Merging r204155:

------------------------------------------------------------------------
r204155 | wschmidt | 2014-03-18 10:32:50 -0400 (Tue, 18 Mar 2014) | 16 lines

Fix PR19144: Incorrect offset generated for int-to-fp conversion at -O0.

When converting a signed 32-bit integer to double-precision floating point on
hardware without a lfiwax instruction, we have to instead use a lfd followed
by fcfid.  We were erroneously offsetting the address by 4 bytes in
preparation for either a lfiwax or lfiwzx when generating the lfd.  This fixes
that silly error.

This was not caught in the test suite since the conversion tests were run with
-mcpu=pwr7, which implies availability of lfiwax.  I've added another test
case for older hardware that checks the code we expect in the absence of
lfiwax and other flavors of fcfid.  There are fewer tests in this test case
because we punt to DAG selection in more cases on older hardware.  (We must
generate complex fiddly sequences in those cases, and there is marginal
benefit in duplicating that logic in fast-isel.)

------------------------------------------------------------------------

llvm-svn: 205824

2 files changed, 158 insertions, 3 deletions

diff --git a/llvm/lib/Target/PowerPC/PPCFastISel.cpp b/llvm/lib/Target/PowerPC/PPCFastISel.cpp
index 09117e7..4e3b0b8 100644
--- a/llvm/lib/Target/PowerPC/PPCFastISel.cpp
+++ b/llvm/lib/Target/PowerPC/PPCFastISel.cpp
@@ -892,11 +892,13 @@ unsigned PPCFastISel::PPCMoveToFPReg(MVT SrcVT, unsigned SrcReg,
   unsigned LoadOpc = PPC::LFD;
 
   if (SrcVT == MVT::i32) {
-    Addr.Offset = 4;
-    if (!IsSigned)
+    if (!IsSigned) {
       LoadOpc = PPC::LFIWZX;
-    else if (PPCSubTarget.hasLFIWAX())
+      Addr.Offset = 4;
+    } else if (PPCSubTarget.hasLFIWAX()) {
       LoadOpc = PPC::LFIWAX;
+      Addr.Offset = 4;
+    }
   }
 
