This is an archive of the discontinued LLVM Phabricator instance.

Differential D105600

[WebAssembly] Optimize out shift masks
ClosedPublic

Authored by tlively on Jul 7 2021, 4:57 PM.

Details

Reviewers
aheejin
dschuff
Commits
rGf8c5a4c67075: [WebAssembly] Optimize out shift masks
Summary

WebAssembly's shift instructions implicitly masks the shift count, so optimize
out redundant explicit masks of the shift count. For vector shifts, this
currently only works if the mask is applied before splatting the shift count,
but this should be addressed in a future commit. Resolves PR49655.

Diff Detail

Event Timeline

tlively created this revision.Jul 7 2021, 4:57 PM
Herald added subscribers: wingo, ecnelises, sunfish and 3 others. · View Herald TranscriptJul 7 2021, 4:57 PM
tlively requested review of this revision.Jul 7 2021, 4:57 PM
Herald added a project: Restricted Project. · View Herald TranscriptJul 7 2021, 4:57 PM
Herald added a subscriber: llvm-commits. · View Herald Transcript
Harbormaster completed remote builds in B112894: Diff 357106.Jul 7 2021, 5:32 PM
aheejin accepted this revision.Jul 7 2021, 10:02 PM
aheejin added inline comments.
llvm/test/CodeGen/WebAssembly/masked-shifts.ll
151

Are these _late functions yet to be optimized? If so How about adding TODO do make sure they are currently unoptimized?

This revision is now accepted and ready to land.Jul 7 2021, 10:02 PM
tlively added inline comments.Jul 7 2021, 11:06 PM
llvm/test/CodeGen/WebAssembly/masked-shifts.ll
151

Sounds good 👍

Closed by commit rGf8c5a4c67075: [WebAssembly] Optimize out shift masks (authored by tlively). · Explain WhyJul 7 2021, 11:14 PM
This revision was automatically updated to reflect the committed changes.
tlively added a commit: rGf8c5a4c67075: [WebAssembly] Optimize out shift masks.

Revision Contents

PathSize
llvm/
lib/
Target/
WebAssembly/
8 lines
29 lines
test/
CodeGen/
WebAssembly/
531 lines

