[AArch64][SVE] Add patterns to select masked add/sub instructions

When lowering add(a, select(mask, b, splat(0))) the sel instruction can
be removed by using predicated add/sub instructions.

Reviewed By: paulwalker-arm

Differential Revision: https://reviews.llvm.org/D129751

diff --git a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
index 4032c46..0436404 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
@@ -337,6 +337,12 @@ def AArch64bic : PatFrags<(ops node:$op1, node:$op2),
 
 def AArch64subr : PatFrag<(ops node:$op1, node:$op2),
                           (sub node:$op2, node:$op1)>;
+def AArch64add_m1 : PatFrags<(ops node:$pred, node:$op1, node:$op2),
+                             [(int_aarch64_sve_add node:$pred, node:$op1, node:$op2),
+                              (add node:$op1, (vselect node:$pred, node:$op2, (SVEDup0)))]>;
+def AArch64sub_m1 : PatFrags<(ops node:$pred, node:$op1, node:$op2),
+                             [(int_aarch64_sve_sub node:$pred, node:$op1, node:$op2),
+                              (sub node:$op1, (vselect node:$pred, node:$op2, (SVEDup0)))]>;
 
 let Predicates = [HasSVE] in {
   defm RDFFR_PPz  : sve_int_rdffr_pred<0b0, "rdffr", int_aarch64_sve_rdffr_z>;
@@ -359,8 +365,8 @@ let Predicates = [HasSVEorSME] in {
   defm EOR_ZZZ : sve_int_bin_cons_log<0b10, "eor", xor>;
   defm BIC_ZZZ : sve_int_bin_cons_log<0b11, "bic", AArch64bic>;
 
-  defm ADD_ZPmZ  : sve_int_bin_pred_arit_0<0b000, "add",  "ADD_ZPZZ", int_aarch64_sve_add, DestructiveBinaryComm>;
-  defm SUB_ZPmZ  : sve_int_bin_pred_arit_0<0b001, "sub",  "SUB_ZPZZ", int_aarch64_sve_sub, DestructiveBinaryCommWithRev, "SUBR_ZPmZ">;
+  defm ADD_ZPmZ  : sve_int_bin_pred_arit_0<0b000, "add",  "ADD_ZPZZ", AArch64add_m1, DestructiveBinaryComm>;
+  defm SUB_ZPmZ  : sve_int_bin_pred_arit_0<0b001, "sub",  "SUB_ZPZZ", AArch64sub_m1, DestructiveBinaryCommWithRev, "SUBR_ZPmZ">;
   defm SUBR_ZPmZ : sve_int_bin_pred_arit_0<0b011, "subr", "SUBR_ZPZZ", int_aarch64_sve_subr, DestructiveBinaryCommWithRev, "SUB_ZPmZ", /*isReverseInstr*/ 1>;
 
   defm ORR_ZPmZ : sve_int_bin_pred_log<0b000, "orr", "ORR_ZPZZ", int_aarch64_sve_orr, DestructiveBinaryComm>;

