[AArch64] Prefer zip over ushll for anyext. (#133433)

Many CPUs have a higher throughput of ZIP instructions vs USHLL. This
adds some tablegen patterns for preferring zip in anyext patterns.

10 files changed, 45 insertions, 26 deletions

diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 6c61e3a..f291589 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -6751,6 +6751,23 @@ def : Pat<(v4i32 (concat_vectors
                  (v2i32 (trunc (AArch64vlshr (v2i64 V128:$Vm), (i32 32)))))),
           (UZP2v4i32 V128:$Vn, V128:$Vm)>;
 
+// extract_subvec(anyext) can use zip. Check for one use on the anyext, otherwise
+// the extract_subvector can be free.
+let HasOneUse = 1 in
+def anyext_oneuse: PatFrag<(ops node:$src0), (anyext $src0)>;
+def : Pat<(v4i16 (extract_subvector (v8i16 (anyext_oneuse (v8i8 V64:$Vn))), (i64 0))),
+          (ZIP1v8i8 V64:$Vn, V64:$Vn)>;
+def : Pat<(v2i32 (extract_subvector (v4i32 (anyext_oneuse (v4i16 V64:$Vn))), (i64 0))),
+          (ZIP1v4i16 V64:$Vn, V64:$Vn)>;
+def : Pat<(v1i64 (extract_subvector (v2i64 (anyext_oneuse (v2i32 V64:$Vn))), (i64 0))),
+          (ZIP1v2i32 V64:$Vn, V64:$Vn)>;
+def : Pat<(v4i16 (extract_subvector (v8i16 (anyext_oneuse (v8i8 V64:$Vn))), (i64 4))),
+          (ZIP2v8i8 V64:$Vn, V64:$Vn)>;
+def : Pat<(v2i32 (extract_subvector (v4i32 (anyext_oneuse (v4i16 V64:$Vn))), (i64 2))),
+          (ZIP2v4i16 V64:$Vn, V64:$Vn)>;
+def : Pat<(v1i64 (extract_subvector (v2i64 (anyext_oneuse (v2i32 V64:$Vn))), (i64 1))),
+          (ZIP2v2i32 V64:$Vn, V64:$Vn)>;
+
 //----------------------------------------------------------------------------
 // AdvSIMD TBL/TBX instructions
 //----------------------------------------------------------------------------
diff --git a/llvm/test/CodeGen/AArch64/andorxor.ll b/llvm/test/CodeGen/AArch64/andorxor.ll
index 24f2549..0384848 100644
--- a/llvm/test/CodeGen/AArch64/andorxor.ll
+++ b/llvm/test/CodeGen/AArch64/andorxor.ll
@@ -433,8 +433,8 @@ define void @and_v4i8(ptr %p1, ptr %p2) {
 ; CHECK-SD:       // %bb.0: // %entry
 ; CHECK-SD-NEXT:    ldr s0, [x0]
 ; CHECK-SD-NEXT:    ldr s1, [x1]
-; CHECK-SD-NEXT:    ushll v0.8h, v0.8b, #0
-; CHECK-SD-NEXT:    ushll v1.8h, v1.8b, #0
+; CHECK-SD-NEXT:    zip1 v0.8b, v0.8b, v0.8b
+; CHECK-SD-NEXT:    zip1 v1.8b, v1.8b, v1.8b
 ; CHECK-SD-NEXT:    and v0.8b, v0.8b, v1.8b
 ; CHECK-SD-NEXT:    uzp1 v0.8b, v0.8b, v0.8b
 ; CHECK-SD-NEXT:    str s0, [x0]
@@ -482,8 +482,8 @@ define void @or_v4i8(ptr %p1, ptr %p2) {