Diff 357137

llvm/lib/Target/WebAssembly/WebAssemblyInstrInteger.td

Show First 20 Lines • Show All 87 Lines • ▼ Show 20 Lines
defm EQZ_I32 : I<(outs I32:$dst), (ins I32:$src), (outs), (ins), defm EQZ_I32 : I<(outs I32:$dst), (ins I32:$src), (outs), (ins),
[(set I32:$dst, (setcc I32:$src, 0, SETEQ))], [(set I32:$dst, (setcc I32:$src, 0, SETEQ))],
"i32.eqz \t$dst, $src", "i32.eqz", 0x45>; "i32.eqz \t$dst, $src", "i32.eqz", 0x45>;
defm EQZ_I64 : I<(outs I32:$dst), (ins I64:$src), (outs), (ins), defm EQZ_I64 : I<(outs I32:$dst), (ins I64:$src), (outs), (ins),
[(set I32:$dst, (setcc I64:$src, 0, SETEQ))], [(set I32:$dst, (setcc I64:$src, 0, SETEQ))],
"i64.eqz \t$dst, $src", "i64.eqz", 0x50>; "i64.eqz \t$dst, $src", "i64.eqz", 0x50>;
// Optimize away an explicit mask on a shift count.
def : Pat<(shl I32:$lhs, (and I32:$rhs, 31)), (SHL_I32 I32:$lhs, I32:$rhs)>;
def : Pat<(sra I32:$lhs, (and I32:$rhs, 31)), (SHR_S_I32 I32:$lhs, I32:$rhs)>;
def : Pat<(srl I32:$lhs, (and I32:$rhs, 31)), (SHR_U_I32 I32:$lhs, I32:$rhs)>;
def : Pat<(shl I64:$lhs, (and I64:$rhs, 63)), (SHL_I64 I64:$lhs, I64:$rhs)>;
def : Pat<(sra I64:$lhs, (and I64:$rhs, 63)), (SHR_S_I64 I64:$lhs, I64:$rhs)>;
def : Pat<(srl I64:$lhs, (and I64:$rhs, 63)), (SHR_U_I64 I64:$lhs, I64:$rhs)>;
// Optimize away an explicit mask on a rotate count. // Optimize away an explicit mask on a rotate count.
def : Pat<(rotl I32:$lhs, (and I32:$rhs, 31)), (ROTL_I32 I32:$lhs, I32:$rhs)>; def : Pat<(rotl I32:$lhs, (and I32:$rhs, 31)), (ROTL_I32 I32:$lhs, I32:$rhs)>;
def : Pat<(rotr I32:$lhs, (and I32:$rhs, 31)), (ROTR_I32 I32:$lhs, I32:$rhs)>; def : Pat<(rotr I32:$lhs, (and I32:$rhs, 31)), (ROTR_I32 I32:$lhs, I32:$rhs)>;
def : Pat<(rotl I64:$lhs, (and I64:$rhs, 63)), (ROTL_I64 I64:$lhs, I64:$rhs)>; def : Pat<(rotl I64:$lhs, (and I64:$rhs, 63)), (ROTL_I64 I64:$lhs, I64:$rhs)>;
def : Pat<(rotr I64:$lhs, (and I64:$rhs, 63)), (ROTR_I64 I64:$lhs, I64:$rhs)>; def : Pat<(rotr I64:$lhs, (and I64:$rhs, 63)), (ROTR_I64 I64:$lhs, I64:$rhs)>;
defm SELECT_I32 : I<(outs I32:$dst), (ins I32:$lhs, I32:$rhs, I32:$cond), defm SELECT_I32 : I<(outs I32:$dst), (ins I32:$lhs, I32:$rhs, I32:$cond),
(outs), (ins), (outs), (ins),
Show All 20 Lines

llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td

Show First 20 Lines • Show All 892 Lines • ▼ Show 20 Lines
// Left shift by scalar: shl // Left shift by scalar: shl
defm SHL : SIMDShiftInt<wasm_shl, "shl", 107>; defm SHL : SIMDShiftInt<wasm_shl, "shl", 107>;
