blob: 74128ab28c5572fa1865ab9cda592b3125fafbd2 [file] [edit]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc -mtriple=amdgpu12.50 < %s | FileCheck %s
; Two vectors are built outside the loop with the same values in lanes 1 and 3,
; and selected inside it. After isel the select is per lane, and only
; PeepholeOptimizer reveals that lanes 1 and 3 select between the same values.
; Those selects fold away, so only lane 0 needs an s_cselect in the loop.
define amdgpu_kernel void @select_vectors_with_identical_lanes(i32 %a, i32 %b, i32 %c0, i32 %n) {
; CHECK-LABEL: select_vectors_with_identical_lanes:
; CHECK: ; %bb.0: ; %entry
; CHECK-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; CHECK-NEXT: s_mov_b64 s[64:65], 0
; CHECK-NEXT: v_nop
; CHECK-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; CHECK-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 nv
; CHECK-NEXT: s_mov_b32 s6, 0
; CHECK-NEXT: s_wait_xcnt 0x0
; CHECK-NEXT: s_mov_b32 s5, 16
; CHECK-NEXT: s_wait_kmcnt 0x0
; CHECK-NEXT: s_mov_b32 s7, s2
; CHECK-NEXT: .LBB0_1: ; %loop
; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1
; CHECK-NEXT: s_cmp_eq_u32 s6, 7
; CHECK-NEXT: s_cselect_b32 s4, s0, s1
; CHECK-NEXT: ;;#ASMSTART
; CHECK-NEXT: ; use s[4:7]
; CHECK-NEXT: ;;#ASMEND
; CHECK-NEXT: s_add_co_i32 s6, s6, 1
; CHECK-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; CHECK-NEXT: s_cmp_lt_u32 s6, s3
; CHECK-NEXT: s_cbranch_scc1 .LBB0_1
; CHECK-NEXT: ; %bb.2: ; %exit
; CHECK-NEXT: s_endpgm
entry:
%t0 = insertelement <4 x i32> <i32 poison, i32 16, i32 0, i32 poison>, i32 %a, i64 0
%b0 = insertelement <4 x i32> %t0, i32 %c0, i64 3
%t1 = insertelement <4 x i32> <i32 poison, i32 16, i32 0, i32 poison>, i32 %b, i64 0
%b1 = insertelement <4 x i32> %t1, i32 %c0, i64 3
br label %loop
loop:
%i = phi i32 [ 0, %entry ], [ %i.next, %loop ]
%w0 = insertelement <4 x i32> %b0, i32 %i, i64 2
%w1 = insertelement <4 x i32> %b1, i32 %i, i64 2
%c = icmp eq i32 %i, 7
%s = select i1 %c, <4 x i32> %w0, <4 x i32> %w1
call void asm sideeffect "; use $0", "s"(<4 x i32> %s)
%i.next = add i32 %i, 1
%cmp = icmp ult i32 %i.next, %n
br i1 %cmp, label %loop, label %exit
exit:
ret void
}