TheRock/compiler/amd-llvm/llvm/test/CodeGen/AArch64/GlobalISel/legalize-log.mir

# RUN: llc -verify-machineinstrs -mtriple aarch64--- \
# RUN: -run-pass=legalizer -mattr=+fullfp16 -global-isel %s -o - \
# RUN: | FileCheck %s
...
---
name:            test_v4f16.log
alignment:       4
tracksRegLiveness: true
registers:
  - { id: 0, class: _ }
  - { id: 1, class: _ }
body:             |
  bb.0:
    liveins: $d0
    ; CHECK-LABEL: name:            test_v4f16.log
    ; CHECK: [[V1:%[0-9]+]]:_(f16), [[V2:%[0-9]+]]:_(f16), [[V3:%[0-9]+]]:_(f16), [[V4:%[0-9]+]]:_(f16) = G_UNMERGE_VALUES %{{[0-9]+}}(<4 x f16>)

    ; CHECK-DAG: [[V1_S32:%[0-9]+]]:_(f32) = G_FPEXT [[V1]](f16)
    ; CHECK-NEXT: ADJCALLSTACKDOWN
    ; CHECK-NEXT: $s0 = COPY [[V1_S32]](f32)
    ; CHECK-NEXT: BL &logf
    ; CHECK-NEXT: ADJCALLSTACKUP
    ; CHECK-NEXT: [[ELT1_S32:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-NEXT: [[ELT1:%[0-9]+]]:_(f16) = G_FPTRUNC [[ELT1_S32]](f32)

    ; CHECK-DAG: [[V2_S32:%[0-9]+]]:_(f32) = G_FPEXT [[V2]](f16)
    ; CHECK-NEXT: ADJCALLSTACKDOWN
    ; CHECK-NEXT: $s0 = COPY [[V2_S32]](f32)
    ; CHECK-NEXT: BL &logf
    ; CHECK-NEXT: ADJCALLSTACKUP
    ; CHECK-NEXT: [[ELT2_S32:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-NEXT: [[ELT2:%[0-9]+]]:_(f16) = G_FPTRUNC [[ELT2_S32]](f32)

    ; CHECK-DAG: [[V3_S32:%[0-9]+]]:_(f32) = G_FPEXT [[V3]](f16)
    ; CHECK-NEXT: ADJCALLSTACKDOWN
    ; CHECK-NEXT: $s0 = COPY [[V3_S32]](f32)
    ; CHECK-NEXT: BL &logf
    ; CHECK-NEXT: ADJCALLSTACKUP
    ; CHECK-NEXT: [[ELT3_S32:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-NEXT: [[ELT3:%[0-9]+]]:_(f16) = G_FPTRUNC [[ELT3_S32]](f32)

    ; CHECK-DAG: [[V4_S32:%[0-9]+]]:_(f32) = G_FPEXT [[V4]](f16)
    ; CHECK-NEXT: ADJCALLSTACKDOWN
    ; CHECK-NEXT: $s0 = COPY [[V4_S32]](f32)
    ; CHECK-NEXT: BL &logf
    ; CHECK-NEXT: ADJCALLSTACKUP
    ; CHECK-NEXT: [[ELT4_S32:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-NEXT: [[ELT4:%[0-9]+]]:_(f16) = G_FPTRUNC [[ELT4_S32]](f32)

    ; CHECK-DAG: %{{[0-9]+}}:_(<4 x f16>) = G_BUILD_VECTOR [[ELT1]](f16), [[ELT2]](f16), [[ELT3]](f16), [[ELT4]](f16)

    %0:_(<4 x f16>) = COPY $d0
    %1:_(<4 x f16>) = G_FLOG %0
    $d0 = COPY %1(<4 x f16>)
    RET_ReallyLR implicit $d0

...
---
name:            test_v8f16.log
alignment:       4
tracksRegLiveness: true
registers:
  - { id: 0, class: _ }
  - { id: 1, class: _ }
body:             |
  bb.0:
    liveins: $q0

    ; CHECK-LABEL: name:            test_v8f16.log

    ; This is big, so let's just check for the 8 calls to logf, the the
    ; G_UNMERGE_VALUES, and the G_BUILD_VECTOR. The other instructions ought
    ; to be covered by the other tests.

    ; CHECK: G_UNMERGE_VALUES
    ; CHECK: BL &logf
    ; CHECK: BL &logf
    ; CHECK: BL &logf
    ; CHECK: BL &logf
    ; CHECK: BL &logf
    ; CHECK: BL &logf
    ; CHECK: BL &logf
    ; CHECK: BL &logf
    ; CHECK-DAG: G_BUILD_VECTOR

    %0:_(<8 x f16>) = COPY $q0
    %1:_(<8 x f16>) = G_FLOG %0
    $q0 = COPY %1(<8 x f16>)
    RET_ReallyLR implicit $q0

...
---
name:            test_v2f32.log
alignment:       4
tracksRegLiveness: true
registers:
  - { id: 0, class: _ }
  - { id: 1, class: _ }
body:             |
  bb.0:
    liveins: $d0

    ; CHECK-LABEL: name:            test_v2f32.log
    ; CHECK: [[V1:%[0-9]+]]:_(f32), [[V2:%[0-9]+]]:_(f32) = G_UNMERGE_VALUES %{{[0-9]+}}(<2 x f32>)

    ; CHECK-NEXT: ADJCALLSTACKDOWN
    ; CHECK-DAG: $s0 = COPY [[V1]](f32)
    ; CHECK-DAG: BL &logf
    ; CHECK-DAG: [[ELT1:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-DAG: ADJCALLSTACKUP