// Right shift by scalar: shr_s / shr_u // Right shift by scalar: shr_s / shr_u
defm SHR_S : SIMDShiftInt<wasm_shr_s, "shr_s", 108>; defm SHR_S : SIMDShiftInt<wasm_shr_s, "shr_s", 108>;
defm SHR_U : SIMDShiftInt<wasm_shr_u, "shr_u", 109>; defm SHR_U : SIMDShiftInt<wasm_shr_u, "shr_u", 109>;
// Optimize away an explicit mask on a shift count.
def : Pat<(wasm_shl (v16i8 V128:$lhs), (and I32:$rhs, 7)),
(SHL_I8x16 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_s (v16i8 V128:$lhs), (and I32:$rhs, 7)),
(SHR_S_I8x16 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_u (v16i8 V128:$lhs), (and I32:$rhs, 7)),
(SHR_U_I8x16 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shl (v8i16 V128:$lhs), (and I32:$rhs, 15)),
(SHL_I16x8 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_s (v8i16 V128:$lhs), (and I32:$rhs, 15)),
(SHR_S_I16x8 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_u (v8i16 V128:$lhs), (and I32:$rhs, 15)),
(SHR_U_I16x8 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shl (v4i32 V128:$lhs), (and I32:$rhs, 31)),
(SHL_I32x4 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_s (v4i32 V128:$lhs), (and I32:$rhs, 31)),
(SHR_S_I32x4 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_u (v4i32 V128:$lhs), (and I32:$rhs, 31)),
(SHR_U_I32x4 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shl (v2i64 V128:$lhs), (trunc (and I64:$rhs, 63))),
(SHL_I64x2 V128:$lhs, (I32_WRAP_I64 I64:$rhs))>;
def : Pat<(wasm_shr_s (v2i64 V128:$lhs), (trunc (and I64:$rhs, 63))),
(SHR_S_I64x2 V128:$lhs, (I32_WRAP_I64 I64:$rhs))>;
def : Pat<(wasm_shr_u (v2i64 V128:$lhs), (trunc (and I64:$rhs, 63))),
(SHR_U_I64x2 V128:$lhs, (I32_WRAP_I64 I64:$rhs))>;
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
// Integer binary arithmetic // Integer binary arithmetic
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
multiclass SIMDBinaryIntNoI8x16<SDPatternOperator node, string name, bits<32> baseInst> { multiclass SIMDBinaryIntNoI8x16<SDPatternOperator node, string name, bits<32> baseInst> {
defm "" : SIMDBinary<I16x8, node, name, !add(baseInst, 32)>; defm "" : SIMDBinary<I16x8, node, name, !add(baseInst, 32)>;
defm "" : SIMDBinary<I32x4, node, name, !add(baseInst, 64)>; defm "" : SIMDBinary<I32x4, node, name, !add(baseInst, 64)>;
defm "" : SIMDBinary<I64x2, node, name, !add(baseInst, 96)>; defm "" : SIMDBinary<I64x2, node, name, !add(baseInst, 96)>;
▲ Show 20 Lines • Show All 365 Lines • Show Last 20 Lines

