[ARM] Remove VHADD patterns

These instructions do not work quite like I expected them to. They
perform the addition and then shift in a higher precision integer, so do
not match up with the patterns that we added.

For example with s8s, adding 100 and 100 should wrap leaving the shift
to work on a negative number. VHADD will instead do the arithmetic in
higher precision, giving 100 overall. The vhadd gives a "better" result,
but not one that matches up with the input.

I am just removing the patterns here. We might be able to re-add them in
the future by checking for wrap flags or changing bitwidths. But for the
moment just remove them to remove the problem cases.

2 files changed, 48 insertions, 78 deletions

diff --git a/llvm/lib/Target/ARM/ARMInstrMVE.td b/llvm/lib/Target/ARM/ARMInstrMVE.td
index df385034589..cc3a8ee77d7 100644
--- a/llvm/lib/Target/ARM/ARMInstrMVE.td
+++ b/llvm/lib/Target/ARM/ARMInstrMVE.td
@@ -1785,60 +1785,6 @@ def MVE_VHSUBu8  : MVE_VHSUB<"u8",  0b1, 0b00>;
 def MVE_VHSUBu16 : MVE_VHSUB<"u16", 0b1, 0b01>;
 def MVE_VHSUBu32 : MVE_VHSUB<"u32", 0b1, 0b10>;
 
-let Predicates = [HasMVEInt] in {
-  def : Pat<(v16i8 (ARMvshrsImm
-                     (v16i8 (add (v16i8 MQPR:$v1), (v16i8 MQPR:$v2))), 1)),
-            (v16i8 (MVE_VHADDs8
-                     (v16i8 MQPR:$v1), (v16i8 MQPR:$v2)))>;
-  def : Pat<(v8i16 (ARMvshrsImm
-                     (v8i16 (add (v8i16 MQPR:$v1), (v8i16 MQPR:$v2))), 1)),
-            (v8i16 (MVE_VHADDs16
-                     (v8i16 MQPR:$v1), (v8i16 MQPR:$v2)))>;
-  def : Pat<(v4i32 (ARMvshrsImm
-                     (v4i32 (add (v4i32 MQPR:$v1), (v4i32 MQPR:$v2))), 1)),
-            (v4i32 (MVE_VHADDs32
-                     (v4i32 MQPR:$v1), (v4i32 MQPR:$v2)))>;
-
-  def : Pat<(v16i8 (ARMvshruImm
-                     (v16i8 (add (v16i8 MQPR:$v1), (v16i8 MQPR:$v2))), 1)),
-            (v16i8 (MVE_VHADDu8
-                     (v16i8 MQPR:$v1), (v16i8 MQPR:$v2)))>;
-  def : Pat<(v8i16 (ARMvshruImm
-                     (v8i16 (add (v8i16 MQPR:$v1), (v8i16 MQPR:$v2))), 1)),
-            (v8i16 (MVE_VHADDu16
-                     (v8i16 MQPR:$v1), (v8i16 MQPR:$v2)))>;
-  def : Pat<(v4i32 (ARMvshruImm
-                     (v4i32 (add (v4i32 MQPR:$v1), (v4i32 MQPR:$v2))), 1)),
-            (v4i32 (MVE_VHADDu32
-                     (v4i32 MQPR:$v1), (v4i32 MQPR:$v2)))>;
-
-  def : Pat<(v16i8 (ARMvshrsImm
-                     (v16i8 (sub (v16i8 MQPR:$v1), (v16i8 MQPR:$v2))), 1)),
-            (v16i8 (MVE_VHSUBs8
-                     (v16i8 MQPR:$v1), (v16i8 MQPR:$v2)))>;
-  def : Pat<(v8i16 (ARMvshrsImm
-                     (v8i16 (sub (v8i16 MQPR:$v1), (v8i16 MQPR:$v2))), 1)),
-            (v8i16 (MVE_VHSUBs16
-                     (v8i16 MQPR:$v1), (v8i16 MQPR:$v2)))>;
-  def : Pat<(v4i32 (ARMvshrsImm
-                     (v4i32 (sub (v4i32 MQPR:$v1), (v4i32 MQPR:$v2))), 1)),
-            (v4i32 (MVE_VHSUBs32
-                     (v4i32 MQPR:$v1), (v4i32 MQPR:$v2)))>;
-
-  def : Pat<(v16i8 (ARMvshruImm
-                     (v16i8 (sub (v16i8 MQPR:$v1), (v16i8 MQPR:$v2))), 1)),
-            (v16i8 (MVE_VHSUBu8
-                     (v16i8 MQPR:$v1), (v16i8 MQPR:$v2)))>;
-  def : Pat<(v8i16 (ARMvshruImm