    ; CHECK-DAG: ADJCALLSTACKDOWN
    ; CHECK-DAG: $s0 = COPY [[V2]](f32)
    ; CHECK-DAG: BL &logf
    ; CHECK-DAG: [[ELT2:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-DAG: ADJCALLSTACKUP

    ; CHECK-DAG: %1:_(<2 x f32>) = G_BUILD_VECTOR [[ELT1]](f32), [[ELT2]](f32)

    %0:_(<2 x f32>) = COPY $d0
    %1:_(<2 x f32>) = G_FLOG %0
    $d0 = COPY %1(<2 x f32>)
    RET_ReallyLR implicit $d0

...
---
name:            test_v4f32.log
alignment:       4
tracksRegLiveness: true
registers:
  - { id: 0, class: _ }
  - { id: 1, class: _ }
body:             |
  bb.0:
    liveins: $q0
    ; CHECK-LABEL: name:            test_v4f32.log
    ; CHECK: [[V1:%[0-9]+]]:_(f32), [[V2:%[0-9]+]]:_(f32), [[V3:%[0-9]+]]:_(f32), [[V4:%[0-9]+]]:_(f32) = G_UNMERGE_VALUES %{{[0-9]+}}(<4 x f32>)

    ; CHECK-NEXT: ADJCALLSTACKDOWN
    ; CHECK-DAG: $s0 = COPY [[V1]](f32)
    ; CHECK-DAG: BL &logf
    ; CHECK-DAG: [[ELT1:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-DAG: ADJCALLSTACKUP

    ; CHECK-DAG: ADJCALLSTACKDOWN
    ; CHECK-DAG: $s0 = COPY [[V2]](f32)
    ; CHECK-DAG: BL &logf
    ; CHECK-DAG: [[ELT2:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-DAG: ADJCALLSTACKUP

    ; CHECK-DAG: ADJCALLSTACKDOWN
    ; CHECK-DAG: $s0 = COPY [[V3]](f32)
    ; CHECK-DAG: BL &logf
    ; CHECK-DAG: [[ELT3:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-DAG: ADJCALLSTACKUP

    ; CHECK-DAG: ADJCALLSTACKDOWN
    ; CHECK-DAG: $s0 = COPY [[V4]](f32)
    ; CHECK-DAG: BL &logf
    ; CHECK-DAG: [[ELT4:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-DAG: ADJCALLSTACKUP

    ; CHECK-DAG: %1:_(<4 x f32>) = G_BUILD_VECTOR [[ELT1]](f32), [[ELT2]](f32), [[ELT3]](f32), [[ELT4]](f32)

    %0:_(<4 x f32>) = COPY $q0
    %1:_(<4 x f32>) = G_FLOG %0
    $q0 = COPY %1(<4 x f32>)
    RET_ReallyLR implicit $q0

...
---
name:            test_v2f64.log
alignment:       4
tracksRegLiveness: true
registers:
  - { id: 0, class: _ }
  - { id: 1, class: _ }
body:             |
  bb.0:
    liveins: $q0

    ; CHECK-LABEL: name:            test_v2f64.log
    ; CHECK: [[V1:%[0-9]+]]:_(f64), [[V2:%[0-9]+]]:_(f64) = G_UNMERGE_VALUES %{{[0-9]+}}(<2 x f64>)

    ; CHECK-NEXT: ADJCALLSTACKDOWN
    ; CHECK-DAG: $d0 = COPY [[V1]](f64)
    ; CHECK-DAG: BL &log
    ; CHECK-DAG: [[ELT1:%[0-9]+]]:_(f64) = COPY $d0
    ; CHECK-DAG: ADJCALLSTACKUP

    ; CHECK-DAG: ADJCALLSTACKDOWN
    ; CHECK-DAG: $d0 = COPY [[V2]](f64)
    ; CHECK-DAG: BL &log
    ; CHECK-DAG: [[ELT2:%[0-9]+]]:_(f64) = COPY $d0
    ; CHECK-DAG: ADJCALLSTACKUP

    ; CHECK-DAG: %1:_(<2 x f64>) = G_BUILD_VECTOR [[ELT1]](f64), [[ELT2]](f64)

    %0:_(<2 x f64>) = COPY $q0
    %1:_(<2 x f64>) = G_FLOG %0
    $q0 = COPY %1(<2 x f64>)
    RET_ReallyLR implicit $q0

...
---
name:            test_log_half
alignment:       4
tracksRegLiveness: true
registers:
  - { id: 0, class: _ }
  - { id: 1, class: _ }
body:             |
  bb.0:
    liveins: $h0
    ; CHECK-LABEL: name:            test_log_half
    ; CHECK: [[REG1:%[0-9]+]]:_(f32) = G_FPEXT %0(f16)
    ; CHECK-NEXT: ADJCALLSTACKDOWN
    ; CHECK-NEXT: $s0 = COPY [[REG1]](f32)
    ; CHECK-NEXT: BL &logf
    ; CHECK-NEXT: ADJCALLSTACKUP
    ; CHECK-NEXT: [[REG2:%[0-9]+]]:_(f32) = COPY $s0
    ; CHECK-NEXT: [[RES:%[0-9]+]]:_(f16) = G_FPTRUNC [[REG2]](f32)

    %0:_(f16) = COPY $h0
    %1:_(f16) = G_FLOG %0
    $h0 = COPY %1(f16)
    RET_ReallyLR implicit $h0