   const TargetRegisterClass *RC = &PPC::F8RCRegClass;
diff --git a/llvm/test/CodeGen/PowerPC/fast-isel-conversion-p5.ll b/llvm/test/CodeGen/PowerPC/fast-isel-conversion-p5.ll
new file mode 100644
index 0000000..db0d8ed
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/fast-isel-conversion-p5.ll
@@ -0,0 +1,153 @@
+; RUN: llc < %s -O0 -verify-machineinstrs -fast-isel-abort -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr5 | FileCheck %s --check-prefix=ELF64
+
+; Test sitofp
+
+define void @sitofp_double_i32(i32 %a, double %b) nounwind ssp {
+entry:
+; ELF64: sitofp_double_i32
+  %b.addr = alloca double, align 8
+  %conv = sitofp i32 %a to double
+; ELF64: std {{[0-9]+}}, -[[OFFSET:[0-9]+]](1)
+; ELF64: lfd {{[0-9]+}}, -[[OFFSET]](1)
+; ELF64: fcfid
+  store double %conv, double* %b.addr, align 8
+  ret void
+}
+
+define void @sitofp_double_i64(i64 %a, double %b) nounwind ssp {
+entry:
+; ELF64: sitofp_double_i64
+  %b.addr = alloca double, align 8
+  %conv = sitofp i64 %a to double
+; ELF64: std {{[0-9]+}}, -[[OFFSET:[0-9]+]](1)
+; ELF64: lfd {{[0-9]+}}, -[[OFFSET]](1)
+; ELF64: fcfid
+  store double %conv, double* %b.addr, align 8
+  ret void
+}
+
+define void @sitofp_double_i16(i16 %a, double %b) nounwind ssp {
+entry:
+; ELF64: sitofp_double_i16
+  %b.addr = alloca double, align 8
+  %conv = sitofp i16 %a to double
+; ELF64: extsh
+; ELF64: std {{[0-9]+}}, -[[OFFSET:[0-9]+]](1)
+; ELF64: lfd {{[0-9]+}}, -[[OFFSET]](1)
+; ELF64: fcfid
+  store double %conv, double* %b.addr, align 8
+  ret void
+}
+
+define void @sitofp_double_i8(i8 %a, double %b) nounwind ssp {
+entry:
+; ELF64: sitofp_double_i8
+  %b.addr = alloca double, align 8
+  %conv = sitofp i8 %a to double
+; ELF64: extsb
+; ELF64: std {{[0-9]+}}, -[[OFFSET:[0-9]+]](1)
+; ELF64: lfd {{[0-9]+}}, -[[OFFSET]](1)
+; ELF64: fcfid
+  store double %conv, double* %b.addr, align 8
+  ret void
+}
+
+; Test fptosi
+
+define void @fptosi_float_i32(float %a) nounwind ssp {
+entry:
+; ELF64: fptosi_float_i32
+  %b.addr = alloca i32, align 4
+  %conv = fptosi float %a to i32
+; ELF64: fctiwz
+; ELF64: stfd
+; ELF64: lwa
+  store i32 %conv, i32* %b.addr, align 4
+  ret void
+}
+
+define void @fptosi_float_i64(float %a) nounwind ssp {
+entry:
+; ELF64: fptosi_float_i64
+  %b.addr = alloca i64, align 4
+  %conv = fptosi float %a to i64
+; ELF64: fctidz
+; ELF64: stfd
+; ELF64: ld
+  store i64 %conv, i64* %b.addr, align 4
+  ret void
+}
+
+define void @fptosi_double_i32(double %a) nounwind ssp {
+entry:
+; ELF64: fptosi_double_i32
+  %b.addr = alloca i32, align 8
+  %conv = fptosi double %a to i32
+; ELF64: fctiwz
+; ELF64: stfd
+; ELF64: lwa
+  store i32 %conv, i32* %b.addr, align 8
+  ret void
+}
+
+define void @fptosi_double_i64(double %a) nounwind ssp {
+entry:
+; ELF64: fptosi_double_i64
+  %b.addr = alloca i64, align 8
+  %conv = fptosi double %a to i64
+; ELF64: fctidz
+; ELF64: stfd
+; ELF64: ld
+  store i64 %conv, i64* %b.addr, align 8
+  ret void
+}
+
+; Test fptoui
+
+define void @fptoui_float_i32(float %a) nounwind ssp {
+entry:
+; ELF64: fptoui_float_i32
+  %b.addr = alloca i32, align 4
+  %conv = fptoui float %a to i32
+; ELF64: fctidz
+; ELF64: stfd
+; ELF64: lwz
+  store i32 %conv, i32* %b.addr, align 4
+  ret void
+}
+
+define void @fptoui_float_i64(float %a) nounwind ssp {
+entry:
+; ELF64: fptoui_float_i64
+  %b.addr = alloca i64, align 4
+  %conv = fptoui float %a to i64
+; ELF64: fctiduz
+; ELF64: stfd
+; ELF64: ld
+  store i64 %conv, i64* %b.addr, align 4
+  ret void
+}
+
+define void @fptoui_double_i32(double %a) nounwind ssp {
+entry:
+; ELF64: fptoui_double_i32
+  %b.addr = alloca i32, align 8
+  %conv = fptoui double %a to i32
+; ELF64: fctidz
+; ELF64: stfd
+; ELF64: lwz
+  store i32 %conv, i32* %b.addr, align 8
+  ret void
+}
+
+define void @fptoui_double_i64(double %a) nounwind ssp {
+entry:
+; ELF64: fptoui_double_i64
+  %b.addr = alloca i64, align 8
+  %conv = fptoui double %a to i64
+; ELF64: fctiduz
+; ELF64: stfd
+; ELF64: ld
+  store i64 %conv, i64* %b.addr, align 8
+  ret void
+}