[llvm] [AArch64][GlobalISel] Select SHL({Z|S}EXT, DUP Imm) into {U|S}HLL Imm (PR #96782)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Aug 5 05:56:22 PDT 2024
https://github.com/chuongg3 updated https://github.com/llvm/llvm-project/pull/96782
>From 452e6a291a797cbb4e485ba3e9f42057541bf78b Mon Sep 17 00:00:00 2001
From: Tuan Chuong Goh <chuong.goh at arm.com>
Date: Thu, 20 Jun 2024 12:40:51 +0000
Subject: [PATCH] [AArch64][GlobalISel] Select SHL(ZEXT, DUP imm) into {U/S}HLL
imm
---
.../lib/Target/AArch64/AArch64InstrFormats.td | 7 ++
llvm/lib/Target/AArch64/AArch64InstrInfo.td | 14 +++
.../CodeGen/AArch64/neon-shift-left-long.ll | 108 ++++++------------
3 files changed, 57 insertions(+), 72 deletions(-)
diff --git a/llvm/lib/Target/AArch64/AArch64InstrFormats.td b/llvm/lib/Target/AArch64/AArch64InstrFormats.td
index e1ecc5a57dd26..ab8251dc83014 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrFormats.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrFormats.td
@@ -1108,6 +1108,13 @@ def timm32_1_7 : Operand<i32>, TImmLeaf<i32, [{
let ParserMatchClass = Imm1_7Operand;
}
+// imm32_0_7 predicate - True if the 32-bit immediate is in the range [0,7]
+def imm32_0_7 : Operand<i32>, ImmLeaf<i32, [{
+ return ((uint32_t)Imm) < 8;
+}]> {
+ let ParserMatchClass = Imm0_7Operand;
+}
+
// imm32_0_15 predicate - True if the 32-bit immediate is in the range [0,15]
def imm32_0_15 : Operand<i32>, ImmLeaf<i32, [{
return ((uint32_t)Imm) < 16;
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 1053ba9242768..e720c6b21a97b 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -8129,6 +8129,20 @@ def : Pat<(v4i32 (concat_vectors (v2i32 V64:$Rd),
(SHRNv4i32_shift (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
V128:$Rn, vecshiftR32Narrow:$imm)>;
+def : Pat<(shl (v8i16 (zext (v8i8 V64:$Rm))), (v8i16 (AArch64dup (i32 imm32_0_7:$size)))),
+ (USHLLv8i8_shift V64:$Rm, (i32 imm32_0_7:$size))>;
+def : Pat<(shl (v4i32 (zext (v4i16 V64:$Rm))), (v4i32 (AArch64dup (i32 imm32_0_15:$size)))),
+ (USHLLv4i16_shift V64:$Rm, (i32 imm32_0_15:$size))>;
+def : Pat<(shl (v2i64 (zext (v2i32 V64:$Rm))), (v2i64 (AArch64dup (i64 imm0_31:$size)))),
+ (USHLLv2i32_shift V64:$Rm, (trunc_imm imm0_31:$size))>;
+
+def : Pat<(shl (v8i16 (sext (v8i8 V64:$Rm))), (v8i16 (AArch64dup (i32 imm32_0_7:$size)))),
+ (SSHLLv8i8_shift V64:$Rm, (i32 imm32_0_7:$size))>;
+def : Pat<(shl (v4i32 (sext (v4i16 V64:$Rm))), (v4i32 (AArch64dup (i32 imm32_0_15:$size)))),
+ (SSHLLv4i16_shift V64:$Rm, (i32 imm32_0_15:$size))>;
+def : Pat<(shl (v2i64 (sext (v2i32 V64:$Rm))), (v2i64 (AArch64dup (i64 imm0_31:$size)))),
+ (SSHLLv2i32_shift V64:$Rm, (trunc_imm imm0_31:$size))>;
+
// Vector sign and zero extensions are implemented with SSHLL and USSHLL.