-                     (v8i16 (sub (v8i16 MQPR:$v1), (v8i16 MQPR:$v2))), 1)),
-            (v8i16 (MVE_VHSUBu16
-                     (v8i16 MQPR:$v1), (v8i16 MQPR:$v2)))>;
-  def : Pat<(v4i32 (ARMvshruImm
-                     (v4i32 (sub (v4i32 MQPR:$v1), (v4i32 MQPR:$v2))), 1)),
-            (v4i32 (MVE_VHSUBu32
-                     (v4i32 MQPR:$v1), (v4i32 MQPR:$v2)))>;
-}
-
 class MVE_VDUP<string suffix, bit B, bit E, list<dag> pattern=[]>
   : MVE_p<(outs MQPR:$Qd), (ins rGPR:$Rt), NoItinerary,
           "vdup", suffix, "$Qd, $Rt", vpred_r, "", pattern> {
diff --git a/llvm/test/CodeGen/Thumb2/mve-vhaddsub.ll b/llvm/test/CodeGen/Thumb2/mve-vhaddsub.ll
index 19979f203f1..83534e2c3e8 100644
--- a/llvm/test/CodeGen/Thumb2/mve-vhaddsub.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-vhaddsub.ll
@@ -4,7 +4,8 @@
 define arm_aapcs_vfpcc <16 x i8> @add_ashr_v16i8(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: add_ashr_v16i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhadd.s8 q0, q0, q1
+; CHECK-NEXT:    vadd.i8 q0, q0, q1
+; CHECK-NEXT:    vshr.s8 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add <16 x i8> %src1, %src2
@@ -15,7 +16,8 @@ entry:
 define arm_aapcs_vfpcc <8 x i16> @add_ashr_v8i16(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: add_ashr_v8i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhadd.s16 q0, q0, q1
+; CHECK-NEXT:    vadd.i16 q0, q0, q1
+; CHECK-NEXT:    vshr.s16 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add <8 x i16> %src1, %src2
@@ -26,7 +28,8 @@ entry:
 define arm_aapcs_vfpcc <4 x i32> @add_ashr_v4i32(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: add_ashr_v4i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhadd.s32 q0, q0, q1
+; CHECK-NEXT:    vadd.i32 q0, q0, q1
+; CHECK-NEXT:    vshr.s32 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add nsw <4 x i32> %src1, %src2
@@ -37,7 +40,8 @@ entry:
 define arm_aapcs_vfpcc <16 x i8> @add_lshr_v16i8(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: add_lshr_v16i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhadd.u8 q0, q0, q1
+; CHECK-NEXT:    vadd.i8 q0, q0, q1
+; CHECK-NEXT:    vshr.u8 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add <16 x i8> %src1, %src2
@@ -48,7 +52,8 @@ entry:
 define arm_aapcs_vfpcc <8 x i16> @add_lshr_v8i16(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: add_lshr_v8i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhadd.u16 q0, q0, q1
+; CHECK-NEXT:    vadd.i16 q0, q0, q1
+; CHECK-NEXT:    vshr.u16 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add <8 x i16> %src1, %src2
@@ -59,7 +64,8 @@ entry:
 define arm_aapcs_vfpcc <4 x i32> @add_lshr_v4i32(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: add_lshr_v4i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhadd.u32 q0, q0, q1
+; CHECK-NEXT:    vadd.i32 q0, q0, q1
+; CHECK-NEXT:    vshr.u32 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add nsw <4 x i32> %src1, %src2
@@ -70,7 +76,8 @@ entry:
 define arm_aapcs_vfpcc <16 x i8> @sub_ashr_v16i8(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: sub_ashr_v16i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhsub.s8 q0, q0, q1
+; CHECK-NEXT:    vsub.i8 q0, q0, q1
