[AArch64] Expand UADDLV patterns to SADDLV

We already had some patterns for UADDV(UADDLP(x)) -> UADDLV(x), this
simply expands them to the signed instructions by re-using the tablegen
patterns.

Differential Revision: https://reviews.llvm.org/D118133

diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 8dc91d6..42c1afc 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -5860,24 +5860,28 @@ defm FMAXV   : SIMDFPAcrossLanes<0b01111, 0, "fmaxv", int_aarch64_neon_fmaxv>;
 defm FMINNMV : SIMDFPAcrossLanes<0b01100, 1, "fminnmv", int_aarch64_neon_fminnmv>;
 defm FMINV   : SIMDFPAcrossLanes<0b01111, 1, "fminv", int_aarch64_neon_fminv>;
 
-// Patterns for uaddv(uaddlp(x)) ==> uaddlv
-def : Pat<(i32 (vector_extract (v8i16 (insert_subvector undef,
-            (v4i16 (AArch64uaddv (v4i16 (AArch64uaddlp (v8i8 V64:$op))))),
-            (i64 0))), (i64 0))),
-          (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
-           (UADDLVv8i8v V64:$op), hsub), ssub)>;
-def : Pat<(i32 (vector_extract (v8i16 (AArch64uaddv (v8i16 (AArch64uaddlp
-           (v16i8 V128:$op))))), (i64 0))),
-          (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
-           (UADDLVv16i8v V128:$op), hsub), ssub)>;
-def : Pat<(v4i32 (AArch64uaddv (v4i32 (AArch64uaddlp (v8i16 V128:$op))))),
-          (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), (UADDLVv8i16v V128:$op), ssub)>;
-
-// Patterns for addp(uaddlp(x))) ==> uaddlv
-def : Pat<(v2i32 (AArch64uaddv (v2i32 (AArch64uaddlp (v4i16 V64:$op))))),
-          (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)), (UADDLVv4i16v V64:$op), ssub)>;
-def : Pat<(v2i64 (AArch64uaddv (v2i64 (AArch64uaddlp (v4i32 V128:$op))))),
-          (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (UADDLVv4i32v V128:$op), dsub)>;
+multiclass SIMDAcrossLaneLongPairIntrinsic<string Opc, SDPatternOperator addlp> {
+  // Patterns for addv(addlp(x)) ==> addlv
+  def : Pat<(i32 (vector_extract (v8i16 (insert_subvector undef,
+              (v4i16 (AArch64uaddv (v4i16 (addlp (v8i8 V64:$op))))),
+              (i64 0))), (i64 0))),
+            (EXTRACT_SUBREG (INSERT_SUBREG (v4i16 (IMPLICIT_DEF)),
+              (!cast<Instruction>(Opc#"v8i8v") V64:$op), hsub), ssub)>;
+  def : Pat<(i32 (vector_extract (v8i16 (AArch64uaddv (v8i16 (addlp (v16i8 V128:$op))))), (i64 0))),
+            (EXTRACT_SUBREG (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
+              (!cast<Instruction>(Opc#"v16i8v") V128:$op), hsub), ssub)>;
+  def : Pat<(v4i32 (AArch64uaddv (v4i32 (addlp (v8i16 V128:$op))))),
+            (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), (!cast<Instruction>(Opc#"v8i16v") V128:$op), ssub)>;
+
+  // Patterns for addp(addlp(x))) ==> addlv
+  def : Pat<(v2i32 (AArch64uaddv (v2i32 (addlp (v4i16 V64:$op))))),
+            (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)), (!cast<Instruction>(Opc#"v4i16v") V64:$op), ssub)>;
+  def : Pat<(v2i64 (AArch64uaddv (v2i64 (addlp (v4i32 V128:$op))))),
+            (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), (!cast<Instruction>(Opc#"v4i32v") V128:$op), dsub)>;
+}
+
+defm : SIMDAcrossLaneLongPairIntrinsic<"UADDLV", AArch64uaddlp>;
+defm : SIMDAcrossLaneLongPairIntrinsic<"SADDLV", AArch64saddlp>;
 
 // Patterns for across-vector intrinsics, that have a node equivalent, that
 // returns a vector (with only the low lane defined) instead of a scalar.

