[llvm] ee44896 - [AArch64] Add missing HasNEON predicate in scalar FABD patterns

[Alex Richardson via llvm-commits]

Author: Alex Richardson
Date: 2022-04-13T09:30:11Z
New Revision: ee44896cf42d3360f7e0229c360258c92262a47b

URL: https://github.com/llvm/llvm-project/commit/ee44896cf42d3360f7e0229c360258c92262a47b
DIFF: https://github.com/llvm/llvm-project/commit/ee44896cf42d3360f7e0229c360258c92262a47b.diff

LOG: [AArch64] Add missing HasNEON predicate in scalar FABD patterns

I was trying to compile with -march=+nosimd and hit the following assertion:
`Attempting to emit FABD64 instruction but the Feature_HasNEON predicate(s) are not met`.
This adds a HasNEON predicate to the patterns which was omitted in commit
21d9b33d62772c58267cc0aa725e35ac9a4661db for some reason.
The new code generation matches GCC with -mcpu=<cpu>+nosimd:
https://godbolt.org/z/n1Y7xh5jo

Differential Revision: https://reviews.llvm.org/D123491

Added: 
    

Modified: 
    llvm/lib/Target/AArch64/AArch64InstrInfo.td
    llvm/test/CodeGen/AArch64/fabd-no-neon.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 8d4a1ed3aabf7..e2a4581201bda 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -4780,11 +4780,13 @@ defm CMTST    : SIMDThreeScalarD<0, 0b10001, "cmtst", AArch64cmtst>;
 defm FABD     : SIMDFPThreeScalar<1, 1, 0b010, "fabd", int_aarch64_sisd_fabd>;
 def : Pat<(v1f64 (int_aarch64_neon_fabd (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
           (FABD64 FPR64:$Rn, FPR64:$Rm)>;
-let Predicates = [HasFullFP16] in {
+let Predicates = [HasNEON, HasFullFP16] in {
 def : Pat<(fabs (fsub f16:$Rn, f16:$Rm)), (FABD16 f16:$Rn, f16:$Rm)>;
 }
+let Predicates = [HasNEON] in {
 def : Pat<(fabs (fsub f32:$Rn, f32:$Rm)), (FABD32 f32:$Rn, f32:$Rm)>;
 def : Pat<(fabs (fsub f64:$Rn, f64:$Rm)), (FABD64 f64:$Rn, f64:$Rm)>;
+}
 defm FACGE    : SIMDThreeScalarFPCmp<1, 0, 0b101, "facge",
                                      int_aarch64_neon_facge>;
 defm FACGT    : SIMDThreeScalarFPCmp<1, 1, 0b101, "facgt",

diff  --git a/llvm/test/CodeGen/AArch64/fabd-no-neon.ll b/llvm/test/CodeGen/AArch64/fabd-no-neon.ll
index 0e484d79f9581..4966ef75b6767 100644
--- a/llvm/test/CodeGen/AArch64/fabd-no-neon.ll
+++ b/llvm/test/CodeGen/AArch64/fabd-no-neon.ll
@@ -5,7 +5,7 @@
 ; RUN: llc -mtriple=aarch64 -mattr=+fullfp16,-neon < %s | FileCheck %s --check-prefix NEON-DISABLED
 ; Note: We need to use -filetype=obj to trigger verifyInstructionPredicates()
 ; checks since it is not called when emitting assembly output.
-; FIXME: llc -mtriple=aarch64 -mattr=+fullfp16,-neon -o /dev/null %s -filetype=obj
+; RUN: llc -mtriple=aarch64 -mattr=+fullfp16,-neon -o /dev/null %s -filetype=obj
 
 declare half @llvm.fabs.f16(half)
 declare float @llvm.fabs.f32(float)
@@ -19,7 +19,8 @@ define half @fabd16(half %f1, half %f2) {
 ;
 ; NEON-DISABLED-LABEL: fabd16:
 ; NEON-DISABLED:       // %bb.0:
-; NEON-DISABLED-NEXT:    fabd h0, h0, h1
+; NEON-DISABLED-NEXT:    fsub h0, h0, h1
+; NEON-DISABLED-NEXT:    fabs h0, h0
 ; NEON-DISABLED-NEXT:    ret
   %sub = fsub half %f1, %f2
   %abs = tail call half @llvm.fabs.f16(half %sub)
@@ -34,7 +35,8 @@ define float @fabd32(float %f1, float %f2) {
 ;
 ; NEON-DISABLED-LABEL: fabd32:
 ; NEON-DISABLED:       // %bb.0:
-; NEON-DISABLED-NEXT:    fabd s0, s0, s1
+; NEON-DISABLED-NEXT:    fsub s0, s0, s1
+; NEON-DISABLED-NEXT:    fabs s0, s0
 ; NEON-DISABLED-NEXT:    ret
   %sub = fsub float %f1, %f2
   %abs = tail call float @llvm.fabs.f32(float %sub)
@@ -49,7 +51,8 @@ define double @fabd64(double %f1, double %f2) {
 ;
 ; NEON-DISABLED-LABEL: fabd64:
 ; NEON-DISABLED:       // %bb.0:
-; NEON-DISABLED-NEXT:    fabd d0, d0, d1
+; NEON-DISABLED-NEXT:    fsub d0, d0, d1
+; NEON-DISABLED-NEXT:    fabs d0, d0
 ; NEON-DISABLED-NEXT:    ret
   %sub = fsub double %f1, %f2
   %abs = tail call double @llvm.fabs.f64(double %sub)