MarkMurrayARM created this revision.
MarkMurrayARM added reviewers: dmgreen, simon_tatham.
Herald added subscribers: llvm-commits, cfe-commits, hiraditya, kristof.beyls.
Herald added projects: clang, LLVM.
Fix VMINNMAQ, VMAXNMAQ intrinsics; BOTH arguments have the absolute values
taken.
Repository:
rG LLVM Github Monorepo
https://reviews.llvm.org/D72830
Files:
clang/include/clang/Basic/arm_mve.td
clang/test/CodeGen/arm-mve-intrinsics/vmaxnmaq.c
clang/test/CodeGen/arm-mve-intrinsics/vminnmaq.c
llvm/lib/Target/ARM/ARMInstrMVE.td
llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmaq.ll
llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmaq.ll
Index: llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmaq.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmaq.ll
+++ llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmaq.ll
@@ -7,9 +7,10 @@
; CHECK-NEXT: vminnma.f16 q0, q1
; CHECK-NEXT: bx lr
entry:
- %0 = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> %b)
- %1 = tail call <8 x half> @llvm.minnum.v8f16(<8 x half> %a, <8 x half> %0)
- ret <8 x half> %1
+ %0 = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> %a)
+ %1 = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> %b)
+ %2 = tail call <8 x half> @llvm.minnum.v8f16(<8 x half> %0, <8 x half> %1)
+ ret <8 x half> %2
}
declare <8 x half> @llvm.fabs.v8f16(<8 x half>) #1
@@ -22,9 +23,10 @@
; CHECK-NEXT: vminnma.f32 q0, q1
; CHECK-NEXT: bx lr
entry:
- %0 = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %b)
- %1 = tail call <4 x float> @llvm.minnum.v4f32(<4 x float> %a, <4 x float> %0)
- ret <4 x float> %1
+ %0 = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %a)
+ %1 = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %b)
+ %2 = tail call <4 x float> @llvm.minnum.v4f32(<4 x float> %0, <4 x float> %1)
+ ret <4 x float> %2
}
declare <4 x float> @llvm.fabs.v4f32(<4 x float>) #1
Index: llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmaq.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmaq.ll
+++ llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmaq.ll
@@ -7,9 +7,10 @@
; CHECK-NEXT: vmaxnma.f16 q0, q1
; CHECK-NEXT: bx lr
entry:
- %0 = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> %b)
- %1 = tail call <8 x half> @llvm.maxnum.v8f16(<8 x half> %a, <8 x half> %0)
- ret <8 x half> %1
+ %0 = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> %a)
+ %1 = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> %b)
+ %2 = tail call <8 x half> @llvm.maxnum.v8f16(<8 x half> %0, <8 x half> %1)
+ ret <8 x half> %2
}
declare <8 x half> @llvm.fabs.v8f16(<8 x half>) #1
@@ -22,9 +23,10 @@
; CHECK-NEXT: vmaxnma.f32 q0, q1
; CHECK-NEXT: bx lr
entry:
- %0 = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %b)
- %1 = tail call <4 x float> @llvm.maxnum.v4f32(<4 x float> %a, <4 x float> %0)
- ret <4 x float> %1
+ %0 = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %a)
+ %1 = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %b)
+ %2 = tail call <4 x float> @llvm.maxnum.v4f32(<4 x float> %0, <4 x float> %1)
+ ret <4 x float> %2
}
declare <4 x float> @llvm.fabs.v4f32(<4 x float>) #1
Index: llvm/lib/Target/ARM/ARMInstrMVE.td
===================================================================
--- llvm/lib/Target/ARM/ARMInstrMVE.td
+++ llvm/lib/Target/ARM/ARMInstrMVE.td
@@ -3655,7 +3655,8 @@
let Predicates = [HasMVEInt] in {
// Unpredicated v(max|min)nma
- def : Pat<(VTI.Vec (unpred_op (VTI.Vec MQPR:$Qd), (fabs (VTI.Vec MQPR:$Qm)))),
+ def : Pat<(VTI.Vec (unpred_op (fabs (VTI.Vec MQPR:$Qd)),
+ (fabs (VTI.Vec MQPR:$Qm)))),
(VTI.Vec (Inst (VTI.Vec MQPR:$Qd), (VTI.Vec MQPR:$Qm)))>;
// Predicated v(max|min)nma
Index: clang/test/CodeGen/arm-mve-intrinsics/vminnmaq.c
===================================================================
--- clang/test/CodeGen/arm-mve-intrinsics/vminnmaq.c
