Add support for passing i1, i8, and i16 call parameters. Also, be sure to

zero-extend the constant integer encoding.  Test case provides testing for
both call parameters and materialization of i1, i8, and i16 types.

llvm-svn: 143821

lib/Target/ARM/ARMFastISel.cpp

@@ -557,7 +557,7 @@ unsigned ARMFastISel::ARMMaterializeInt(const Constant *C, EVT VT) {
     unsigned ImmReg = createResultReg(TLI.getRegClassFor(SrcVT));
     AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                             TII.get(Opc), ImmReg)
-                    .addImm(CI->getSExtValue()));
+                    .addImm(CI->getZExtValue()));
     return ImmReg;
   }
 
@@ -1599,33 +1599,21 @@ bool ARMFastISel::ProcessCallArgs(SmallVectorImpl<Value*> &Args,
     switch (VA.getLocInfo()) {
       case CCValAssign::Full: break;
       case CCValAssign::SExt: {
-        bool Emitted = FastEmitExtend(ISD::SIGN_EXTEND, VA.getLocVT(),
+        EVT DestVT = VA.getLocVT();
-                                         Arg, ArgVT, Arg);
+        unsigned ResultReg = ARMEmitIntExt(ArgVT, Arg, DestVT,
-        assert(Emitted && "Failed to emit a sext!"); (void)Emitted;
+                                           /*isZExt*/false);
-        Emitted = true;
+        assert (ResultReg != 0 && "Failed to emit a sext");
-        ArgVT = VA.getLocVT();
+        Arg = ResultReg;
         break;
       }
+      case CCValAssign::AExt:
+        // Intentional fall-through.  Handle AExt and ZExt.
       case CCValAssign::ZExt: {
-        bool Emitted = FastEmitExtend(ISD::ZERO_EXTEND, VA.getLocVT(),
+        EVT DestVT = VA.getLocVT();
-                                         Arg, ArgVT, Arg);
+        unsigned ResultReg = ARMEmitIntExt(ArgVT, Arg, DestVT,
-        assert(Emitted && "Failed to emit a zext!"); (void)Emitted;
+                                           /*isZExt*/true);
-        Emitted = true;
+        assert (ResultReg != 0 && "Failed to emit a sext");
-        ArgVT = VA.getLocVT();
+        Arg = ResultReg;
-        break;
-      }
-      case CCValAssign::AExt: {
-        bool Emitted = FastEmitExtend(ISD::ANY_EXTEND, VA.getLocVT(),
-                                         Arg, ArgVT, Arg);
-        if (!Emitted)
-          Emitted = FastEmitExtend(ISD::ZERO_EXTEND, VA.getLocVT(),
-                                      Arg, ArgVT, Arg);
-        if (!Emitted)
-          Emitted = FastEmitExtend(ISD::SIGN_EXTEND, VA.getLocVT(),
-                                      Arg, ArgVT, Arg);
-
-        assert(Emitted && "Failed to emit a aext!"); (void)Emitted;
-        ArgVT = VA.getLocVT();
         break;
       }
       case CCValAssign::BCvt: {
@@ -1643,7 +1631,7 @@ bool ARMFastISel::ProcessCallArgs(SmallVectorImpl<Value*> &Args,
     if (VA.isRegLoc() && !VA.needsCustom()) {
       BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
               VA.getLocReg())
-      .addReg(Arg);
+        .addReg(Arg);
       RegArgs.push_back(VA.getLocReg());
     } else if (VA.needsCustom()) {
       // TODO: We need custom lowering for vector (v2f64) args.
@@ -1962,8 +1950,8 @@ bool ARMFastISel::SelectCall(const Instruction *I) {
 
     Type *ArgTy = (*i)->getType();
     MVT ArgVT;
-    // FIXME: Should be able to handle i1, i8, and/or i16 parameters.
+    if (!isTypeLegal(ArgTy, ArgVT) && ArgVT != MVT::i16 && ArgVT != MVT::i8 &&
-    if (!isTypeLegal(ArgTy, ArgVT))
+        ArgVT != MVT::i1)
       return false;
     unsigned OriginalAlignment = TD.getABITypeAlignment(ArgTy);
     Flags.setOrigAlign(OriginalAlignment);

test/CodeGen/ARM/fast-isel-call.ll

@@ -0,0 +1,67 @@
+; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=armv7-apple-darwin | FileCheck %s --check-prefix=ARM
+; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=thumbv7-apple-darwin | FileCheck %s --check-prefix=THUMB
+
+define i32 @t0(i1 zeroext %a) nounwind {
+  %1 = zext i1 %a to i32
+  ret i32 %1
+}
+
+define i32 @t1(i8 signext %a) nounwind {
+  %1 = sext i8 %a to i32
+  ret i32 %1
+}
+
+define i32 @t2(i8 zeroext %a) nounwind {
+  %1 = zext i8 %a to i32
+  ret i32 %1
+}
+
+define i32 @t3(i16 signext %a) nounwind {
+  %1 = sext i16 %a to i32
+  ret i32 %1
+}
+
+define i32 @t4(i16 zeroext %a) nounwind {
+  %1 = zext i16 %a to i32
+  ret i32 %1
+}
+
+define void @foo(i8 %a, i16 %b) nounwind {
+; ARM: foo
+; THUMB: foo
+;; Materialize i1 1
+; ARM: movw r2, #1
+;; zero-ext
+; ARM: and r2, r2, #1
+; THUMB: and r2, r2, #1
+  %1 = call i32 @t0(i1 zeroext 1)
+; ARM: sxtb	r2, r1
+; ARM: mov r0, r2
+; THUMB: sxtb	r2, r1
+; THUMB: mov r0, r2
+  %2 = call i32 @t1(i8 signext %a)
+; ARM: uxtb	r2, r1
+; ARM: mov r0, r2
+; THUMB: uxtb	r2, r1
+; THUMB: mov r0, r2
+  %3 = call i32 @t2(i8 zeroext %a)
+; ARM: sxth	r2, r1
+; ARM: mov r0, r2
+; THUMB: sxth	r2, r1
+; THUMB: mov r0, r2
+  %4 = call i32 @t3(i16 signext %b)
+; ARM: uxth	r2, r1
+; ARM: mov r0, r2
+; THUMB: uxth	r2, r1
+; THUMB: mov r0, r2
+  %5 = call i32 @t4(i16 zeroext %b)
+
+;; A few test to check materialization
+;; Note: i1 1 was materialized with t1 call
+; ARM: movw r1, #255
+%6 = call i32 @t2(i8 zeroext 255)
+; ARM: movw r1, #65535
+; THUMB: movw r1, #65535
+%7 = call i32 @t4(i16 zeroext 65535)
+  ret void
+}