llvm/test/CodeGen/WebAssembly/masked-shifts.ll

  • This file was added.
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -verify-machineinstrs -mattr=+simd128 | FileCheck %s
;; Check that masked shift counts are optimized out.
;; TODO: optimize the *_late functions.
target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
target triple = "wasm32-unknown-unknown"
define i32 @shl_i32(i32 %v, i32 %x) {
; CHECK-LABEL: shl_i32:
; CHECK: .functype shl_i32 (i32, i32) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32.shl
; CHECK-NEXT: # fallthrough-return
%m = and i32 %x, 31
%a = shl i32 %v, %m
ret i32 %a
}
define i32 @sra_i32(i32 %v, i32 %x) {
; CHECK-LABEL: sra_i32:
; CHECK: .functype sra_i32 (i32, i32) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32.shr_s
; CHECK-NEXT: # fallthrough-return
%m = and i32 %x, 31
%a = ashr i32 %v, %m
ret i32 %a
}
define i32 @srl_i32(i32 %v, i32 %x) {
; CHECK-LABEL: srl_i32:
; CHECK: .functype srl_i32 (i32, i32) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32.shr_u
; CHECK-NEXT: # fallthrough-return
%m = and i32 %x, 31
%a = lshr i32 %v, %m
ret i32 %a
}
define i64 @shl_i64(i64 %v, i64 %x) {
; CHECK-LABEL: shl_i64:
; CHECK: .functype shl_i64 (i64, i64) -> (i64)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.shl
; CHECK-NEXT: # fallthrough-return
%m = and i64 %x, 63
%a = shl i64 %v, %m
ret i64 %a
}
define i64 @sra_i64(i64 %v, i64 %x) {
; CHECK-LABEL: sra_i64:
; CHECK: .functype sra_i64 (i64, i64) -> (i64)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.shr_s
; CHECK-NEXT: # fallthrough-return
%m = and i64 %x, 63
%a = ashr i64 %v, %m
ret i64 %a
}
define i64 @srl_i64(i64 %v, i64 %x) {
; CHECK-LABEL: srl_i64:
; CHECK: .functype srl_i64 (i64, i64) -> (i64)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.shr_u
; CHECK-NEXT: # fallthrough-return
%m = and i64 %x, 63
%a = lshr i64 %v, %m
ret i64 %a
}
define <16 x i8> @shl_v16i8(<16 x i8> %v, i8 %x) {
; CHECK-LABEL: shl_v16i8:
; CHECK: .functype shl_v16i8 (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.shl
; CHECK-NEXT: # fallthrough-return
%m = and i8 %x, 7
%t = insertelement <16 x i8> undef, i8 %m, i32 0
%s = shufflevector <16 x i8> %t, <16 x i8> undef,
<16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0,
i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%a = shl <16 x i8> %v, %s
ret <16 x i8> %a
}
define <16 x i8> @shl_v16i8_late(<16 x i8> %v, i8 %x) {
; CHECK-LABEL: shl_v16i8_late:
; CHECK: .functype shl_v16i8_late (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.splat
; CHECK-NEXT: v128.const 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7
; CHECK-NEXT: v128.and
; CHECK-NEXT: i8x16.extract_lane_u 0
; CHECK-NEXT: i8x16.shl
; CHECK-NEXT: # fallthrough-return
%t = insertelement <16 x i8> undef, i8 %x, i32 0
%s = shufflevector <16 x i8> %t, <16 x i8> undef,
<16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0,
i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%m = and <16 x i8> %s, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7,
i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
%a = shl <16 x i8> %v, %m
ret <16 x i8> %a
}
define <16 x i8> @ashr_v16i8(<16 x i8> %v, i8 %x) {
; CHECK-LABEL: ashr_v16i8:
; CHECK: .functype ashr_v16i8 (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.shr_s
; CHECK-NEXT: # fallthrough-return
%m = and i8 %x, 7
%t = insertelement <16 x i8> undef, i8 %m, i32 0
%s = shufflevector <16 x i8> %t, <16 x i8> undef,
<16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0,
i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%a = ashr <16 x i8> %v, %s
ret <16 x i8> %a
}
define <16 x i8> @ashr_v16i8_late(<16 x i8> %v, i8 %x) {
; CHECK-LABEL: ashr_v16i8_late:
; CHECK: .functype ashr_v16i8_late (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.splat
aheejinUnsubmitted
Not Done
ReplyInline Actions