diff --git a/llvm/test/CodeGen/AArch64/neon-addlv.ll b/llvm/test/CodeGen/AArch64/neon-addlv.ll
index bf23e94..1e4931e 100644 (file)
--- a/llvm/test/CodeGen/AArch64/neon-addlv.ll
+++ b/llvm/test/CodeGen/AArch64/neon-addlv.ll
@@ -90,8 +90,7 @@ define i16 @saddlv4h_from_v8i8(<8 x i8>* %A) nounwind {
 ; CHECK-LABEL: saddlv4h_from_v8i8:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    saddlp v0.4h, v0.8b
-; CHECK-NEXT:    addv h0, v0.4h
+; CHECK-NEXT:    saddlv h0, v0.8b
 ; CHECK-NEXT:    fmov w0, s0
 ; CHECK-NEXT:    ret
   %tmp1 = load <8 x i8>, <8 x i8>* %A
@@ -104,8 +103,7 @@ define i16 @saddlv16b_from_v16i8(<16 x i8>* %A) nounwind {
 ; CHECK-LABEL: saddlv16b_from_v16i8:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ldr q0, [x0]
-; CHECK-NEXT:    saddlp v0.8h, v0.16b
-; CHECK-NEXT:    addv h0, v0.8h
+; CHECK-NEXT:    saddlv h0, v0.16b
 ; CHECK-NEXT:    fmov w0, s0
 ; CHECK-NEXT:    ret
   %tmp1 = load <16 x i8>, <16 x i8>* %A
@@ -118,8 +116,7 @@ define i32 @saddlv8h_from_v8i16(<8 x i16>* %A) nounwind {
 ; CHECK-LABEL: saddlv8h_from_v8i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ldr q0, [x0]
-; CHECK-NEXT:    saddlp v0.4s, v0.8h
-; CHECK-NEXT:    addv s0, v0.4s
+; CHECK-NEXT:    saddlv s0, v0.8h
 ; CHECK-NEXT:    fmov w0, s0
 ; CHECK-NEXT:    ret
   %tmp1 = load <8 x i16>, <8 x i16>* %A
@@ -132,8 +129,7 @@ define i64 @saddlv4s_from_v4i32(<4 x i32>* %A) nounwind {
 ; CHECK-LABEL: saddlv4s_from_v4i32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ldr q0, [x0]
-; CHECK-NEXT:    saddlp v0.2d, v0.4s
-; CHECK-NEXT:    addp d0, v0.2d
+; CHECK-NEXT:    saddlv d0, v0.4s
 ; CHECK-NEXT:    fmov x0, d0
 ; CHECK-NEXT:    ret
   %tmp1 = load <4 x i32>, <4 x i32>* %A
@@ -146,8 +142,7 @@ define i32 @saddlv4h_from_v4i16(<4 x i16>* %A) nounwind {
 ; CHECK-LABEL: saddlv4h_from_v4i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    saddlp v0.2s, v0.4h
-; CHECK-NEXT:    addp v0.2s, v0.2s, v0.2s
+; CHECK-NEXT:    saddlv s0, v0.4h
 ; CHECK-NEXT:    fmov w0, s0
 ; CHECK-NEXT:    ret
   %tmp1 = load <4 x i16>, <4 x i16>* %A

