ void SelectionDAGLowering::visitInlineAsm(CallSite CS) {
     // width/num elts.  Make sure to convert it to the right type with
     // bit_convert.
     if (MVT::isVector(Val.getValueType())) {
-      const VectorType *VTy = cast<VectorType>(CS.getType());
-      MVT::ValueType DesiredVT = TLI.getValueType(VTy);
-      
+      MVT::ValueType DesiredVT = TLI.getValueType(CS.getType());
       Val = DAG.getNode(ISD::BIT_CONVERT, DesiredVT, Val);
     }
     
diff --git a/test/CodeGen/X86/inline-asm-mrv.ll b/test/CodeGen/X86/inline-asm-mrv.ll
new file mode 100644
index 00000000000..1d83a16ef21
--- /dev/null
+++ b/test/CodeGen/X86/inline-asm-mrv.ll
@@ -0,0 +1,17 @@
+; PR2094
+; RUN: llvm-as < %s | llc -march=x86-64 | grep movslq
+; RUN: llvm-as < %s | llc -march=x86-64 | not grep movq
+
+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"
+target triple = "x86_64-apple-darwin8"
+
+define i32 @sad16_sse2(i8* %v, i8* %blk2, i8* %blk1, i32 %stride, i32 %h) nounwind  {
+entry:
+	%tmp12 = sext i32 %stride to i64		; <i64> [#uses=1]
+	%mrv = call {i32, i8*, i8*} asm sideeffect "$0 $1 $2 $3 $4 $5 $6",
+         "=r,=r,=r,r,r,r,r"( i64 %tmp12, i32 %h, i8* %blk1, i8* %blk2 ) nounwind
+        %tmp6 = getresult {i32, i8*, i8*} %mrv, 0
+	%tmp7 = call i32 asm sideeffect "set $0",
+             "=r,~{dirflag},~{fpsr},~{flags}"( ) nounwind
+	ret i32 %tmp7
+}
-- 
2.34.1