+; CHECK-NEXT:    vshr.s8 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub <16 x i8> %src1, %src2
@@ -81,7 +88,8 @@ entry:
 define arm_aapcs_vfpcc <8 x i16> @sub_ashr_v8i16(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: sub_ashr_v8i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhsub.s16 q0, q0, q1
+; CHECK-NEXT:    vsub.i16 q0, q0, q1
+; CHECK-NEXT:    vshr.s16 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub <8 x i16> %src1, %src2
@@ -92,7 +100,8 @@ entry:
 define arm_aapcs_vfpcc <4 x i32> @sub_ashr_v4i32(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: sub_ashr_v4i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhsub.s32 q0, q0, q1
+; CHECK-NEXT:    vsub.i32 q0, q0, q1
+; CHECK-NEXT:    vshr.s32 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub nsw <4 x i32> %src1, %src2
@@ -103,7 +112,8 @@ entry:
 define arm_aapcs_vfpcc <16 x i8> @sub_lshr_v16i8(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: sub_lshr_v16i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhsub.u8 q0, q0, q1
+; CHECK-NEXT:    vsub.i8 q0, q0, q1
+; CHECK-NEXT:    vshr.u8 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub <16 x i8> %src1, %src2
@@ -114,7 +124,8 @@ entry:
 define arm_aapcs_vfpcc <8 x i16> @sub_lshr_v8i16(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: sub_lshr_v8i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhsub.u16 q0, q0, q1
+; CHECK-NEXT:    vsub.i16 q0, q0, q1
+; CHECK-NEXT:    vshr.u16 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub <8 x i16> %src1, %src2
@@ -125,7 +136,8 @@ entry:
 define arm_aapcs_vfpcc <4 x i32> @sub_lshr_v4i32(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: sub_lshr_v4i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhsub.u32 q0, q0, q1
+; CHECK-NEXT:    vsub.i32 q0, q0, q1
+; CHECK-NEXT:    vshr.u32 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub nsw <4 x i32> %src1, %src2
@@ -140,7 +152,8 @@ define arm_aapcs_vfpcc <16 x i8> @add_sdiv_v16i8(<16 x i8> %src1, <16 x i8> %src
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vadd.i8 q0, q0, q1
 ; CHECK-NEXT:    vshr.u8 q1, q0, #7
-; CHECK-NEXT:    vhadd.s8 q0, q0, q1
+; CHECK-NEXT:    vadd.i8 q0, q0, q1
+; CHECK-NEXT:    vshr.s8 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add <16 x i8> %src1, %src2
@@ -153,7 +166,8 @@ define arm_aapcs_vfpcc <8 x i16> @add_sdiv_v8i16(<8 x i16> %src1, <8 x i16> %src
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vadd.i16 q0, q0, q1
 ; CHECK-NEXT:    vshr.u16 q1, q0, #15
-; CHECK-NEXT:    vhadd.s16 q0, q0, q1
+; CHECK-NEXT:    vadd.i16 q0, q0, q1
+; CHECK-NEXT:    vshr.s16 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add <8 x i16> %src1, %src2
@@ -166,7 +180,8 @@ define arm_aapcs_vfpcc <4 x i32> @add_sdiv_v4i32(<4 x i32> %src1, <4 x i32> %src
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vadd.i32 q0, q0, q1
 ; CHECK-NEXT:    vshr.u32 q1, q0, #31
-; CHECK-NEXT:    vhadd.s32 q0, q0, q1
+; CHECK-NEXT:    vadd.i32 q0, q0, q1
+; CHECK-NEXT:    vshr.s32 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add nsw <4 x i32> %src1, %src2
@@ -177,7 +192,8 @@ entry:
 define arm_aapcs_vfpcc <16 x i8> @add_udiv_v16i8(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: add_udiv_v16i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhadd.u8 q0, q0, q1
