[PATCH] D43291: [ARM] f16 vcmp fixes

[Sjoerd Meijer via Phabricator via llvm-commits]

SjoerdMeijer created this revision.
SjoerdMeijer added reviewers: samparker, olista01.
Herald added subscribers: kristof.beyls, javed.absar.

This adds a VCMPEZH match rule and fixes the f16 VCMP test cases.


https://reviews.llvm.org/D43291

Files:
  lib/Target/ARM/ARMInstrVFP.td
  test/CodeGen/ARM/fp16-instructions.ll


Index: test/CodeGen/ARM/fp16-instructions.ll
===================================================================
--- test/CodeGen/ARM/fp16-instructions.ll
+++ test/CodeGen/ARM/fp16-instructions.ll
@@ -104,43 +104,47 @@
   %2 = bitcast float %G.coerce to i32
   %tmp1.0.extract.trunc = trunc i32 %2 to i16
   %3 = bitcast i16 %tmp1.0.extract.trunc to half
-  %cmp = fcmp ogt half %1, %3
+  %cmp = fcmp une half %1, %3
   ret i1 %cmp
 
 ; CHECK-LABEL:            VCMP:
 
-; CHECK-SOFT:             bl  __aeabi_fcmpgt
+; CHECK-SOFT:             bl  __aeabi_fcmpeq
 
 ; CHECK-SOFTFP-VFP3:      bl  __aeabi_h2f
 ; CHECK-SOFTFP-VFP3:      bl  __aeabi_h2f
-; CHECK-SOFTFP-VFP3:      vcmpe.f32 s{{.}}, s{{.}}
+; CHECK-SOFTFP-VFP3:      vcmp.f32 s{{.}}, s{{.}}
 
 ; CHECK-SOFTFP-FP16:      vcvtb.f32.f16 s{{.}}, s{{.}}
 ; CHECK-SOFTFP-FP16:      vcvtb.f32.f16 s{{.}}, s{{.}}
-; CHECK-SOFTFP-FP16:      vcmpe.f32 s{{.}}, s{{.}}
+; CHECK-SOFTFP-FP16:      vcmp.f32 s{{.}}, s{{.}}
 
 ; CHECK-SOFTFP-FULLFP16:  vmov.f16  [[S2:s[0-9]]], r0
 ; CHECK-SOFTFP-FULLFP16:  vmov.f16 [[S0:s[0-9]]], r1
-; CHECK-SOFTFP-FULLFP16:  vcmpe.f16 [[S2]], [[S0]]
+; CHECK-SOFTFP-FULLFP16:  vcmp.f16 [[S2]], [[S0]]
 
 ; CHECK-SOFTFP-FULLFP16-NOT:  vmov.f16  s{{.}}, r0
 ; CHECK-SOFTFP-FULLFP16-NOT:  vmov.f16  s{{.}}, r1
-; CHECK-HARDFP-FULLFP16:      vcmpe.f16  s0, s1
+; CHECK-HARDFP-FULLFP16:      vcmp.f16  s0, s1
 }
 
 ; 4. VCMPE
+define i32 @VCMPE_IMM(float %F.coerce) {
+entry:
+  %0 = bitcast float %F.coerce to i32
+  %tmp.0.extract.trunc = trunc i32 %0 to i16
+  %1 = bitcast i16 %tmp.0.extract.trunc to half
+  %tmp = fcmp olt half %1, 0.000000e+00
+  %tmp1 = zext i1 %tmp to i32
+  ret i32 %tmp1
 
-; FIXME: enable when constant pool is fixed
-;
-;define i32 @VCMPE_IMM(float %F.coerce) {
-;entry:
-;  %0 = bitcast float %F.coerce to i32
-;  %tmp.0.extract.trunc = trunc i32 %0 to i16
-;  %1 = bitcast i16 %tmp.0.extract.trunc to half
-;  %tmp = fcmp olt half %1, 1.000000e+00
-;  %tmp1 = zext i1 %tmp to i32
-;  ret i32 %tmp1
-;}
+; CHECK-LABEL:             VCMPE_IMM:
+
+; CHECK-SOFT:              bl  __aeabi_fcmplt
+; CHECK-SOFTFP-FP16:       vcmpe.f32 s0, #0
+; CHECK-SOFTFP-FULLFP16:   vcmpe.f16 s0, #0
+; CHECK-HARDFP-FULLFP16:   vcmpe.f16 s0, #0
+}
 
 define i32 @VCMPE(float %F.coerce, float %G.coerce) {
 entry:
@@ -155,6 +159,11 @@
   ret i32 %tmp1
 
 ; CHECK-LABEL:  VCMPE:
+
+; CHECK-SOFT:              bl  __aeabi_fcmplt
+; CHECK-SOFTFP-FP16:       vcmpe.f32 s{{.}}, s{{.}}
+; CHECK-SOFTFP-FULLFP16:   vcmpe.f16 s{{.}}, s{{.}}
+; CHECK-HARDFP-FULLFP16:   vcmpe.f16 s{{.}}, s{{.}}
 }
 
 ; 5. VCVT (between floating-point and fixed-point)
Index: lib/Target/ARM/ARMInstrVFP.td
===================================================================
--- lib/Target/ARM/ARMInstrVFP.td
+++ lib/Target/ARM/ARMInstrVFP.td
@@ -594,9 +594,9 @@
 }
 
 def VCMPEZH : AHuI<0b11101, 0b11, 0b0101, 0b11, 0,
-                   (outs), (ins SPR:$Sd),
+                   (outs), (ins HPR:$Sd),
                    IIC_fpCMP16, "vcmpe", ".f16\t$Sd, #0",
-                   []> {
+                   [(arm_cmpfp0 HPR:$Sd, (i32 1))]> {
   let Inst{3-0} = 0b0000;
   let Inst{5}   = 0;
 }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D43291.134227.patch
Type: text/x-patch
Size: 3159 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180214/f949664c/attachment.bin>