 ; CHECK-SD:       // %bb.0: // %entry
 ; CHECK-SD-NEXT:    ldr s0, [x0]
 ; CHECK-SD-NEXT:    ldr s1, [x1]
-; CHECK-SD-NEXT:    ushll v0.8h, v0.8b, #0
-; CHECK-SD-NEXT:    ushll v1.8h, v1.8b, #0
+; CHECK-SD-NEXT:    zip1 v0.8b, v0.8b, v0.8b
+; CHECK-SD-NEXT:    zip1 v1.8b, v1.8b, v1.8b
 ; CHECK-SD-NEXT:    orr v0.8b, v0.8b, v1.8b
 ; CHECK-SD-NEXT:    uzp1 v0.8b, v0.8b, v0.8b
 ; CHECK-SD-NEXT:    str s0, [x0]
@@ -531,8 +531,8 @@ define void @xor_v4i8(ptr %p1, ptr %p2) {
 ; CHECK-SD:       // %bb.0: // %entry
 ; CHECK-SD-NEXT:    ldr s0, [x0]
 ; CHECK-SD-NEXT:    ldr s1, [x1]
-; CHECK-SD-NEXT:    ushll v0.8h, v0.8b, #0
-; CHECK-SD-NEXT:    ushll v1.8h, v1.8b, #0
+; CHECK-SD-NEXT:    zip1 v0.8b, v0.8b, v0.8b
+; CHECK-SD-NEXT:    zip1 v1.8b, v1.8b, v1.8b
 ; CHECK-SD-NEXT:    eor v0.8b, v0.8b, v1.8b
 ; CHECK-SD-NEXT:    uzp1 v0.8b, v0.8b, v0.8b
 ; CHECK-SD-NEXT:    str s0, [x0]
diff --git a/llvm/test/CodeGen/AArch64/bitcast-promote-widen.ll b/llvm/test/CodeGen/AArch64/bitcast-promote-widen.ll
index 864ddc2..90fa294 100644
--- a/llvm/test/CodeGen/AArch64/bitcast-promote-widen.ll
+++ b/llvm/test/CodeGen/AArch64/bitcast-promote-widen.ll
@@ -6,8 +6,7 @@
 define <2 x i16> @bitcast_v2i16_v2f16(<2 x half> %x) {
 ; CHECK-LABEL: bitcast_v2i16_v2f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    zip1 v0.4h, v0.4h, v0.4h
 ; CHECK-NEXT:    ret
   %y = bitcast <2 x half> %x to <2 x i16>
   ret <2 x i16> %y
diff --git a/llvm/test/CodeGen/AArch64/bitcast.ll b/llvm/test/CodeGen/AArch64/bitcast.ll
index d9199ce2..d54cc4a 100644
--- a/llvm/test/CodeGen/AArch64/bitcast.ll
+++ b/llvm/test/CodeGen/AArch64/bitcast.ll
@@ -125,8 +125,7 @@ define <2 x i16> @bitcast_i32_v2i16(i32 %a, i32 %b){
 ; CHECK-SD:       // %bb.0:
 ; CHECK-SD-NEXT:    add w8, w0, w1
 ; CHECK-SD-NEXT:    fmov s0, w8
-; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
-; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT:    zip1 v0.4h, v0.4h, v0.4h
 ; CHECK-SD-NEXT:    ret
 ;
 ; CHECK-GI-LABEL: bitcast_i32_v2i16:
diff --git a/llvm/test/CodeGen/AArch64/extbinopload.ll b/llvm/test/CodeGen/AArch64/extbinopload.ll
index 72f4d58..82114d60 100644
--- a/llvm/test/CodeGen/AArch64/extbinopload.ll
+++ b/llvm/test/CodeGen/AArch64/extbinopload.ll
@@ -649,7 +649,7 @@ define <16 x i32> @extrause_load(ptr %p, ptr %q, ptr %r, ptr %s, ptr %z) {
 ; CHECK-NEXT:    add x8, x3, #8
 ; CHECK-NEXT:    add x11, x1, #12
 ; CHECK-NEXT:    str s1, [x4]
-; CHECK-NEXT:    ushll v1.8h, v1.8b, #0
+; CHECK-NEXT:    zip1 v1.8b, v1.8b, v1.8b
