File: bfe_uint.ll

package info (click to toggle)
llvm-toolchain-17 1%3A17.0.6-22
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid, trixie
  • size: 1,799,624 kB
  • sloc: cpp: 6,428,607; ansic: 1,383,196; asm: 793,408; python: 223,504; objc: 75,364; f90: 60,502; lisp: 33,869; pascal: 15,282; sh: 9,684; perl: 7,453; ml: 4,937; awk: 3,523; makefile: 2,889; javascript: 2,149; xml: 888; fortran: 619; cs: 573
file content (26 lines) | stat: -rw-r--r-- 780 bytes parent folder | download | duplicates (4) 
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
 
; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s

; CHECK: {{^}}bfe_def:
; CHECK: BFE_UINT
define amdgpu_kernel void @bfe_def(ptr addrspace(1) %out, i32 %x) {
entry:
  %0 = lshr i32 %x, 5
  %1 = and i32 %0, 15 ; 0xf
  store i32 %1, ptr addrspace(1) %out
  ret void
}

; This program could be implemented using a BFE_UINT instruction, however
; since the lshr constant + number of bits in the mask is >= 32, it can also be
; implmented with a LSHR instruction, which is better, because LSHR has less
; operands and requires less constants.

; CHECK: {{^}}bfe_shift:
; CHECK-NOT: BFE_UINT
define amdgpu_kernel void @bfe_shift(ptr addrspace(1) %out, i32 %x) {
entry:
  %0 = lshr i32 %x, 16
  %1 = and i32 %0, 65535 ; 0xffff
  store i32 %1, ptr addrspace(1) %out
  ret void
}