[ARM][MVE][Intrinsics] Take abs() of VMINNMAQ, VMAXNMAQ intrinsics' first arguments.

Summary: Fix VMINNMAQ, VMAXNMAQ intrinsics; BOTH arguments have the absolute values taken.

Reviewers: dmgreen, simon_tatham

Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D72830

diff --git a/clang/include/clang/Basic/arm_mve.td b/clang/include/clang/Basic/arm_mve.td
index a348713..d28f973 100644 (file)
--- a/clang/include/clang/Basic/arm_mve.td
+++ b/clang/include/clang/Basic/arm_mve.td
@@ -307,10 +307,12 @@ let params = T.Float in {
                                  (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),

diff --git a/clang/test/CodeGen/arm-mve-intrinsics/vmaxnmaq.c b/clang/test/CodeGen/arm-mve-intrinsics/vmaxnmaq.c
index ae963e2..b7e85ff 100644 (file)
--- a/clang/test/CodeGen/arm-mve-intrinsics/vmaxnmaq.c
+++ b/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 @@ float16x8_t test_vmaxnmaq_f16(float16x8_t a, float16x8_t b)
 
 // 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)
 {

diff --git a/clang/test/CodeGen/arm-mve-intrinsics/vminnmaq.c b/clang/test/CodeGen/arm-mve-intrinsics/vminnmaq.c
index e16ede0..955e8fe 100644 (file)
--- a/clang/test/CodeGen/arm-mve-intrinsics/vminnmaq.c
+++ b/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 @@ float16x8_t test_vminnmaq_f16(float16x8_t a, float16x8_t b)
 
 // 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)
 {

diff --git a/llvm/lib/Target/ARM/ARMInstrMVE.td b/llvm/lib/Target/ARM/ARMInstrMVE.td
index 70f1d75..1cccad2 100644 (file)
--- a/llvm/lib/Target/ARM/ARMInstrMVE.td
+++ b/llvm/lib/Target/ARM/ARMInstrMVE.td
@@ -3655,7 +3655,8 @@ multiclass MVE_VMAXMINNMA_m<string iname, MVEVectorVTInfo VTI,
 
   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

diff --git a/llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmaq.ll b/llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmaq.ll
index df0367e..0f412ca 100644 (file)
--- a/llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmaq.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmaq.ll
@@ -7,9 +7,10 @@ define arm_aapcs_vfpcc <8 x half> @test_vmaxnmaq_f16(<8 x half> %a, <8 x half> %
 ; 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 @@ define arm_aapcs_vfpcc <4 x float> @test_vmaxnmaq_f32(<4 x float> %a, <4 x float
 ; 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

diff --git a/llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmaq.ll b/llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmaq.ll
index dec0d9e..cff73c8 100644 (file)
--- a/llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmaq.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmaq.ll
@@ -7,9 +7,10 @@ define arm_aapcs_vfpcc <8 x half> @test_vminnmaq_f16(<8 x half> %a, <8 x half> %
 ; 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 @@ define arm_aapcs_vfpcc <4 x float> @test_vminnmaq_f32(<4 x float> %a, <4 x float
 ; 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