 ; CHECK-NEXT:    ldr s0, [x2]
 ; CHECK-NEXT:    ushll v2.8h, v0.8b, #0
 ; CHECK-NEXT:    umov w9, v2.h[0]
@@ -659,7 +659,7 @@ define <16 x i32> @extrause_load(ptr %p, ptr %q, ptr %r, ptr %s, ptr %z) {
 ; CHECK-NEXT:    mov v0.b[9], w10
 ; CHECK-NEXT:    umov w10, v2.h[3]
 ; CHECK-NEXT:    ldr s2, [x1]
-; CHECK-NEXT:    ushll v2.8h, v2.8b, #0
+; CHECK-NEXT:    zip1 v2.8b, v2.8b, v2.8b
 ; CHECK-NEXT:    mov v0.b[10], w9
 ; CHECK-NEXT:    add x9, x1, #4
 ; CHECK-NEXT:    mov v1.d[1], v2.d[0]
diff --git a/llvm/test/CodeGen/AArch64/extract-subvec-combine.ll b/llvm/test/CodeGen/AArch64/extract-subvec-combine.ll
index 75d55773..368103b 100644
--- a/llvm/test/CodeGen/AArch64/extract-subvec-combine.ll
+++ b/llvm/test/CodeGen/AArch64/extract-subvec-combine.ll
@@ -104,12 +104,19 @@ define <2 x i32> @sext_extract_zext_idx0(<4 x i16> %vec) nounwind {
 
 ; Negative test, combine should not fire if sign extension is for a different width.
 define <2 x i32> @sext_extract_zext_idx0_negtest(<4 x i16> %vec) nounwind {
-; CHECK-LABEL: sext_extract_zext_idx0_negtest:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
-; CHECK-NEXT:    shl v0.2s, v0.2s, #17
-; CHECK-NEXT:    sshr v0.2s, v0.2s, #17
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: sext_extract_zext_idx0_negtest:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    zip1 v0.4h, v0.4h, v0.4h
+; CHECK-SD-NEXT:    shl v0.2s, v0.2s, #17
+; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #17
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: sext_extract_zext_idx0_negtest:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    ushll v0.4s, v0.4h, #0
+; CHECK-GI-NEXT:    shl v0.2s, v0.2s, #17
+; CHECK-GI-NEXT:    sshr v0.2s, v0.2s, #17
+; CHECK-GI-NEXT:    ret
   %zext = zext <4 x i16> %vec to <4 x i32>
   %extract = call <2 x i32> @llvm.vector.extract.v2i32.v4i32(<4 x i32> %zext, i64 0)
   %sext_inreg_step0 = shl <2 x i32> %extract, <i32 17, i32 17>
diff --git a/llvm/test/CodeGen/AArch64/neon-bitcast.ll b/llvm/test/CodeGen/AArch64/neon-bitcast.ll
index d06612e..07772b7 100644
--- a/llvm/test/CodeGen/AArch64/neon-bitcast.ll
+++ b/llvm/test/CodeGen/AArch64/neon-bitcast.ll
@@ -518,15 +518,14 @@ define <2 x i16> @bitcast_i32_to_v2i16(i32 %word) {
 ; CHECK-LE-LABEL: bitcast_i32_to_v2i16:
 ; CHECK-LE:       // %bb.0:
 ; CHECK-LE-NEXT:    fmov s0, w0
-; CHECK-LE-NEXT:    ushll v0.4s, v0.4h, #0
-; CHECK-LE-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-LE-NEXT:    zip1 v0.4h, v0.4h, v0.4h
