[ARM] VQADD instructions

This selects MVE VQADD from the vector llvm.sadd.sat or llvm.uadd.sat
intrinsics.

Differential Revision: https://reviews.llvm.org/D68566

llvm-svn: 374336

3 files changed, 42 insertions, 74 deletions

diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index c1365f58930..90b709867fe 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -265,6 +265,8 @@ void ARMTargetLowering::addMVEVectorTypes(bool HasMVEFP) {
     setOperationAction(ISD::CTTZ, VT, Custom);
     setOperationAction(ISD::BITREVERSE, VT, Legal);
     setOperationAction(ISD::BSWAP, VT, Legal);
+    setOperationAction(ISD::SADDSAT, VT, Legal);
+    setOperationAction(ISD::UADDSAT, VT, Legal);
 
     // No native support for these.
     setOperationAction(ISD::UDIV, VT, Expand);
diff --git a/llvm/lib/Target/ARM/ARMInstrMVE.td b/llvm/lib/Target/ARM/ARMInstrMVE.td
index 3ba23d9812c..28a32e1263d 100644
--- a/llvm/lib/Target/ARM/ARMInstrMVE.td
+++ b/llvm/lib/Target/ARM/ARMInstrMVE.td
@@ -1524,8 +1524,8 @@ let Predicates = [HasMVEInt] in {
 }
 
 class MVE_VQADDSUB<string iname, string suffix, bit U, bit subtract,
-                   bits<2> size, list<dag> pattern=[]>
-  : MVE_int<iname, suffix, size, pattern> {
+                   bits<2> size, ValueType vt>
+  : MVE_int<iname, suffix, size, []> {
 
   let Inst{28} = U;
   let Inst{25-23} = 0b110;
@@ -1535,26 +1535,40 @@ class MVE_VQADDSUB<string iname, string suffix, bit U, bit subtract,
   let Inst{8} = 0b0;
   let Inst{4} = 0b1;
   let Inst{0} = 0b0;
+
+  ValueType VT = vt;
+}
+
+class MVE_VQADD<string suffix, bit U, bits<2> size, ValueType VT>
+  : MVE_VQADDSUB<"vqadd", suffix, U, 0b0, size, VT>;
+class MVE_VQSUB<string suffix, bit U, bits<2> size, ValueType VT>
+  : MVE_VQADDSUB<"vqsub", suffix, U, 0b1, size, VT>;
+
+def MVE_VQADDs8  : MVE_VQADD<"s8",  0b0, 0b00, v16i8>;
+def MVE_VQADDs16 : MVE_VQADD<"s16", 0b0, 0b01, v8i16>;
+def MVE_VQADDs32 : MVE_VQADD<"s32", 0b0, 0b10, v4i32>;
+def MVE_VQADDu8  : MVE_VQADD<"u8",  0b1, 0b00, v16i8>;
+def MVE_VQADDu16 : MVE_VQADD<"u16", 0b1, 0b01, v8i16>;
+def MVE_VQADDu32 : MVE_VQADD<"u32", 0b1, 0b10, v4i32>;
+
+def MVE_VQSUBs8  : MVE_VQSUB<"s8",  0b0, 0b00, v16i8>;
+def MVE_VQSUBs16 : MVE_VQSUB<"s16", 0b0, 0b01, v8i16>;
+def MVE_VQSUBs32 : MVE_VQSUB<"s32", 0b0, 0b10, v4i32>;
+def MVE_VQSUBu8  : MVE_VQSUB<"u8",  0b1, 0b00, v16i8>;
+def MVE_VQSUBu16 : MVE_VQSUB<"u16", 0b1, 0b01, v8i16>;
+def MVE_VQSUBu32 : MVE_VQSUB<"u32", 0b1, 0b10, v4i32>;
+
+let Predicates = [HasMVEInt] in {
+  foreach instr = [MVE_VQADDu8, MVE_VQADDu16, MVE_VQADDu32] in
+    foreach VT = [instr.VT] in
+      def : Pat<(VT (uaddsat (VT MQPR:$Qm), (VT MQPR:$Qn))),
+                (VT (instr (VT MQPR:$Qm), (VT MQPR:$Qn)))>;
+  foreach instr = [MVE_VQADDs8, MVE_VQADDs16, MVE_VQADDs32] in
+    foreach VT = [instr.VT] in
+      def : Pat<(VT (saddsat (VT MQPR:$Qm), (VT MQPR:$Qn))),
+                (VT (instr (VT MQPR:$Qm), (VT MQPR:$Qn)))>;
 }
 
