[AArch64] Handle lowering lround on windows, where long is 32 bit

[mstorsjo]

Summary

Currently, building code with lround for windows with -fno-math-errno yields this error:

fatal error: error in backend: Cannot select: t2: i32 = lround ConstantFP:f32<1.000000e+02>

t1: f32 = ConstantFP<1.000000e+02>

[efriedma]

LGTM

llvm/trunk/lib/Target/AArch64/AArch64InstrInfo.td

 defm : FPToIntegerPats<fp_to_uint, fceil,  "FCVTPU">;
 defm : FPToIntegerPats<fp_to_sint, ffloor, "FCVTMS">;
 defm : FPToIntegerPats<fp_to_uint, ffloor, "FCVTMU">;
 defm : FPToIntegerPats<fp_to_sint, ftrunc, "FCVTZS">;
 defm : FPToIntegerPats<fp_to_uint, ftrunc, "FCVTZU">;
 defm : FPToIntegerPats<fp_to_sint, fround, "FCVTAS">;
 defm : FPToIntegerPats<fp_to_uint, fround, "FCVTAU">;
 
+def : Pat<(i32 (lround f32:$Rn)),
+          (!cast<Instruction>(FCVTASUWSr) f32:$Rn)>;
+def : Pat<(i32 (lround f64:$Rn)),
+          (!cast<Instruction>(FCVTASUWDr) f64:$Rn)>;
 def : Pat<(i64 (lround f32:$Rn)),
           (!cast<Instruction>(FCVTASUXSr) f32:$Rn)>;
 def : Pat<(i64 (lround f64:$Rn)),
           (!cast<Instruction>(FCVTASUXDr) f64:$Rn)>;
 def : Pat<(i64 (llround f32:$Rn)),
           (!cast<Instruction>(FCVTASUXSr) f32:$Rn)>;
 def : Pat<(i64 (llround f64:$Rn)),
           (!cast<Instruction>(FCVTASUXDr) f64:$Rn)>;

llvm/trunk/test/CodeGen/AArch64/lround-conv-win.ll

+; RUN: llc < %s -mtriple=aarch64-windows -mattr=+neon | FileCheck %s
+
+; CHECK-LABEL: testmsxs:
+; CHECK:       fcvtas  w8, s0
+; CHECK-NEXT:  sxtw    x0, w8
+; CHECK-NEXT:  ret
+define i64 @testmsxs(float %x) {
+entry:
+  %0 = tail call i32 @llvm.lround.i32.f32(float %x)
+  %conv = sext i32 %0 to i64
+  ret i64 %conv
+}
+
+; CHECK-LABEL: testmsws:
+; CHECK:       fcvtas  w0, s0
+; CHECK-NEXT:  ret
+define i32 @testmsws(float %x) {
+entry:
+  %0 = tail call i32 @llvm.lround.i32.f32(float %x)
+  ret i32 %0
+}
+
+; CHECK-LABEL: testmsxd:
+; CHECK:       fcvtas  w8, d0
+; CHECK-NEXT:  sxtw    x0, w8
+; CHECK-NEXT:  ret
+define i64 @testmsxd(double %x) {
+entry:
+  %0 = tail call i32 @llvm.lround.i32.f64(double %x)
+  %conv = sext i32 %0 to i64
+  ret i64 %conv
+}
+
+; CHECK-LABEL: testmswd:
+; CHECK:       fcvtas  w0, d0
+; CHECK-NEXT:  ret
+define i32 @testmswd(double %x) {
+entry:
+  %0 = tail call i32 @llvm.lround.i32.f64(double %x)
+  ret i32 %0
+}
+
+declare i32 @llvm.lround.i32.f32(float) nounwind readnone
+declare i32 @llvm.lround.i32.f64(double) nounwind readnone