[ARM,MVE] Generate the right instruction for vmaxnmq_m_f16.

Summary:
Due to a copy-paste error in the isel patterns, the predicated version
of this intrinsic was expanding to the `VMAXNMT.F32` instruction
instead of `VMAXNMT.F16`. Similarly for vminnm.

Reviewers: dmgreen, miyuki, MarkMurrayARM

Reviewed By: dmgreen

Subscribers: kristof.beyls, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D72269

diff --git a/llvm/lib/Target/ARM/ARMInstrMVE.td b/llvm/lib/Target/ARM/ARMInstrMVE.td
index 8945806..3189caf 100644 (file)
--- a/llvm/lib/Target/ARM/ARMInstrMVE.td
+++ b/llvm/lib/Target/ARM/ARMInstrMVE.td
@@ -1097,7 +1097,7 @@ let Predicates = [HasMVEFloat] in {
                           (v4f32 MQPR:$inactive)))>;
   def : Pat<(v8f16 (int_arm_mve_max_predicated (v8f16 MQPR:$val1), (v8f16 MQPR:$val2),
                           (v8i1 VCCR:$mask), (v8f16 MQPR:$inactive))),
-            (v8f16 (MVE_VMAXNMf32 (v8f16 MQPR:$val1), (v8f16 MQPR:$val2),
+            (v8f16 (MVE_VMAXNMf16 (v8f16 MQPR:$val1), (v8f16 MQPR:$val2),
                           ARMVCCThen, (v8i1 VCCR:$mask),
                           (v8f16 MQPR:$inactive)))>;
 }
@@ -1117,7 +1117,7 @@ let Predicates = [HasMVEFloat] in {
                           (v4f32 MQPR:$inactive)))>;
   def : Pat<(v8f16 (int_arm_mve_min_predicated (v8f16 MQPR:$val1), (v8f16 MQPR:$val2),
                           (v8i1 VCCR:$mask), (v8f16 MQPR:$inactive))),
-            (v8f16 (MVE_VMINNMf32 (v8f16 MQPR:$val1), (v8f16 MQPR:$val2),
+            (v8f16 (MVE_VMINNMf16 (v8f16 MQPR:$val1), (v8f16 MQPR:$val2),
                           ARMVCCThen, (v8i1 VCCR:$mask),
                           (v8f16 MQPR:$inactive)))>;
 }

diff --git a/llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmq.ll b/llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmq.ll
index 54a1400..c1e45b8 100644 (file)
--- a/llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmq.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-intrinsics/vmaxnmq.ll
@@ -30,7 +30,7 @@ define arm_aapcs_vfpcc <8 x half> @test_vmaxnmq_m_f16(<8 x half> %inactive, <8 x
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vmsr p0, r0
 ; CHECK-NEXT:    vpst
-; CHECK-NEXT:    vmaxnmt.f32 q0, q1, q2
+; CHECK-NEXT:    vmaxnmt.f16 q0, q1, q2
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = zext i16 %p to i32
@@ -66,7 +66,7 @@ define arm_aapcs_vfpcc <8 x half> @test_vmaxnmq_x_f16(<8 x half> %a, <8 x half>
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vmsr p0, r0
 ; CHECK-NEXT:    vpst
-; CHECK-NEXT:    vmaxnmt.f32 q0, q0, q1
+; CHECK-NEXT:    vmaxnmt.f16 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = zext i16 %p to i32

diff --git a/llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmq.ll b/llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmq.ll
index ae44909..9bbc2a9 100644 (file)
--- a/llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmq.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-intrinsics/vminnmq.ll
@@ -30,7 +30,7 @@ define arm_aapcs_vfpcc <8 x half> @test_vminnmq_m_f16(<8 x half> %inactive, <8 x
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vmsr p0, r0
 ; CHECK-NEXT:    vpst
-; CHECK-NEXT:    vminnmt.f32 q0, q1, q2
+; CHECK-NEXT:    vminnmt.f16 q0, q1, q2
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = zext i16 %p to i32
@@ -66,7 +66,7 @@ define arm_aapcs_vfpcc <8 x half> @test_vminnmq_x_f16(<8 x half> %a, <8 x half>
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vmsr p0, r0
 ; CHECK-NEXT:    vpst
-; CHECK-NEXT:    vminnmt.f32 q0, q0, q1
+; CHECK-NEXT:    vminnmt.f16 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = zext i16 %p to i32