-class MVE_VQADD<string suffix, bit U, bits<2> size, list<dag> pattern=[]>
-  : MVE_VQADDSUB<"vqadd", suffix, U, 0b0, size, pattern>;
-class MVE_VQSUB<string suffix, bit U, bits<2> size, list<dag> pattern=[]>
-  : MVE_VQADDSUB<"vqsub", suffix, U, 0b1, size, pattern>;
-
-def MVE_VQADDs8  : MVE_VQADD<"s8",  0b0, 0b00>;
-def MVE_VQADDs16 : MVE_VQADD<"s16", 0b0, 0b01>;
-def MVE_VQADDs32 : MVE_VQADD<"s32", 0b0, 0b10>;
-def MVE_VQADDu8  : MVE_VQADD<"u8",  0b1, 0b00>;
-def MVE_VQADDu16 : MVE_VQADD<"u16", 0b1, 0b01>;
-def MVE_VQADDu32 : MVE_VQADD<"u32", 0b1, 0b10>;
-
-def MVE_VQSUBs8  : MVE_VQSUB<"s8",  0b0, 0b00>;
-def MVE_VQSUBs16 : MVE_VQSUB<"s16", 0b0, 0b01>;
-def MVE_VQSUBs32 : MVE_VQSUB<"s32", 0b0, 0b10>;
-def MVE_VQSUBu8  : MVE_VQSUB<"u8",  0b1, 0b00>;
-def MVE_VQSUBu16 : MVE_VQSUB<"u16", 0b1, 0b01>;
-def MVE_VQSUBu32 : MVE_VQSUB<"u32", 0b1, 0b10>;
 
 class MVE_VABD_int<string suffix, bit U, bits<2> size, list<dag> pattern=[]>
   : MVE_int<"vabd", suffix, size, pattern> {
diff --git a/llvm/test/CodeGen/Thumb2/mve-saturating-arith.ll b/llvm/test/CodeGen/Thumb2/mve-saturating-arith.ll
index f6a9af42f6a..2a775fe9543 100644
--- a/llvm/test/CodeGen/Thumb2/mve-saturating-arith.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-saturating-arith.ll
@@ -4,21 +4,7 @@
 define arm_aapcs_vfpcc <16 x i8> @sadd_int8_t(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: sadd_int8_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9}
-; CHECK-NEXT:    vpush {d8, d9}
-; CHECK-NEXT:    vadd.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 lt, 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:    vqadd.s8 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %src1, <16 x i8> %src2)
@@ -28,21 +14,7 @@ entry:
 define arm_aapcs_vfpcc <8 x i16> @sadd_int16_t(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: sadd_int16_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9}
-; CHECK-NEXT:    vpush {d8, d9}
-; CHECK-NEXT:    vadd.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 lt, 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:    vqadd.s16 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %src1, <8 x i16> %src2)
@@ -52,21 +24,7 @@ entry:
 define arm_aapcs_vfpcc <4 x i32> @sadd_int32_t(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: sadd_int32_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9}
-; CHECK-NEXT:    vpush {d8, d9}
-; CHECK-NEXT:    vadd.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 lt, 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:    vqadd.s32 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> %src1, <4 x i32> %src2)
@@ -156,9 +114,7 @@ entry:
 define arm_aapcs_vfpcc <16 x i8> @uadd_int8_t(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: uadd_int8_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmvn q2, q1
-; CHECK-NEXT:    vmin.u8 q0, q0, q2
-; CHECK-NEXT:    vadd.i8 q0, q0, q1
+; CHECK-NEXT:    vqadd.u8 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <16 x i8> @llvm.uadd.sat.v16i8(<16 x i8> %src1, <16 x i8> %src2)
@@ -168,9 +124,7 @@ entry:
 define arm_aapcs_vfpcc <8 x i16> @uadd_int16_t(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: uadd_int16_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmvn q2, q1
-; CHECK-NEXT:    vmin.u16 q0, q0, q2
-; CHECK-NEXT:    vadd.i16 q0, q0, q1
+; CHECK-NEXT:    vqadd.u16 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <8 x i16> @llvm.uadd.sat.v8i16(<8 x i16> %src1, <8 x i16> %src2)
@@ -180,9 +134,7 @@ entry:
 define arm_aapcs_vfpcc <4 x i32> @uadd_int32_t(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: uadd_int32_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmvn q2, q1
-; CHECK-NEXT:    vmin.u32 q0, q0, q2
-; CHECK-NEXT:    vadd.i32 q0, q0, q1
+; CHECK-NEXT:    vqadd.u32 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = call <4 x i32> @llvm.uadd.sat.v4i32(<4 x i32> %src1, <4 x i32> %src2)