llvm/test/CodeGen/AMDGPU/waitcnt-wcg-attributes.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs -run-pass si-insert-waitcnts -o - %s | FileCheck %s

# Checks that the WaitcntGenerator updates its behavior when we switch MFs
# with different attributes.
# There are 2 pairs of functions, each testing an attribute:
# The first pair is testing amdgpu-expert-scheduling-mode
# and the second is testing target-cpu
#
# In each function pair the function bodies are identical and the only
# difference is the attribute.


--- |
    define amdgpu_kernel void @test-wcg-attributes-noexpert() #0 { ret void }
    define amdgpu_kernel void @test-wcg-attributes-expert() #1 { ret void }
    define amdgpu_kernel void @test-wcg-attributes-gfx900() #2 { ret void }
    define amdgpu_kernel void @test-wcg-attributes-gfx1200() #3 { ret void }
    attributes #0 = { "amdgpu-expert-scheduling-mode"="false" }
    attributes #1 = { "amdgpu-expert-scheduling-mode"="true" }
    attributes #2 = { "target-cpu"="gfx900" }
    attributes #3 = { "target-cpu"="gfx1200" }
...

---
name:            test-wcg-attributes-noexpert
body:             |
  bb.0:
    ; CHECK-LABEL: name: test-wcg-attributes-noexpert
    ; CHECK: S_WAIT_LOADCNT_DSCNT 0
    ; CHECK-NEXT: S_WAIT_EXPCNT 0
    ; CHECK-NEXT: S_WAIT_SAMPLECNT 0
    ; CHECK-NEXT: S_WAIT_BVHCNT 0
    ; CHECK-NEXT: S_WAIT_KMCNT 0
    ; CHECK-NEXT: $vgpr1 = nofpexcept V_EXP_F32_e32 $vgpr0, implicit $mode, implicit $exec
    ; CHECK-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 renamable $vgpr36, 8224, 0, implicit $exec
    ; CHECK-NEXT: S_ENDPGM 0
    $vgpr1 = nofpexcept V_EXP_F32_e32 $vgpr0, implicit $mode, implicit $exec
    $vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 renamable $vgpr36, 8224, 0, implicit $exec
    S_ENDPGM 0
...

---
name:            test-wcg-attributes-expert
body:             |
  bb.0:
    ; CHECK-LABEL: name: test-wcg-attributes-expert
    ; CHECK: S_SETREG_IMM32_B32 2, 2074, implicit-def $mode, implicit $mode
    ; CHECK-NEXT: S_WAIT_LOADCNT_DSCNT 0
    ; CHECK-NEXT: S_WAIT_EXPCNT 0
    ; CHECK-NEXT: S_WAIT_SAMPLECNT 0
    ; CHECK-NEXT: S_WAIT_BVHCNT 0
    ; CHECK-NEXT: S_WAIT_KMCNT 0
    ; CHECK-NEXT: $vgpr1 = nofpexcept V_EXP_F32_e32 $vgpr0, implicit $mode, implicit $exec
    ; CHECK-NEXT: S_WAITCNT_DEPCTR .VaVdst_0
    ; CHECK-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 renamable $vgpr36, 8224, 0, implicit $exec
    ; CHECK-NEXT: S_ENDPGM 0
    $vgpr1 = nofpexcept V_EXP_F32_e32 $vgpr0, implicit $mode, implicit $exec
    $vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 renamable $vgpr36, 8224, 0, implicit $exec
    S_ENDPGM 0
...

---
name:            test-wcg-attributes-gfx900
body:             |
  bb.0:
    ; CHECK-LABEL: name: test-wcg-attributes-gfx900
    ; CHECK: S_WAITCNT 0
    ; CHECK-NEXT: $vgpr1 = nofpexcept V_EXP_F32_e32 $vgpr0, implicit $mode, implicit $exec
    ; CHECK-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 renamable $vgpr36, 8224, 0, implicit $exec
    ; CHECK-NEXT: S_ENDPGM 0
    $vgpr1 = nofpexcept V_EXP_F32_e32 $vgpr0, implicit $mode, implicit $exec
    $vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 renamable $vgpr36, 8224, 0, implicit $exec
    S_ENDPGM 0
...

---
name:            test-wcg-attributes-gfx1200
body:             |
  bb.0:
    ; CHECK-LABEL: name: test-wcg-attributes-gfx1200
    ; CHECK: S_WAIT_LOADCNT_DSCNT 0
    ; CHECK-NEXT: S_WAIT_EXPCNT 0
    ; CHECK-NEXT: S_WAIT_SAMPLECNT 0
    ; CHECK-NEXT: S_WAIT_BVHCNT 0
    ; CHECK-NEXT: S_WAIT_KMCNT 0
    ; CHECK-NEXT: $vgpr1 = nofpexcept V_EXP_F32_e32 $vgpr0, implicit $mode, implicit $exec
    ; CHECK-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 renamable $vgpr36, 8224, 0, implicit $exec
    ; CHECK-NEXT: S_ENDPGM 0
    $vgpr1 = nofpexcept V_EXP_F32_e32 $vgpr0, implicit $mode, implicit $exec
    $vgpr0_vgpr1_vgpr2_vgpr3 = DS_READ_B128_gfx9 renamable $vgpr36, 8224, 0, implicit $exec
    S_ENDPGM 0
...