diff --git a/llvm/test/CodeGen/AArch64/vecreduce-add.ll b/llvm/test/CodeGen/AArch64/vecreduce-add.ll
index 273f2e0..3b946b8 100644 (file)
--- a/llvm/test/CodeGen/AArch64/vecreduce-add.ll
+++ b/llvm/test/CodeGen/AArch64/vecreduce-add.ll
@@ -28,8 +28,7 @@ entry:
 define i64 @add_v4i32_v4i64_sext(<4 x i32> %x) {
 ; CHECK-LABEL: add_v4i32_v4i64_sext:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    saddlp v0.2d, v0.4s
-; CHECK-NEXT:    addp d0, v0.2d
+; CHECK-NEXT:    saddlv d0, v0.4s
 ; CHECK-NEXT:    fmov x0, d0
 ; CHECK-NEXT:    ret
 entry:
@@ -79,8 +78,7 @@ entry:
 define i32 @add_v8i16_v8i32_sext(<8 x i16> %x) {
 ; CHECK-LABEL: add_v8i16_v8i32_sext:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    saddlp v0.4s, v0.8h
-; CHECK-NEXT:    addv s0, v0.4s
+; CHECK-NEXT:    saddlv s0, v0.8h
 ; CHECK-NEXT:    fmov w0, s0
 ; CHECK-NEXT:    ret
 entry:
@@ -177,8 +175,7 @@ define i64 @add_v4i16_v4i64_sext(<4 x i16> %x) {
 ; CHECK-LABEL: add_v4i16_v4i64_sext:
 ; CHECK:       // %bb.0: // %entry
 ; CHECK-NEXT:    sshll v0.4s, v0.4h, #0
-; CHECK-NEXT:    saddlp v0.2d, v0.4s
-; CHECK-NEXT:    addp d0, v0.2d
+; CHECK-NEXT:    saddlv d0, v0.4s
 ; CHECK-NEXT:    fmov x0, d0
 ; CHECK-NEXT:    ret
 entry:
@@ -295,8 +292,7 @@ define i32 @add_v8i8_v8i32_sext(<8 x i8> %x) {
 ; CHECK-BASE-LABEL: add_v8i8_v8i32_sext:
 ; CHECK-BASE:       // %bb.0: // %entry
 ; CHECK-BASE-NEXT:    sshll v0.8h, v0.8b, #0
-; CHECK-BASE-NEXT:    saddlp v0.4s, v0.8h
-; CHECK-BASE-NEXT:    addv s0, v0.4s
+; CHECK-BASE-NEXT:    saddlv s0, v0.8h
 ; CHECK-BASE-NEXT:    fmov w0, s0
 ; CHECK-BASE-NEXT:    ret
 ;
@@ -596,8 +592,7 @@ entry:
 define i64 @add_v4i32_v4i64_acc_sext(<4 x i32> %x, i64 %a) {
 ; CHECK-LABEL: add_v4i32_v4i64_acc_sext:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    saddlp v0.2d, v0.4s
-; CHECK-NEXT:    addp d0, v0.2d
+; CHECK-NEXT:    saddlv d0, v0.4s
 ; CHECK-NEXT:    fmov x8, d0
 ; CHECK-NEXT:    add x0, x8, x0
 ; CHECK-NEXT:    ret
@@ -655,8 +650,7 @@ entry:
 define i32 @add_v8i16_v8i32_acc_sext(<8 x i16> %x, i32 %a) {
 ; CHECK-LABEL: add_v8i16_v8i32_acc_sext:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    saddlp v0.4s, v0.8h
-; CHECK-NEXT:    addv s0, v0.4s
+; CHECK-NEXT:    saddlv s0, v0.8h
 ; CHECK-NEXT:    fmov w8, s0
 ; CHECK-NEXT:    add w0, w8, w0
 ; CHECK-NEXT:    ret
@@ -768,8 +762,7 @@ define i64 @add_v4i16_v4i64_acc_sext(<4 x i16> %x, i64 %a) {
 ; CHECK-LABEL: add_v4i16_v4i64_acc_sext:
 ; CHECK:       // %bb.0: // %entry
 ; CHECK-NEXT:    sshll v0.4s, v0.4h, #0
-; CHECK-NEXT:    saddlp v0.2d, v0.4s
-; CHECK-NEXT:    addp d0, v0.2d
+; CHECK-NEXT:    saddlv d0, v0.4s
 ; CHECK-NEXT:    fmov x8, d0
 ; CHECK-NEXT:    add x0, x8, x0
 ; CHECK-NEXT:    ret
@@ -901,8 +894,7 @@ define i32 @add_v8i8_v8i32_acc_sext(<8 x i8> %x, i32 %a) {
 ; CHECK-BASE-LABEL: add_v8i8_v8i32_acc_sext:
 ; CHECK-BASE:       // %bb.0: // %entry
 ; CHECK-BASE-NEXT:    sshll v0.8h, v0.8b, #0
-; CHECK-BASE-NEXT:    saddlp v0.4s, v0.8h
-; CHECK-BASE-NEXT:    addv s0, v0.4s
+; CHECK-BASE-NEXT:    saddlv s0, v0.8h
 ; CHECK-BASE-NEXT:    fmov w8, s0
 ; CHECK-BASE-NEXT:    add w0, w8, w0
 ; CHECK-BASE-NEXT:    ret
@@ -974,8 +966,7 @@ entry:
 define signext i16 @add_v16i8_v16i16_acc_sext(<16 x i8> %x, i16 %a) {
 ; CHECK-LABEL: add_v16i8_v16i16_acc_sext:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    saddlp v0.8h, v0.16b
-; CHECK-NEXT:    addv h0, v0.8h
+; CHECK-NEXT:    saddlv h0, v0.16b
 ; CHECK-NEXT:    fmov w8, s0
 ; CHECK-NEXT:    add w8, w8, w0
 ; CHECK-NEXT:    sxth w0, w8
@@ -1695,10 +1686,8 @@ entry:
 define signext i16 @add_pair_v16i8_v16i16_sext(<16 x i8> %x, <16 x i8> %y) {
 ; CHECK-LABEL: add_pair_v16i8_v16i16_sext:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    saddlp v0.8h, v0.16b
-; CHECK-NEXT:    saddlp v1.8h, v1.16b
-; CHECK-NEXT:    addv h0, v0.8h
-; CHECK-NEXT:    addv h1, v1.8h
+; CHECK-NEXT:    saddlv h0, v0.16b
+; CHECK-NEXT:    saddlv h1, v1.16b
 ; CHECK-NEXT:    fmov w8, s0
 ; CHECK-NEXT:    fmov w9, s1
 ; CHECK-NEXT:    add w8, w8, w9