[llvm] r374377 - [ARM] VQSUB instruction

David Green via llvm-commits llvm-commits at lists.llvm.org
Thu Oct 10 09:34:30 PDT 2019 

Author: dmgreen
Date: Thu Oct 10 09:34:30 2019
New Revision: 374377

URL: http://llvm.org/viewvc/llvm-project?rev=374377&view=rev
Log:
[ARM] VQSUB instruction

Same as VQADD, VQSUB can be selected from llvm.ssub.sat intrinsics.

Differential Revision: https://reviews.llvm.org/D68567

Modified:
    llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
    llvm/trunk/lib/Target/ARM/ARMInstrMVE.td
    llvm/trunk/test/CodeGen/Thumb2/mve-saturating-arith.ll

Modified: llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp?rev=374377&r1=374376&r2=374377&view=diff
==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp Thu Oct 10 09:34:30 2019
@@ -267,6 +267,8 @@ void ARMTargetLowering::addMVEVectorType
     setOperationAction(ISD::BSWAP, VT, Legal);
     setOperationAction(ISD::SADDSAT, VT, Legal);
     setOperationAction(ISD::UADDSAT, VT, Legal);
+    setOperationAction(ISD::SSUBSAT, VT, Legal);
+    setOperationAction(ISD::USUBSAT, VT, Legal);
 
     // No native support for these.
     setOperationAction(ISD::UDIV, VT, Expand);

Modified: llvm/trunk/lib/Target/ARM/ARMInstrMVE.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrMVE.td?rev=374377&r1=374376&r2=374377&view=diff
==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrMVE.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrMVE.td Thu Oct 10 09:34:30 2019
@@ -1567,6 +1567,14 @@ let Predicates = [HasMVEInt] in {
     foreach VT = [instr.VT] in
       def : Pat<(VT (saddsat (VT MQPR:$Qm), (VT MQPR:$Qn))),
                 (VT (instr (VT MQPR:$Qm), (VT MQPR:$Qn)))>;
+  foreach instr = [MVE_VQSUBu8, MVE_VQSUBu16, MVE_VQSUBu32] in
+    foreach VT = [instr.VT] in
+      def : Pat<(VT (usubsat (VT MQPR:$Qm), (VT MQPR:$Qn))),
+                (VT (instr (VT MQPR:$Qm), (VT MQPR:$Qn)))>;
+  foreach instr = [MVE_VQSUBs8, MVE_VQSUBs16, MVE_VQSUBs32] in
+    foreach VT = [instr.VT] in
+      def : Pat<(VT (ssubsat (VT MQPR:$Qm), (VT MQPR:$Qn))),
+                (VT (instr (VT MQPR:$Qm), (VT MQPR:$Qn)))>;
 }
 
 