diff --git a/llvm/test/CodeGen/AArch64/sve-masked-int-arith.ll b/llvm/test/CodeGen/AArch64/sve-masked-int-arith.ll
index 10b0de8..e532390 100644 (file)
--- a/llvm/test/CodeGen/AArch64/sve-masked-int-arith.ll
+++ b/llvm/test/CodeGen/AArch64/sve-masked-int-arith.ll
@@ -8,9 +8,7 @@
 define <vscale x 16 x i8> @masked_add_nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b, <vscale x 16 x i1> %mask) {
 ; CHECK-LABEL: masked_add_nxv16i8:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z2.b, #0 // =0x0
-; CHECK-NEXT:    sel z1.b, p0, z1.b, z2.b
-; CHECK-NEXT:    add z0.b, z0.b, z1.b
+; CHECK-NEXT:    add z0.b, p0/m, z0.b, z1.b
 ; CHECK-NEXT:    ret
   %select = select <vscale x 16 x i1> %mask, <vscale x 16 x i8> %b, <vscale x 16 x i8> zeroinitializer
   %ret = add <vscale x 16 x i8> %a, %select
@@ -20,9 +18,7 @@ define <vscale x 16 x i8> @masked_add_nxv16i8(<vscale x 16 x i8> %a, <vscale x 1
 define <vscale x 8 x i16> @masked_add_nxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b, <vscale x 8 x i1> %mask) {
 ; CHECK-LABEL: masked_add_nxv8i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z2.h, #0 // =0x0
-; CHECK-NEXT:    sel z1.h, p0, z1.h, z2.h
-; CHECK-NEXT:    add z0.h, z0.h, z1.h
+; CHECK-NEXT:    add z0.h, p0/m, z0.h, z1.h
 ; CHECK-NEXT:    ret
   %select = select <vscale x 8 x i1> %mask, <vscale x 8 x i16> %b, <vscale x 8 x i16> zeroinitializer
   %ret = add <vscale x 8 x i16> %a, %select
@@ -32,9 +28,7 @@ define <vscale x 8 x i16> @masked_add_nxv8i16(<vscale x 8 x i16> %a, <vscale x 8
 define <vscale x 4 x i32> @masked_add_nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, <vscale x 4 x i1> %mask) {
 ; CHECK-LABEL: masked_add_nxv4i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z2.s, #0 // =0x0
-; CHECK-NEXT:    sel z1.s, p0, z1.s, z2.s
-; CHECK-NEXT:    add z0.s, z0.s, z1.s
+; CHECK-NEXT:    add z0.s, p0/m, z0.s, z1.s
 ; CHECK-NEXT:    ret
   %select = select <vscale x 4 x i1> %mask, <vscale x 4 x i32> %b, <vscale x 4 x i32> zeroinitializer
   %ret = add <vscale x 4 x i32> %a, %select
@@ -44,9 +38,7 @@ define <vscale x 4 x i32> @masked_add_nxv4i32(<vscale x 4 x i32> %a, <vscale x 4
 define <vscale x 2 x i64> @masked_add_nxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b, <vscale x 2 x i1> %mask) {
 ; CHECK-LABEL: masked_add_nxv2i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z2.d, #0 // =0x0
-; CHECK-NEXT:    sel z1.d, p0, z1.d, z2.d
-; CHECK-NEXT:    add z0.d, z0.d, z1.d
+; CHECK-NEXT:    add z0.d, p0/m, z0.d, z1.d
 ; CHECK-NEXT:    ret
   %select = select <vscale x 2 x i1> %mask, <vscale x 2 x i64> %b, <vscale x 2 x i64> zeroinitializer
   %ret = add <vscale x 2 x i64> %a, %select
@@ -60,9 +52,7 @@ define <vscale x 2 x i64> @masked_add_nxv2i64(<vscale x 2 x i64> %a, <vscale x 2
 define <vscale x 16 x i8> @masked_sub_nxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b, <vscale x 16 x i1> %mask) {
 ; CHECK-LABEL: masked_sub_nxv16i8:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z2.b, #0 // =0x0
-; CHECK-NEXT:    sel z1.b, p0, z1.b, z2.b
-; CHECK-NEXT:    sub z0.b, z0.b, z1.b
+; CHECK-NEXT:    sub z0.b, p0/m, z0.b, z1.b
 ; CHECK-NEXT:    ret
   %select = select <vscale x 16 x i1> %mask, <vscale x 16 x i8> %b, <vscale x 16 x i8> zeroinitializer
   %ret = sub <vscale x 16 x i8> %a, %select
@@ -72,9 +62,7 @@ define <vscale x 16 x i8> @masked_sub_nxv16i8(<vscale x 16 x i8> %a, <vscale x 1
 define <vscale x 8 x i16> @masked_sub_nxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b, <vscale x 8 x i1> %mask) {
 ; CHECK-LABEL: masked_sub_nxv8i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z2.h, #0 // =0x0
-; CHECK-NEXT:    sel z1.h, p0, z1.h, z2.h
-; CHECK-NEXT:    sub z0.h, z0.h, z1.h
+; CHECK-NEXT:    sub z0.h, p0/m, z0.h, z1.h
 ; CHECK-NEXT:    ret
   %select = select <vscale x 8 x i1> %mask, <vscale x 8 x i16> %b, <vscale x 8 x i16> zeroinitializer
   %ret = sub <vscale x 8 x i16> %a, %select
@@ -84,9 +72,7 @@ define <vscale x 8 x i16> @masked_sub_nxv8i16(<vscale x 8 x i16> %a, <vscale x 8
 define <vscale x 4 x i32> @masked_sub_nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, <vscale x 4 x i1> %mask) {
 ; CHECK-LABEL: masked_sub_nxv4i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z2.s, #0 // =0x0
-; CHECK-NEXT:    sel z1.s, p0, z1.s, z2.s
-; CHECK-NEXT:    sub z0.s, z0.s, z1.s
+; CHECK-NEXT:    sub z0.s, p0/m, z0.s, z1.s
 ; CHECK-NEXT:    ret
   %select = select <vscale x 4 x i1> %mask, <vscale x 4 x i32> %b, <vscale x 4 x i32> zeroinitializer
   %ret = sub <vscale x 4 x i32> %a, %select
@@ -96,9 +82,7 @@ define <vscale x 4 x i32> @masked_sub_nxv4i32(<vscale x 4 x i32> %a, <vscale x 4
 define <vscale x 2 x i64> @masked_sub_nxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b, <vscale x 2 x i1> %mask) {
 ; CHECK-LABEL: masked_sub_nxv2i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov z2.d, #0 // =0x0
-; CHECK-NEXT:    sel z1.d, p0, z1.d, z2.d
-; CHECK-NEXT:    sub z0.d, z0.d, z1.d
+; CHECK-NEXT:    sub z0.d, p0/m, z0.d, z1.d
 ; CHECK-NEXT:    ret
   %select = select <vscale x 2 x i1> %mask, <vscale x 2 x i64> %b, <vscale x 2 x i64> zeroinitializer
   %ret = sub <vscale x 2 x i64> %a, %select