File: sgpr-copy-duplicate-operand.ll

package info (click to toggle)
llvm-toolchain-16 1%3A16.0.6-15~deb12u1
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 1,634,792 kB
  • sloc: cpp: 6,179,261; ansic: 1,216,205; asm: 741,319; python: 196,614; objc: 75,325; f90: 49,640; lisp: 32,396; pascal: 12,286; sh: 9,394; perl: 7,442; ml: 5,494; awk: 3,523; makefile: 2,723; javascript: 1,206; xml: 886; fortran: 581; cs: 573
file content (19 lines) | stat: -rw-r--r-- 897 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
 
; RUN:  llc -amdgpu-scalarize-global-loads=false  -march=amdgcn -verify-machineinstrs< %s | FileCheck -check-prefix=SI %s
; RUN:  llc -amdgpu-scalarize-global-loads=false  -march=amdgcn -mcpu=tonga -verify-machineinstrs< %s | FileCheck -check-prefix=SI %s

; Copy VGPR -> SGPR used twice as an instruction operand, which is then
; used in an REG_SEQUENCE that also needs to be handled.

; SI-LABEL: {{^}}test_dup_operands:
; SI: v_add_{{[iu]}}32_e32
define amdgpu_kernel void @test_dup_operands(ptr addrspace(1) noalias %out, ptr addrspace(1) noalias %in) {
  %a = load <2 x i32>, ptr addrspace(1) %in
  %lo = extractelement <2 x i32> %a, i32 0
  %hi = extractelement <2 x i32> %a, i32 1
  %add = add i32 %lo, %lo
  %vec0 = insertelement <2 x i32> undef, i32 %add, i32 0
  %vec1 = insertelement <2 x i32> %vec0, i32 %hi, i32 1
  store <2 x i32> %vec1, ptr addrspace(1) %out, align 8
  ret void
}