Are these _late functions yet to be optimized? If so How about adding TODO do make sure they are currently unoptimized?

aheejin: Are these `_late` functions yet to be optimized? If so How about adding `TODO` do make sure…
tlivelyAuthorUnsubmitted
Done
ReplyInline Actions

Sounds good 👍

tlively: Sounds good 👍
; CHECK-NEXT: v128.const 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7
; CHECK-NEXT: v128.and
; CHECK-NEXT: i8x16.extract_lane_u 0
; CHECK-NEXT: i8x16.shr_s
; CHECK-NEXT: # fallthrough-return
%t = insertelement <16 x i8> undef, i8 %x, i32 0
%s = shufflevector <16 x i8> %t, <16 x i8> undef,
<16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0,
i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%m = and <16 x i8> %s, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7,
i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
%a = ashr <16 x i8> %v, %m
ret <16 x i8> %a
}
define <16 x i8> @lshr_v16i8(<16 x i8> %v, i8 %x) {
; CHECK-LABEL: lshr_v16i8:
; CHECK: .functype lshr_v16i8 (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.shr_u
; CHECK-NEXT: # fallthrough-return
%m = and i8 %x, 7
%t = insertelement <16 x i8> undef, i8 %m, i32 0
%s = shufflevector <16 x i8> %t, <16 x i8> undef,
<16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0,
i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%a = lshr <16 x i8> %v, %s
ret <16 x i8> %a
}
define <16 x i8> @lshr_v16i8_late(<16 x i8> %v, i8 %x) {
; CHECK-LABEL: lshr_v16i8_late:
; CHECK: .functype lshr_v16i8_late (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.splat
; CHECK-NEXT: v128.const 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7
; CHECK-NEXT: v128.and
; CHECK-NEXT: i8x16.extract_lane_u 0
; CHECK-NEXT: i8x16.shr_u
; CHECK-NEXT: # fallthrough-return
%t = insertelement <16 x i8> undef, i8 %x, i32 0
%s = shufflevector <16 x i8> %t, <16 x i8> undef,
<16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0,
i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%m = and <16 x i8> %s, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7,
i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
%a = lshr <16 x i8> %v, %m
ret <16 x i8> %a
}
define <8 x i16> @shl_v8i16(<8 x i16> %v, i16 %x) {
; CHECK-LABEL: shl_v8i16:
; CHECK: .functype shl_v8i16 (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i16x8.shl
; CHECK-NEXT: # fallthrough-return
%m = and i16 %x, 15
%t = insertelement <8 x i16> undef, i16 %m, i32 0
%s = shufflevector <8 x i16> %t, <8 x i16> undef,
<8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%a = shl <8 x i16> %v, %s
ret <8 x i16> %a
}
define <8 x i16> @shl_v8i16_late(<8 x i16> %v, i16 %x) {
; CHECK-LABEL: shl_v8i16_late:
; CHECK: .functype shl_v8i16_late (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i16x8.splat
; CHECK-NEXT: v128.const 15, 15, 15, 15, 15, 15, 15, 15
; CHECK-NEXT: v128.and
; CHECK-NEXT: i16x8.extract_lane_u 0
; CHECK-NEXT: i16x8.shl
; CHECK-NEXT: # fallthrough-return
%t = insertelement <8 x i16> undef, i16 %x, i32 0
%s = shufflevector <8 x i16> %t, <8 x i16> undef,
<8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%m = and <8 x i16> %s,
<i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
%a = shl <8 x i16> %v, %m
ret <8 x i16> %a
}
define <8 x i16> @ashr_v8i16(<8 x i16> %v, i16 %x) {
; CHECK-LABEL: ashr_v8i16:
; CHECK: .functype ashr_v8i16 (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i16x8.shr_s
; CHECK-NEXT: # fallthrough-return
%m = and i16 %x, 15
%t = insertelement <8 x i16> undef, i16 %m, i32 0
%s = shufflevector <8 x i16> %t, <8 x i16> undef,
<8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%a = ashr <8 x i16> %v, %s
ret <8 x i16> %a
}
define <8 x i16> @ashr_v8i16_late(<8 x i16> %v, i16 %x) {
; CHECK-LABEL: ashr_v8i16_late:
; CHECK: .functype ashr_v8i16_late (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i16x8.splat