+++ clang/test/CodeGen/arm-mve-intrinsics/vminnmaq.c
@@ -6,9 +6,10 @@
// CHECK-LABEL: @test_vminnmaq_f16(
// CHECK-NEXT: entry:
-// CHECK-NEXT: [[TMP0:%.*]] = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> [[B:%.*]])
-// CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x half> @llvm.minnum.v8f16(<8 x half> [[A:%.*]], <8 x half> [[TMP0]])
-// CHECK-NEXT: ret <8 x half> [[TMP1]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> [[A:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> [[B:%.*]])
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <8 x half> @llvm.minnum.v8f16(<8 x half> [[TMP0]], <8 x half> [[TMP1]])
+// CHECK-NEXT: ret <8 x half> [[TMP2]]
//
float16x8_t test_vminnmaq_f16(float16x8_t a, float16x8_t b)
{
@@ -21,9 +22,10 @@
// CHECK-LABEL: @test_vminnmaq_f32(
// CHECK-NEXT: entry:
-// CHECK-NEXT: [[TMP0:%.*]] = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> [[B:%.*]])
-// CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x float> @llvm.minnum.v4f32(<4 x float> [[A:%.*]], <4 x float> [[TMP0]])
-// CHECK-NEXT: ret <4 x float> [[TMP1]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> [[A:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> [[B:%.*]])
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <4 x float> @llvm.minnum.v4f32(<4 x float> [[TMP0]], <4 x float> [[TMP1]])
+// CHECK-NEXT: ret <4 x float> [[TMP2]]
//
float32x4_t test_vminnmaq_f32(float32x4_t a, float32x4_t b)
{
Index: clang/test/CodeGen/arm-mve-intrinsics/vmaxnmaq.c
===================================================================
--- clang/test/CodeGen/arm-mve-intrinsics/vmaxnmaq.c
+++ clang/test/CodeGen/arm-mve-intrinsics/vmaxnmaq.c
@@ -6,9 +6,10 @@
// CHECK-LABEL: @test_vmaxnmaq_f16(
// CHECK-NEXT: entry:
-// CHECK-NEXT: [[TMP0:%.*]] = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> [[B:%.*]])
-// CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x half> @llvm.maxnum.v8f16(<8 x half> [[A:%.*]], <8 x half> [[TMP0]])
-// CHECK-NEXT: ret <8 x half> [[TMP1]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> [[A:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> [[B:%.*]])
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <8 x half> @llvm.maxnum.v8f16(<8 x half> [[TMP0]], <8 x half> [[TMP1]])
+// CHECK-NEXT: ret <8 x half> [[TMP2]]
//
float16x8_t test_vmaxnmaq_f16(float16x8_t a, float16x8_t b)
{
@@ -21,9 +22,10 @@
// CHECK-LABEL: @test_vmaxnmaq_f32(
// CHECK-NEXT: entry:
-// CHECK-NEXT: [[TMP0:%.*]] = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> [[B:%.*]])
-// CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x float> @llvm.maxnum.v4f32(<4 x float> [[A:%.*]], <4 x float> [[TMP0]])
-// CHECK-NEXT: ret <4 x float> [[TMP1]]
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> [[A:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> [[B:%.*]])
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <4 x float> @llvm.maxnum.v4f32(<4 x float> [[TMP0]], <4 x float> [[TMP1]])
+// CHECK-NEXT: ret <4 x float> [[TMP2]]
//
float32x4_t test_vmaxnmaq_f32(float32x4_t a, float32x4_t b)
{
Index: clang/include/clang/Basic/arm_mve.td
===================================================================
--- clang/include/clang/Basic/arm_mve.td
+++ clang/include/clang/Basic/arm_mve.td
@@ -307,10 +307,12 @@
(IRIntBase<"maxnum", [Vector]> $a, $b)>;
def vminnmaq: Intrinsic<Vector, (args Vector:$a, Vector:$b),
(IRIntBase<"minnum", [Vector]>
- $a, (IRIntBase<"fabs", [Vector]> $b))>;
+ (IRIntBase<"fabs", [Vector]> $a),
+ (IRIntBase<"fabs", [Vector]> $b))>;
def vmaxnmaq: Intrinsic<Vector, (args Vector:$a, Vector:$b),
(IRIntBase<"maxnum", [Vector]>
- $a, (IRIntBase<"fabs", [Vector]> $b))>;
+ (IRIntBase<"fabs", [Vector]> $a),
+ (IRIntBase<"fabs", [Vector]> $b))>;
}
def vpselq: Intrinsic<Vector, (args Vector:$t, Vector:$f, Predicate:$pred),
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