+; CHECK-NEXT:    vadd.i8 q0, q0, q1
+; CHECK-NEXT:    vshr.u8 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add <16 x i8> %src1, %src2
@@ -188,7 +204,8 @@ entry:
 define arm_aapcs_vfpcc <8 x i16> @add_udiv_v8i16(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: add_udiv_v8i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhadd.u16 q0, q0, q1
+; CHECK-NEXT:    vadd.i16 q0, q0, q1
+; CHECK-NEXT:    vshr.u16 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add <8 x i16> %src1, %src2
@@ -199,7 +216,8 @@ entry:
 define arm_aapcs_vfpcc <4 x i32> @add_udiv_v4i32(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: add_udiv_v4i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhadd.u32 q0, q0, q1
+; CHECK-NEXT:    vadd.i32 q0, q0, q1
+; CHECK-NEXT:    vshr.u32 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = add nsw <4 x i32> %src1, %src2
@@ -212,7 +230,8 @@ define arm_aapcs_vfpcc <16 x i8> @sub_sdiv_v16i8(<16 x i8> %src1, <16 x i8> %src
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vsub.i8 q0, q0, q1
 ; CHECK-NEXT:    vshr.u8 q1, q0, #7
-; CHECK-NEXT:    vhadd.s8 q0, q0, q1
+; CHECK-NEXT:    vadd.i8 q0, q0, q1
+; CHECK-NEXT:    vshr.s8 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub <16 x i8> %src1, %src2
@@ -225,7 +244,8 @@ define arm_aapcs_vfpcc <8 x i16> @sub_sdiv_v8i16(<8 x i16> %src1, <8 x i16> %src
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vsub.i16 q0, q0, q1
 ; CHECK-NEXT:    vshr.u16 q1, q0, #15
-; CHECK-NEXT:    vhadd.s16 q0, q0, q1
+; CHECK-NEXT:    vadd.i16 q0, q0, q1
+; CHECK-NEXT:    vshr.s16 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub <8 x i16> %src1, %src2
@@ -238,7 +258,8 @@ define arm_aapcs_vfpcc <4 x i32> @sub_sdiv_v4i32(<4 x i32> %src1, <4 x i32> %src
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vsub.i32 q0, q0, q1
 ; CHECK-NEXT:    vshr.u32 q1, q0, #31
-; CHECK-NEXT:    vhadd.s32 q0, q0, q1
+; CHECK-NEXT:    vadd.i32 q0, q0, q1
+; CHECK-NEXT:    vshr.s32 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub nsw <4 x i32> %src1, %src2
@@ -249,7 +270,8 @@ entry:
 define arm_aapcs_vfpcc <16 x i8> @sub_udiv_v16i8(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: sub_udiv_v16i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhsub.u8 q0, q0, q1
+; CHECK-NEXT:    vsub.i8 q0, q0, q1
+; CHECK-NEXT:    vshr.u8 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub <16 x i8> %src1, %src2
@@ -260,7 +282,8 @@ entry:
 define arm_aapcs_vfpcc <8 x i16> @sub_udiv_v8i16(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: sub_udiv_v8i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhsub.u16 q0, q0, q1
+; CHECK-NEXT:    vsub.i16 q0, q0, q1
+; CHECK-NEXT:    vshr.u16 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub <8 x i16> %src1, %src2
@@ -271,7 +294,8 @@ entry:
 define arm_aapcs_vfpcc <4 x i32> @sub_udiv_v4i32(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: sub_udiv_v4i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vhsub.u32 q0, q0, q1
+; CHECK-NEXT:    vsub.i32 q0, q0, q1
+; CHECK-NEXT:    vshr.u32 q0, q0, #1
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = sub nsw <4 x i32> %src1, %src2