 ; CHECK-LE-NEXT:    ret
 ;
 ; CHECK-BE-LABEL: bitcast_i32_to_v2i16:
 ; CHECK-BE:       // %bb.0:
 ; CHECK-BE-NEXT:    fmov s0, w0
 ; CHECK-BE-NEXT:    rev32 v0.4h, v0.4h
-; CHECK-BE-NEXT:    ushll v0.4s, v0.4h, #0
+; CHECK-BE-NEXT:    zip1 v0.4h, v0.4h, v0.4h
 ; CHECK-BE-NEXT:    rev64 v0.2s, v0.2s
 ; CHECK-BE-NEXT:    ret
   %ret = bitcast i32 %word to <2 x i16>
diff --git a/llvm/test/CodeGen/AArch64/sve-fixed-length-extract-subvector.ll b/llvm/test/CodeGen/AArch64/sve-fixed-length-extract-subvector.ll
index 8fac0e1..bda7ff9 100644
--- a/llvm/test/CodeGen/AArch64/sve-fixed-length-extract-subvector.ll
+++ b/llvm/test/CodeGen/AArch64/sve-fixed-length-extract-subvector.ll
@@ -88,9 +88,7 @@ define void @extract_subvector_v256i8(ptr %a, ptr %b) vscale_range(16,0) #0 {
 define <2 x i16> @extract_subvector_v4i16(<4 x i16> %op) vscale_range(2,0) #0 {
 ; CHECK-LABEL: extract_subvector_v4i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
-; CHECK-NEXT:    ext v0.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    zip2 v0.4h, v0.4h, v0.4h
 ; CHECK-NEXT:    ret
   %ret = call <2 x i16> @llvm.vector.extract.v2i16.v4i16(<4 x i16> %op, i64 2)
   ret <2 x i16> %ret
diff --git a/llvm/test/CodeGen/AArch64/vec3-loads-ext-trunc-stores.ll b/llvm/test/CodeGen/AArch64/vec3-loads-ext-trunc-stores.ll
index b52cbfe..45b7a27 100644
--- a/llvm/test/CodeGen/AArch64/vec3-loads-ext-trunc-stores.ll
+++ b/llvm/test/CodeGen/AArch64/vec3-loads-ext-trunc-stores.ll
@@ -303,7 +303,7 @@ define <3 x i32> @load_v3i8_zext_to_3xi32(ptr %src) {
 ; BE-NEXT:    add x8, x0, #2
 ; BE-NEXT:    ldr s0, [sp, #12]
 ; BE-NEXT:    rev32 v0.8b, v0.8b
-; BE-NEXT:    ushll v0.8h, v0.8b, #0
+; BE-NEXT:    zip1 v0.8b, v0.8b, v0.8b
 ; BE-NEXT:    ld1 { v0.b }[4], [x8]
 ; BE-NEXT:    ushll v0.4s, v0.4h, #0
 ; BE-NEXT:    and v0.16b, v0.16b, v1.16b
diff --git a/llvm/test/CodeGen/AArch64/zext.ll b/llvm/test/CodeGen/AArch64/zext.ll
index e40b9cb..962486a 100644
--- a/llvm/test/CodeGen/AArch64/zext.ll
+++ b/llvm/test/CodeGen/AArch64/zext.ll
@@ -447,7 +447,7 @@ define <3 x i64> @zext_v3i10_v3i64(<3 x i10> %a) {
 ; CHECK-SD-NEXT:    mov w8, #1023 // =0x3ff
 ; CHECK-SD-NEXT:    dup v2.2d, x8
 ; CHECK-SD-NEXT:    mov v0.s[1], w1
-; CHECK-SD-NEXT:    ushll v3.2d, v1.2s, #0
+; CHECK-SD-NEXT:    zip1 v3.2s, v1.2s, v1.2s
 ; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
 ; CHECK-SD-NEXT:    and v0.16b, v0.16b, v2.16b
 ; CHECK-SD-NEXT:    and v2.8b, v3.8b, v2.8b