[PATCH] D153479: [NFC] Add tests for future commit in DAGCombiner

[Konstantina Mitropoulou via Phabricator via llvm-commits]

kmitropoulou marked 2 inline comments as done.
kmitropoulou added inline comments.


================
Comment at: llvm/test/CodeGen/AMDGPU/combine_andor_with_cmps.ll:2
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -verify-machineinstrs -stop-after=finalize-isel < %s | FileCheck %s
+
----------------
arsenm wrote:
> Why use mir for this?
CSE changes my optimization. Therefore, I need to do the checking earlier. 

For example, the following test:

define i1 @test1(i32 %arg1, i32 %arg2) #0 {
  %cmp1 = icmp slt i32 %arg1, 1000
  %cmp2 = icmp slt i32 %arg2, 1000
  %or  = or i1 %cmp1, %cmp2
  ret i1 %or
}

will be optimized as follows with my optimization:

  bb.0 (%ir-block.0):
    liveins: $vgpr0, $vgpr1
    %1:vgpr_32 = COPY $vgpr1
    %0:vgpr_32 = COPY $vgpr0
    %2:vgpr_32 = V_MIN_I32_e64 %0, %1, implicit $exec
    %3:sreg_32 = S_MOV_B32 1000
    %4:sreg_32_xm0_xexec = V_CMP_LT_I32_e64 killed %2, killed %3, implicit $exec
    %5:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, killed %4, implicit $exec
    $vgpr0 = COPY %5
    SI_RETURN implicit $vgpr0

This is the output after the instruction selection. After CSE, the predicate of the compare instruction changes:

; %bb.0:
        s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
        s_waitcnt_vscnt null, 0x0
        v_min_i32_e32 v0, v0, v1
        s_delay_alu instid0(VALU_DEP_1)
        v_cmp_gt_i32_e32 vcc_lo, 0x3e8, v0
        v_cndmask_b32_e64 v0, 0, 1, vcc_lo
        s_setpc_b64 s[30:31]



Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D153479/new/

https://reviews.llvm.org/D153479