// Anyexts are implemented as zexts.
def : Pat<(v8i16 (sext (v8i8 V64:$Rn))), (SSHLLv8i8_shift V64:$Rn, (i32 0))>;
diff --git a/llvm/test/CodeGen/AArch64/neon-shift-left-long.ll b/llvm/test/CodeGen/AArch64/neon-shift-left-long.ll
index fab3776fed086..282f43763f362 100644
--- a/llvm/test/CodeGen/AArch64/neon-shift-left-long.ll
+++ b/llvm/test/CodeGen/AArch64/neon-shift-left-long.ll
@@ -3,16 +3,10 @@
; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -mattr=+neon -global-isel | FileCheck %s --check-prefixes=CHECK,CHECK-GI
define <8 x i16> @test_sshll_v8i8(<8 x i8> %a) {
-; CHECK-SD-LABEL: test_sshll_v8i8:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: sshll v0.8h, v0.8b, #3
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: test_sshll_v8i8:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: sshll v0.8h, v0.8b, #0
-; CHECK-GI-NEXT: shl v0.8h, v0.8h, #3
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: test_sshll_v8i8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: sshll v0.8h, v0.8b, #3
+; CHECK-NEXT: ret
%1 = sext <8 x i8> %a to <8 x i16>
%tmp = shl <8 x i16> %1, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
ret <8 x i16> %tmp
@@ -36,16 +30,10 @@ define <8 x i16> @test_sshll_v8i8_big(<8 x i8> %a) {
}
define <4 x i32> @test_sshll_v4i16(<4 x i16> %a) {
-; CHECK-SD-LABEL: test_sshll_v4i16:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: sshll v0.4s, v0.4h, #9
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: test_sshll_v4i16:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: sshll v0.4s, v0.4h, #0
-; CHECK-GI-NEXT: shl v0.4s, v0.4s, #9
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: test_sshll_v4i16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: sshll v0.4s, v0.4h, #9
+; CHECK-NEXT: ret
%1 = sext <4 x i16> %a to <4 x i32>
%tmp = shl <4 x i32> %1, <i32 9, i32 9, i32 9, i32 9>
ret <4 x i32> %tmp
@@ -69,16 +57,10 @@ define <4 x i32> @test_sshll_v4i16_big(<4 x i16> %a) {
}
define <2 x i64> @test_sshll_v2i32(<2 x i32> %a) {
-; CHECK-SD-LABEL: test_sshll_v2i32:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: sshll v0.2d, v0.2s, #19
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: test_sshll_v2i32:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: sshll v0.2d, v0.2s, #0
-; CHECK-GI-NEXT: shl v0.2d, v0.2d, #19
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: test_sshll_v2i32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: sshll v0.2d, v0.2s, #19
+; CHECK-NEXT: ret
%1 = sext <2 x i32> %a to <2 x i64>
%tmp = shl <2 x i64> %1, <i64 19, i64 19>
ret <2 x i64> %tmp
@@ -102,16 +84,10 @@ define <2 x i64> @test_sshll_v2i32_big(<2 x i32> %a) {
}
define <8 x i16> @test_ushll_v8i8(<8 x i8> %a) {
-; CHECK-SD-LABEL: test_ushll_v8i8:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: ushll v0.8h, v0.8b, #3
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: test_ushll_v8i8:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: ushll v0.8h, v0.8b, #0
-; CHECK-GI-NEXT: shl v0.8h, v0.8h, #3
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: test_ushll_v8i8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ushll v0.8h, v0.8b, #3
+; CHECK-NEXT: ret
%1 = zext <8 x i8> %a to <8 x i16>
%tmp = shl <8 x i16> %1, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
ret <8 x i16> %tmp
@@ -129,16 +105,10 @@ define <8 x i16> @test_ushll_v8i8_big(<8 x i8> %a) {
}
define <4 x i32> @test_ushll_v4i16(<4 x i16> %a) {
-; CHECK-SD-LABEL: test_ushll_v4i16:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: ushll v0.4s, v0.4h, #9
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: test_ushll_v4i16:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: ushll v0.4s, v0.4h, #0
-; CHECK-GI-NEXT: shl v0.4s, v0.4s, #9
