Merging r324581:

------------------------------------------------------------------------
r324581 | sjoerdmeijer | 2018-02-08 00:39:05 -0800 (Thu, 08 Feb 2018) | 12 lines

[AArch64] Don't materialize 0 with "fmov h0, .." when FullFP16 is not supported

We were generating "fmov h0, wzr" instructions when FullFP16 is not enabled.
I've not added any tests, because the problem was visible in:
test/CodeGen/AArch64/arm64-zero-cycle-zeroing.ll,
which I had to change: I don't think Cyclone has FullFP16 enabled
by default, so it shouldn't be using this v8.2a instruction.

I've also removed these rdar tags, please shout if there are any objections.

Differential Revision: https://reviews.llvm.org/D43020

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_60@332655 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index 6b46f94abe40f7146ca6a81e76c9d764b78d6de1..233d6be247c2c8da6671ab5e47a11d460f038a02 100644 (file)
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -4930,7 +4930,8 @@ bool AArch64TargetLowering::isOffsetFoldingLegal(
 bool AArch64TargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
   // We can materialize #0.0 as fmov $Rd, XZR for 64-bit and 32-bit cases.
   // FIXME: We should be able to handle f128 as well with a clever lowering.
-  if (Imm.isPosZero() && (VT == MVT::f16 || VT == MVT::f64 || VT == MVT::f32)) {
+  if (Imm.isPosZero() && (VT == MVT::f64 || VT == MVT::f32 ||
+                          (VT == MVT::f16 && Subtarget->hasFullFP16()))) {
     DEBUG(dbgs() << "Legal fp imm: materialize 0 using the zero register\n");
     return true;
   }

diff --git a/lib/Target/AArch64/AArch64InstrInfo.td b/lib/Target/AArch64/AArch64InstrInfo.td
index 79826ca2ed8dc6f3d0f111508f3cb638fe2f4281..040011d858e771021341f098d0f852310aba87f1 100644 (file)
--- a/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/lib/Target/AArch64/AArch64InstrInfo.td
@@ -2713,7 +2713,7 @@ defm FMOV : UnscaledConversion<"fmov">;
 // Add pseudo ops for FMOV 0 so we can mark them as isReMaterializable
 let isReMaterializable = 1, isCodeGenOnly = 1, isAsCheapAsAMove = 1 in {
 def FMOVH0 : Pseudo<(outs FPR16:$Rd), (ins), [(set f16:$Rd, (fpimm0))]>,
-    Sched<[WriteF]>;
+    Sched<[WriteF]>, Requires<[HasFullFP16]>;
 def FMOVS0 : Pseudo<(outs FPR32:$Rd), (ins), [(set f32:$Rd, (fpimm0))]>,
     Sched<[WriteF]>;
 def FMOVD0 : Pseudo<(outs FPR64:$Rd), (ins), [(set f64:$Rd, (fpimm0))]>,

diff --git a/test/CodeGen/AArch64/arm64-zero-cycle-zeroing.ll b/test/CodeGen/AArch64/arm64-zero-cycle-zeroing.ll
index 3acb1892dd96dd5cb537c59b0898dbc0bb46b774..664078fb7e94202c43a7282f96fa1e822b33658a 100644 (file)
--- a/test/CodeGen/AArch64/arm64-zero-cycle-zeroing.ll
+++ b/test/CodeGen/AArch64/arm64-zero-cycle-zeroing.ll
@@ -1,12 +1,10 @@
 ; RUN: llc -mtriple=arm64-apple-ios   -mcpu=cyclone   < %s | FileCheck %s -check-prefixes=ALL,CYCLONE
+; RUN: llc -mtriple=arm64-apple-ios   -mcpu=cyclone -mattr=+fullfp16 < %s | FileCheck %s -check-prefixes=CYCLONE-FULLFP16
 ; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=exynos-m1 < %s | FileCheck %s -check-prefixes=ALL,OTHERS
 ; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=exynos-m3 < %s | FileCheck %s -check-prefixes=ALL,OTHERS
 ; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=kryo      < %s | FileCheck %s -check-prefixes=ALL,OTHERS
 ; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=falkor    < %s | FileCheck %s -check-prefixes=ALL,OTHERS
 
-; rdar://11481771
-; rdar://13713797
-
 declare void @bar(half, float, double, <2 x double>)
 declare void @bari(i32, i32)
 declare void @barl(i64, i64)
@@ -16,11 +14,14 @@ define void @t1() nounwind ssp {
 entry:
 ; ALL-LABEL: t1:
 ; ALL-NOT: fmov
-; CYCLONE: fmov h0, wzr
+; ALL:     ldr h0,{{.*}}
 ; CYCLONE: fmov s1, wzr
 ; CYCLONE: fmov d2, xzr
 ; CYCLONE: movi.16b v3, #0
-; OTHERS: movi v{{[0-3]+}}.2d, #0000000000000000
+; CYCLONE-FULLFP16: fmov h0, wzr
+; CYCLONE-FULLFP16: fmov s1, wzr
+; CYCLONE-FULLFP16: fmov d2, xzr
+; CYCLONE-FULLFP16: movi.16b v3, #0
 ; OTHERS: movi v{{[0-3]+}}.2d, #0000000000000000
 ; OTHERS: movi v{{[0-3]+}}.2d, #0000000000000000
 ; OTHERS: movi v{{[0-3]+}}.2d, #0000000000000000
@@ -53,6 +54,8 @@ define void @t4() nounwind ssp {
 ; ALL-NOT: fmov
 ; CYCLONE: fmov s{{[0-3]+}}, wzr
 ; CYCLONE: fmov s{{[0-3]+}}, wzr
+; CYCLONE-FULLFP16: fmov s{{[0-3]+}}, wzr
+; CYCLONE-FULLFP16: fmov s{{[0-3]+}}, wzr
 ; OTHERS: movi v{{[0-3]+}}.2d, #0000000000000000
 ; OTHERS: movi v{{[0-3]+}}.2d, #0000000000000000
   tail call void @barf(float 0.000000e+00, float 0.000000e+00) nounwind