Modified: llvm/trunk/test/CodeGen/Thumb2/mve-saturating-arith.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Thumb2/mve-saturating-arith.ll?rev=374377&r1=374376&r2=374377&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/Thumb2/mve-saturating-arith.ll (original)
+++ llvm/trunk/test/CodeGen/Thumb2/mve-saturating-arith.ll Thu Oct 10 09:34:30 2019
@@ -191,21 +191,7 @@ entry:
 define arm_aapcs_vfpcc <16 x i8> @ssub_int8_t(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: ssub_int8_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9}
-; CHECK-NEXT:    vpush {d8, d9}
-; CHECK-NEXT:    vsub.i8 q2, q0, q1
-; CHECK-NEXT:    vmov.i8 q3, #0x80
-; CHECK-NEXT:    vcmp.s8 lt, q2, zr
-; CHECK-NEXT:    vmov.i8 q4, #0x7f
-; CHECK-NEXT:    vpsel q3, q4, q3
-; CHECK-NEXT:    vcmp.s8 gt, q0, q2
-; CHECK-NEXT:    vmrs r0, p0
-; CHECK-NEXT:    vcmp.s8 gt, q1, zr
-; CHECK-NEXT:    vmrs r1, p0
-; CHECK-NEXT:    eors r0, r1
-; CHECK-NEXT:    vmsr p0, r0
-; CHECK-NEXT:    vpsel q0, q3, q2
-; CHECK-NEXT:    vpop {d8, d9}
+; CHECK-NEXT:    vqsub.s8 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %src1, <16 x i8> %src2)
@@ -215,21 +201,7 @@ entry:
 define arm_aapcs_vfpcc <8 x i16> @ssub_int16_t(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: ssub_int16_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9}
-; CHECK-NEXT:    vpush {d8, d9}
-; CHECK-NEXT:    vsub.i16 q2, q0, q1
-; CHECK-NEXT:    vmov.i16 q3, #0x8000
-; CHECK-NEXT:    vcmp.s16 lt, q2, zr
-; CHECK-NEXT:    vmvn.i16 q4, #0x8000
-; CHECK-NEXT:    vpsel q3, q4, q3
-; CHECK-NEXT:    vcmp.s16 gt, q0, q2
-; CHECK-NEXT:    vmrs r0, p0
-; CHECK-NEXT:    vcmp.s16 gt, q1, zr
-; CHECK-NEXT:    vmrs r1, p0
-; CHECK-NEXT:    eors r0, r1
-; CHECK-NEXT:    vmsr p0, r0
-; CHECK-NEXT:    vpsel q0, q3, q2
-; CHECK-NEXT:    vpop {d8, d9}
+; CHECK-NEXT:    vqsub.s16 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %src1, <8 x i16> %src2)
@@ -239,21 +211,7 @@ entry:
 define arm_aapcs_vfpcc <4 x i32> @ssub_int32_t(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: ssub_int32_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9}
-; CHECK-NEXT:    vpush {d8, d9}
-; CHECK-NEXT:    vsub.i32 q2, q0, q1
-; CHECK-NEXT:    vmov.i32 q3, #0x80000000
-; CHECK-NEXT:    vcmp.s32 lt, q2, zr
-; CHECK-NEXT:    vmvn.i32 q4, #0x80000000
-; CHECK-NEXT:    vpsel q3, q4, q3
-; CHECK-NEXT:    vcmp.s32 gt, q0, q2
-; CHECK-NEXT:    vmrs r0, p0
-; CHECK-NEXT:    vcmp.s32 gt, q1, zr
-; CHECK-NEXT:    vmrs r1, p0
-; CHECK-NEXT:    eors r0, r1
-; CHECK-NEXT:    vmsr p0, r0
-; CHECK-NEXT:    vpsel q0, q3, q2
-; CHECK-NEXT:    vpop {d8, d9}
+; CHECK-NEXT:    vqsub.s32 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <4 x i32> @llvm.ssub.sat.v4i32(<4 x i32> %src1, <4 x i32> %src2)
@@ -358,8 +316,7 @@ entry:
 define arm_aapcs_vfpcc <16 x i8> @usub_int8_t(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: usub_int8_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmax.u8 q0, q0, q1
-; CHECK-NEXT:    vsub.i8 q0, q0, q1
+; CHECK-NEXT:    vqsub.u8 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <16 x i8> @llvm.usub.sat.v16i8(<16 x i8> %src1, <16 x i8> %src2)
@@ -369,8 +326,7 @@ entry:
 define arm_aapcs_vfpcc <8 x i16> @usub_int16_t(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: usub_int16_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmax.u16 q0, q0, q1
-; CHECK-NEXT:    vsub.i16 q0, q0, q1
+; CHECK-NEXT:    vqsub.u16 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <8 x i16> @llvm.usub.sat.v8i16(<8 x i16> %src1, <8 x i16> %src2)
@@ -380,8 +336,7 @@ entry:
 define arm_aapcs_vfpcc <4 x i32> @usub_int32_t(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: usub_int32_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmax.u32 q0, q0, q1
-; CHECK-NEXT:    vsub.i32 q0, q0, q1
+; CHECK-NEXT:    vqsub.u32 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> %src1, <4 x i32> %src2)

More information about the llvm-commits mailing list