; CHECK-NEXT: v128.const 15, 15, 15, 15, 15, 15, 15, 15
; CHECK-NEXT: v128.and
; CHECK-NEXT: i16x8.extract_lane_u 0
; CHECK-NEXT: i16x8.shr_s
; CHECK-NEXT: # fallthrough-return
%t = insertelement <8 x i16> undef, i16 %x, i32 0
%s = shufflevector <8 x i16> %t, <8 x i16> undef,
<8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%m = and <8 x i16> %s,
<i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
%a = ashr <8 x i16> %v, %m
ret <8 x i16> %a
}
define <8 x i16> @lshr_v8i16(<8 x i16> %v, i16 %x) {
; CHECK-LABEL: lshr_v8i16:
; CHECK: .functype lshr_v8i16 (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i16x8.shr_u
; CHECK-NEXT: # fallthrough-return
%m = and i16 %x, 15
%t = insertelement <8 x i16> undef, i16 %m, i32 0
%s = shufflevector <8 x i16> %t, <8 x i16> undef,
<8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%a = lshr <8 x i16> %v, %s
ret <8 x i16> %a
}
define <8 x i16> @lshr_v8i16_late(<8 x i16> %v, i16 %x) {
; CHECK-LABEL: lshr_v8i16_late:
; CHECK: .functype lshr_v8i16_late (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i16x8.splat
; CHECK-NEXT: v128.const 15, 15, 15, 15, 15, 15, 15, 15
; CHECK-NEXT: v128.and
; CHECK-NEXT: i16x8.extract_lane_u 0
; CHECK-NEXT: i16x8.shr_u
; CHECK-NEXT: # fallthrough-return
%t = insertelement <8 x i16> undef, i16 %x, i32 0
%s = shufflevector <8 x i16> %t, <8 x i16> undef,
<8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
%m = and <8 x i16> %s,
<i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
%a = lshr <8 x i16> %v, %m
ret <8 x i16> %a
}
define <4 x i32> @shl_v4i32(<4 x i32> %v, i32 %x) {
; CHECK-LABEL: shl_v4i32:
; CHECK: .functype shl_v4i32 (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32x4.shl
; CHECK-NEXT: # fallthrough-return
%m = and i32 %x, 31
%t = insertelement <4 x i32> undef, i32 %m, i32 0
%s = shufflevector <4 x i32> %t, <4 x i32> undef,
<4 x i32> <i32 0, i32 0, i32 0, i32 0>
%a = shl <4 x i32> %v, %s
ret <4 x i32> %a
}
define <4 x i32> @shl_v4i32_late(<4 x i32> %v, i32 %x) {
; CHECK-LABEL: shl_v4i32_late:
; CHECK: .functype shl_v4i32_late (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32x4.splat
; CHECK-NEXT: v128.const 31, 31, 31, 31
; CHECK-NEXT: v128.and
; CHECK-NEXT: i32x4.extract_lane 0
; CHECK-NEXT: i32x4.shl
; CHECK-NEXT: # fallthrough-return
%t = insertelement <4 x i32> undef, i32 %x, i32 0
%s = shufflevector <4 x i32> %t, <4 x i32> undef,
<4 x i32> <i32 0, i32 0, i32 0, i32 0>
%m = and <4 x i32> %s, <i32 31, i32 31, i32 31, i32 31>
%a = shl <4 x i32> %v, %m
ret <4 x i32> %a
}
define <4 x i32> @ashr_v4i32(<4 x i32> %v, i32 %x) {
; CHECK-LABEL: ashr_v4i32:
; CHECK: .functype ashr_v4i32 (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32x4.shr_s
; CHECK-NEXT: # fallthrough-return
%m = and i32 %x, 31
%t = insertelement <4 x i32> undef, i32 %m, i32 0
%s = shufflevector <4 x i32> %t, <4 x i32> undef,
<4 x i32> <i32 0, i32 0, i32 0, i32 0>
%a = ashr <4 x i32> %v, %s
ret <4 x i32> %a
}
define <4 x i32> @ashr_v4i32_late(<4 x i32> %v, i32 %x) {
; CHECK-LABEL: ashr_v4i32_late:
; CHECK: .functype ashr_v4i32_late (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32x4.splat
; CHECK-NEXT: v128.const 31, 31, 31, 31
; CHECK-NEXT: v128.and
; CHECK-NEXT: i32x4.extract_lane 0
; CHECK-NEXT: i32x4.shr_s
; CHECK-NEXT: # fallthrough-return
%t = insertelement <4 x i32> undef, i32 %x, i32 0
%s = shufflevector <4 x i32> %t, <4 x i32> undef,
<4 x i32> <i32 0, i32 0, i32 0, i32 0>
%m = and <4 x i32> %s, <i32 31, i32 31, i32 31, i32 31>
%a = ashr <4 x i32> %v, %m
ret <4 x i32> %a
}
define <4 x i32> @lshr_v4i32(<4 x i32> %v, i32 %x) {
; CHECK-LABEL: lshr_v4i32:
; CHECK: .functype lshr_v4i32 (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32x4.shr_u
; CHECK-NEXT: # fallthrough-return
%m = and i32 %x, 31
%t = insertelement <4 x i32> undef, i32 %m, i32 0
%s = shufflevector <4 x i32> %t, <4 x i32> undef,