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: test_ushll_v4i16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ushll v0.4s, v0.4h, #9
+; CHECK-NEXT: ret
%1 = zext <4 x i16> %a to <4 x i32>
%tmp = shl <4 x i32> %1, <i32 9, i32 9, i32 9, i32 9>
ret <4 x i32> %tmp
@@ -156,16 +126,10 @@ define <4 x i32> @test_ushll_v4i16_big(<4 x i16> %a) {
}
define <2 x i64> @test_ushll_v2i32(<2 x i32> %a) {
-; CHECK-SD-LABEL: test_ushll_v2i32:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: ushll v0.2d, v0.2s, #19
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: test_ushll_v2i32:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: ushll v0.2d, v0.2s, #0
-; CHECK-GI-NEXT: shl v0.2d, v0.2d, #19
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: test_ushll_v2i32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ushll v0.2d, v0.2s, #19
+; CHECK-NEXT: ret
%1 = zext <2 x i32> %a to <2 x i64>
%tmp = shl <2 x i64> %1, <i64 19, i64 19>
ret <2 x i64> %tmp
@@ -190,8 +154,8 @@ define <8 x i16> @test_sshll2_v16i8(<16 x i8> %a) {
;
; CHECK-GI-LABEL: test_sshll2_v16i8:
; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: sshll2 v0.8h, v0.16b, #0
-; CHECK-GI-NEXT: shl v0.8h, v0.8h, #3
+; CHECK-GI-NEXT: mov d0, v0.d[1]
+; CHECK-GI-NEXT: sshll v0.8h, v0.8b, #3
; CHECK-GI-NEXT: ret
%1 = shufflevector <16 x i8> %a, <16 x i8> undef, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
%2 = sext <8 x i8> %1 to <8 x i16>
@@ -225,8 +189,8 @@ define <4 x i32> @test_sshll2_v8i16(<8 x i16> %a) {
;
; CHECK-GI-LABEL: test_sshll2_v8i16:
; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: sshll2 v0.4s, v0.8h, #0
-; CHECK-GI-NEXT: shl v0.4s, v0.4s, #9
+; CHECK-GI-NEXT: mov d0, v0.d[1]
+; CHECK-GI-NEXT: sshll v0.4s, v0.4h, #9
; CHECK-GI-NEXT: ret
%1 = shufflevector <8 x i16> %a, <8 x i16> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
%2 = sext <4 x i16> %1 to <4 x i32>
@@ -260,8 +224,8 @@ define <2 x i64> @test_sshll2_v4i32(<4 x i32> %a) {
;
; CHECK-GI-LABEL: test_sshll2_v4i32:
; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: sshll2 v0.2d, v0.4s, #0
-; CHECK-GI-NEXT: shl v0.2d, v0.2d, #19
+; CHECK-GI-NEXT: mov d0, v0.d[1]
+; CHECK-GI-NEXT: sshll v0.2d, v0.2s, #19
; CHECK-GI-NEXT: ret
%1 = shufflevector <4 x i32> %a, <4 x i32> undef, <2 x i32> <i32 2, i32 3>
%2 = sext <2 x i32> %1 to <2 x i64>
@@ -295,8 +259,8 @@ define <8 x i16> @test_ushll2_v16i8(<16 x i8> %a) {
;
; CHECK-GI-LABEL: test_ushll2_v16i8:
; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: ushll2 v0.8h, v0.16b, #0
-; CHECK-GI-NEXT: shl v0.8h, v0.8h, #3
+; CHECK-GI-NEXT: mov d0, v0.d[1]
+; CHECK-GI-NEXT: ushll v0.8h, v0.8b, #3
; CHECK-GI-NEXT: ret
%1 = shufflevector <16 x i8> %a, <16 x i8> undef, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
%2 = zext <8 x i8> %1 to <8 x i16>
@@ -324,8 +288,8 @@ define <4 x i32> @test_ushll2_v8i16(<8 x i16> %a) {
;
; CHECK-GI-LABEL: test_ushll2_v8i16:
; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: ushll2 v0.4s, v0.8h, #0
-; CHECK-GI-NEXT: shl v0.4s, v0.4s, #9
+; CHECK-GI-NEXT: mov d0, v0.d[1]
+; CHECK-GI-NEXT: ushll v0.4s, v0.4h, #9
; CHECK-GI-NEXT: ret
%1 = shufflevector <8 x i16> %a, <8 x i16> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
%2 = zext <4 x i16> %1 to <4 x i32>
@@ -353,8 +317,8 @@ define <2 x i64> @test_ushll2_v4i32(<4 x i32> %a) {
;
; CHECK-GI-LABEL: test_ushll2_v4i32:
; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: ushll2 v0.2d, v0.4s, #0
-; CHECK-GI-NEXT: shl v0.2d, v0.2d, #19
+; CHECK-GI-NEXT: mov d0, v0.d[1]
+; CHECK-GI-NEXT: ushll v0.2d, v0.2s, #19
; CHECK-GI-NEXT: ret
%1 = shufflevector <4 x i32> %a, <4 x i32> undef, <2 x i32> <i32 2, i32 3>
%2 = zext <2 x i32> %1 to <2 x i64>
More information about the llvm-commits
mailing list