File: sdwa-peephole-instr-combine-sel.mir

package info (click to toggle)
llvm-toolchain-20 1%3A20.1.8-1~exp1
  • links: PTS, VCS
  • area: main
  • in suites: experimental
  • size: 2,111,388 kB
  • sloc: cpp: 7,438,767; ansic: 1,393,871; asm: 1,012,926; python: 241,728; f90: 86,635; objc: 75,411; 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 (56 lines) | stat: -rw-r--r-- 2,402 bytes parent folder | download | duplicates (2)
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
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
# RUN: llc -mtriple=amdgcn -mcpu=gfx1030 -run-pass=si-peephole-sdwa -o - %s | FileCheck -check-prefix=CHECK %s

# Currently the conversions in si-peephole-sdwa are disabled on preexisting sdwa instructions.
# If they are reenabled, the code matches this pattern instead of the corresponding pattern
# for V_LSHLREV_B32_sdwa further below:
# [[V_LSHLREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_sdwa 0, %{{[0-9]+}}, 0, undef [[GLOBAL_LOAD_DWORD_SADDR]], 0, 6, 0, 6, 5, implicit $exec

# TODO Implement a fix for the incorrect sdwa selection

---
name:            sdwa_opsel_hazard
body:             |
  ; CHECK-LABEL: name: sdwa_opsel_hazard
  ; CHECK: bb.0:
  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_64_xexec_xnull = IMPLICIT_DEF
  ; CHECK-NEXT:   [[DEF2:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
  ; CHECK-NEXT:   [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR killed [[DEF1]], [[DEF2]], 0, 0, implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.2
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.1:
  ; CHECK-NEXT:   [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 undef %5, 255, implicit $exec
  ; CHECK-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2, implicit $exec
  ; CHECK-NEXT:   [[V_LSHLREV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_sdwa 0, [[V_MOV_B32_e32_]], 0, undef %5, 0, 6, 0, 6, 0, implicit $exec
  ; CHECK-NEXT:   S_ENDPGM 0
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT: bb.2:
  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
  ; CHECK-NEXT: {{  $}}
  ; CHECK-NEXT:   [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, undef [[GLOBAL_LOAD_DWORD_SADDR]], implicit $exec
  ; CHECK-NEXT:   S_BRANCH %bb.1
  bb.0:
    successors: %bb.2(0x40000000)
    %0:sreg_32 = IMPLICIT_DEF
    %1:sreg_64_xexec_xnull = IMPLICIT_DEF
    %2:vgpr_32 = IMPLICIT_DEF
    %3:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR killed %1, %2, 0, 0, implicit $exec
    S_BRANCH %bb.2

  bb.1:
    %5:vgpr_32 = V_AND_B32_e64 undef %6, 255, implicit $exec
    %7:vgpr_32 = V_LSHLREV_B32_e64 2, killed undef %5, implicit $exec
    S_ENDPGM 0

  bb.2:
    successors: %bb.1(0x40000000)

    %6:vgpr_32 = V_LSHRREV_B32_e64 16, undef %3, implicit $exec

    S_BRANCH %bb.1

...