<4 x i32> <i32 0, i32 0, i32 0, i32 0>
%a = lshr <4 x i32> %v, %s
ret <4 x i32> %a
}
define <4 x i32> @lshr_v4i32_late(<4 x i32> %v, i32 %x) {
; CHECK-LABEL: lshr_v4i32_late:
; CHECK: .functype lshr_v4i32_late (v128, i32) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32x4.splat
; CHECK-NEXT: v128.const 31, 31, 31, 31
; CHECK-NEXT: v128.and
; CHECK-NEXT: i32x4.extract_lane 0
; CHECK-NEXT: i32x4.shr_u
; CHECK-NEXT: # fallthrough-return
%t = insertelement <4 x i32> undef, i32 %x, i32 0
%s = shufflevector <4 x i32> %t, <4 x i32> undef,
<4 x i32> <i32 0, i32 0, i32 0, i32 0>
%m = and <4 x i32> %s, <i32 31, i32 31, i32 31, i32 31>
%a = lshr <4 x i32> %v, %m
ret <4 x i32> %a
}
define <2 x i64> @shl_v2i64(<2 x i64> %v, i64 %x) {
; CHECK-LABEL: shl_v2i64:
; CHECK: .functype shl_v2i64 (v128, i64) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32.wrap_i64
; CHECK-NEXT: i64x2.shl
; CHECK-NEXT: # fallthrough-return
%m = and i64 %x, 63
%t = insertelement <2 x i64> undef, i64 %m, i32 0
%s = shufflevector <2 x i64> %t, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
%a = shl <2 x i64> %v, %s
ret <2 x i64> %a
}
define <2 x i64> @shl_v2i64_late(<2 x i64> %v, i64 %x) {
; CHECK-LABEL: shl_v2i64_late:
; CHECK: .functype shl_v2i64_late (v128, i64) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64x2.splat
; CHECK-NEXT: v128.const 63, 63
; CHECK-NEXT: v128.and
; CHECK-NEXT: i64x2.extract_lane 0
; CHECK-NEXT: i32.wrap_i64
; CHECK-NEXT: i64x2.shl
; CHECK-NEXT: # fallthrough-return
%t = insertelement <2 x i64> undef, i64 %x, i32 0
%s = shufflevector <2 x i64> %t, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
%m = and <2 x i64> %s, <i64 63, i64 63>
%a = shl <2 x i64> %v, %m
ret <2 x i64> %a
}
define <2 x i64> @ashr_v2i64(<2 x i64> %v, i64 %x) {
; CHECK-LABEL: ashr_v2i64:
; CHECK: .functype ashr_v2i64 (v128, i64) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32.wrap_i64
; CHECK-NEXT: i64x2.shr_s
; CHECK-NEXT: # fallthrough-return
%m = and i64 %x, 63
%t = insertelement <2 x i64> undef, i64 %m, i32 0
%s = shufflevector <2 x i64> %t, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
%a = ashr <2 x i64> %v, %s
ret <2 x i64> %a
}
define <2 x i64> @ashr_v2i64_late(<2 x i64> %v, i64 %x) {
; CHECK-LABEL: ashr_v2i64_late:
; CHECK: .functype ashr_v2i64_late (v128, i64) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64x2.splat
; CHECK-NEXT: v128.const 63, 63
; CHECK-NEXT: v128.and
; CHECK-NEXT: i64x2.extract_lane 0
; CHECK-NEXT: i32.wrap_i64
; CHECK-NEXT: i64x2.shr_s
; CHECK-NEXT: # fallthrough-return
%t = insertelement <2 x i64> undef, i64 %x, i32 0
%s = shufflevector <2 x i64> %t, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
%m = and <2 x i64> %s, <i64 63, i64 63>
%a = ashr <2 x i64> %v, %m
ret <2 x i64> %a
}
define <2 x i64> @lshr_v2i64(<2 x i64> %v, i64 %x) {
; CHECK-LABEL: lshr_v2i64:
; CHECK: .functype lshr_v2i64 (v128, i64) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32.wrap_i64
; CHECK-NEXT: i64x2.shr_u
; CHECK-NEXT: # fallthrough-return
%m = and i64 %x, 63
%t = insertelement <2 x i64> undef, i64 %m, i32 0
%s = shufflevector <2 x i64> %t, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
%a = lshr <2 x i64> %v, %s
ret <2 x i64> %a
}
define <2 x i64> @lshr_v2i64_late(<2 x i64> %v, i64 %x) {
; CHECK-LABEL: lshr_v2i64_late:
; CHECK: .functype lshr_v2i64_late (v128, i64) -> (v128)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64x2.splat
; CHECK-NEXT: v128.const 63, 63
; CHECK-NEXT: v128.and
; CHECK-NEXT: i64x2.extract_lane 0
; CHECK-NEXT: i32.wrap_i64
; CHECK-NEXT: i64x2.shr_u
; CHECK-NEXT: # fallthrough-return
%t = insertelement <2 x i64> undef, i64 %x, i32 0
%s = shufflevector <2 x i64> %t, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
%m = and <2 x i64> %s, <i64 63, i64 63>
%a = lshr <2 x i64> %v, %m
ret <2 x i64> %a
}