File: simd-shift-complex-splats.ll

package info (click to toggle)
llvm-toolchain-20 1%3A20.1.6-1~exp1
  • links: PTS, VCS
  • area: main
  • in suites: experimental
  • size: 2,111,304 kB
  • sloc: cpp: 7,438,677; ansic: 1,393,822; asm: 1,012,926; python: 241,650; f90: 86,635; objc: 75,479; lisp: 42,144; pascal: 17,286; sh: 10,027; ml: 5,082; perl: 4,730; awk: 3,523; makefile: 3,349; javascript: 2,251; xml: 892; fortran: 672
file content (60 lines) | stat: -rw-r--r-- 2,511 bytes parent folder | download | duplicates (6)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=+simd128 | FileCheck %s

; Test that SIMD shifts can be lowered correctly even with shift
; values that are more complex than plain splats.

target triple = "wasm32-unknown-unknown"

; CHECK-LABEL: shl_add:
; CHECK-NEXT: .functype shl_add (v128, i32, i32) -> (v128)
; CHECK-NEXT: i32.add $push0=, $1, $2
; CHECK-NEXT: i8x16.shl $push1=, $0, $pop0
; CHECK-NEXT: return $pop1
define <16 x i8> @shl_add(<16 x i8> %v, i8 %a, i8 %b) {
  %t1 = insertelement <16 x i8> undef, i8 %a, i32 0
  %va = shufflevector <16 x i8> %t1, <16 x i8> undef, <16 x i32> zeroinitializer
  %t2 = insertelement <16 x i8> undef, i8 %b, i32 0
  %vb = shufflevector <16 x i8> %t2, <16 x i8> undef, <16 x i32> zeroinitializer
  %shift = add <16 x i8> %va, %vb
  %r = shl <16 x i8> %v, %shift
  ret <16 x i8> %r
}

; CHECK-LABEL: shl_abs:
; CHECK-NEXT: .functype shl_abs (v128, i32) -> (v128)
; CHECK-NEXT: i8x16.splat $push0=, $1
; CHECK-NEXT: i8x16.abs $push1=, $pop0
; CHECK-NEXT: i8x16.extract_lane_u $push2=, $pop1, 0
; CHECK-NEXT: i8x16.shl $push3=, $0, $pop2
; CHECK-NEXT: return $pop3
define <16 x i8> @shl_abs(<16 x i8> %v, i8 %a) {
  %t1 = insertelement <16 x i8> undef, i8 %a, i32 0
  %va = shufflevector <16 x i8> %t1, <16 x i8> undef, <16 x i32> zeroinitializer
  %nva = sub <16 x i8> zeroinitializer, %va
  %c = icmp sgt <16 x i8> %va, zeroinitializer
  %shift = select <16 x i1> %c, <16 x i8> %va, <16 x i8> %nva
  %r = shl <16 x i8> %v, %shift
  ret <16 x i8> %r
}

; CHECK-LABEL: shl_abs_add:
; CHECK-NEXT: .functype shl_abs_add (v128, i32, i32) -> (v128)
; CHECK-NEXT: i32.add $push0=, $1, $2
; CHECK-NEXT: i8x16.splat $push1=, $pop0
; CHECK-NEXT: i8x16.abs $push2=, $pop1
; CHECK-NEXT: i8x16.extract_lane_u $push3=, $pop2, 0
; CHECK-NEXT: i8x16.shl	$push4=, $0, $pop3
; CHECK-NEXT: return $pop4

define <16 x i8> @shl_abs_add(<16 x i8> %v, i8 %a, i8 %b) {
  %t1 = insertelement <16 x i8> undef, i8 %a, i32 0
  %va = shufflevector <16 x i8> %t1, <16 x i8> undef, <16 x i32> zeroinitializer
  %t2 = insertelement <16 x i8> undef, i8 %b, i32 0
  %vb = shufflevector <16 x i8> %t2, <16 x i8> undef, <16 x i32> zeroinitializer
  %vadd = add <16 x i8> %va, %vb
  %nvadd = sub <16 x i8> zeroinitializer, %vadd
  %c = icmp sgt <16 x i8> %vadd, zeroinitializer
  %shift = select <16 x i1> %c, <16 x i8> %vadd, <16 x i8> %nvadd
  %r = shl <16 x i8> %v, %shift
  ret <16 x i8> %r
}