TheRock/compiler/amd-llvm/llvm/test/Transforms/InstCombine/AMDGPU/llvm.amdgcn.readlane.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -mtriple=amdgcn-- -mcpu=gfx1030 -passes=instcombine -S < %s | FileCheck %s

; The readfirstlane version of this test covers all the interesting cases of the
; shared logic. This testcase focuses on readlane specific pitfalls.

; test unary

define float @hoist_fneg_f32(float %arg, i32 %lane) {
; CHECK-LABEL: define float @hoist_fneg_f32(
; CHECK-SAME: float [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[RL:%.*]] = call float @llvm.amdgcn.readlane.f32(float [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[RFL:%.*]] = fneg float [[RL]]
; CHECK-NEXT:    ret float [[RFL]]
;
bb:
  %val = fneg float %arg
  %rl = call float @llvm.amdgcn.readlane.f32(float %val, i32 %lane)
  ret float %rl
}

define double @hoist_fneg_f64(double %arg, i32 %lane) {
; CHECK-LABEL: define double @hoist_fneg_f64(
; CHECK-SAME: double [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[RL:%.*]] = call double @llvm.amdgcn.readlane.f64(double [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[RFL:%.*]] = fneg double [[RL]]
; CHECK-NEXT:    ret double [[RFL]]
;
bb:
  %val = fneg double %arg
  %rl = call double @llvm.amdgcn.readlane.f64(double %val, i32 %lane)
  ret double %rl
}

; test casts

define i32 @hoist_trunc(i64 %arg, i32 %lane) {
; CHECK-LABEL: define i32 @hoist_trunc(
; CHECK-SAME: i64 [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[RL:%.*]] = call i64 @llvm.amdgcn.readlane.i64(i64 [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[TMP0:%.*]] = trunc i64 [[RL]] to i32
; CHECK-NEXT:    ret i32 [[TMP0]]
;
bb:
  %val = trunc i64 %arg to i32
  %rl = call i32 @llvm.amdgcn.readlane.i32(i32 %val, i32 %lane)
  ret i32 %rl
}

define i64 @hoist_zext(i32 %arg, i32 %lane) {
; CHECK-LABEL: define i64 @hoist_zext(
; CHECK-SAME: i32 [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[RL:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[TMP0:%.*]] = zext i32 [[RL]] to i64
; CHECK-NEXT:    ret i64 [[TMP0]]
;
bb:
  %val = zext i32 %arg to i64
  %rl = call i64 @llvm.amdgcn.readlane.i64(i64 %val, i32 %lane)
  ret i64 %rl
}

; test binary i32

define i32 @hoist_add_i32(i32 %arg, i32 %lane) {
; CHECK-LABEL: define i32 @hoist_add_i32(
; CHECK-SAME: i32 [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[RL:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[RFL:%.*]] = add i32 [[RL]], 16777215
; CHECK-NEXT:    ret i32 [[RFL]]
;
bb:
  %val = add i32 %arg, 16777215
  %rl = call i32 @llvm.amdgcn.readlane.i32(i32 %val, i32 %lane)
  ret i32 %rl
}

define float @hoist_fadd_f32(float %arg, i32 %lane) {
; CHECK-LABEL: define float @hoist_fadd_f32(
; CHECK-SAME: float [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[RL:%.*]] = call float @llvm.amdgcn.readlane.f32(float [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[RFL:%.*]] = fadd float [[RL]], 1.280000e+02
; CHECK-NEXT:    ret float [[RFL]]
;
bb:
  %val = fadd float %arg, 128.0
  %rl = call float @llvm.amdgcn.readlane.f32(float %val, i32 %lane)
  ret float %rl
}

; test binary i64

define i64 @hoist_and_i64(i64 %arg, i32 %lane) {
; CHECK-LABEL: define i64 @hoist_and_i64(
; CHECK-SAME: i64 [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[RL:%.*]] = call i64 @llvm.amdgcn.readlane.i64(i64 [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[RFL:%.*]] = and i64 [[RL]], 16777215
; CHECK-NEXT:    ret i64 [[RFL]]
;
bb:
  %val = and i64 %arg, 16777215
  %rl = call i64 @llvm.amdgcn.readlane.i32(i64 %val, i32 %lane)
  ret i64 %rl
}

define double @hoist_fadd_f64(double %arg, i32 %lane) {
; CHECK-LABEL: define double @hoist_fadd_f64(
; CHECK-SAME: double [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[RL:%.*]] = call double @llvm.amdgcn.readlane.f64(double [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[RFL:%.*]] = fadd double [[RL]], 1.280000e+02
; CHECK-NEXT:    ret double [[RFL]]
;
bb:
  %val = fadd double %arg, 128.0
  %rl = call double @llvm.amdgcn.readlane.f64(double %val, i32 %lane)
  ret double %rl
}

; test constant on LHS

define i32 @hoist_sub_i32_lhs(i32 %arg, i32 %lane) {
; CHECK-LABEL: define i32 @hoist_sub_i32_lhs(
; CHECK-SAME: i32 [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[RL:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[RFL:%.*]] = sub i32 16777215, [[RL]]
; CHECK-NEXT:    ret i32 [[RFL]]
;
bb:
  %val = sub i32 16777215, %arg
  %rl = call i32 @llvm.amdgcn.readlane.i32(i32 %val, i32 %lane)
  ret i32 %rl
}

define float @hoist_fsub_f32_lhs(float %arg, i32 %lane) {
; CHECK-LABEL: define float @hoist_fsub_f32_lhs(
; CHECK-SAME: float [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[RL:%.*]] = call float @llvm.amdgcn.readlane.f32(float [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[RFL:%.*]] = fsub float 1.280000e+02, [[RL]]
; CHECK-NEXT:    ret float [[RFL]]
;
bb:
  %val = fsub float 128.0, %arg
  %rl = call float @llvm.amdgcn.readlane.f32(float %val, i32 %lane)
  ret float %rl
}

define i32 @readlane_lane_op_in_other_block(i1 %cond, i32 %arg, i32 %base) {
; CHECK-LABEL: define i32 @readlane_lane_op_in_other_block(
; CHECK-SAME: i1 [[COND:%.*]], i32 [[ARG:%.*]], i32 [[BASE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*]]:
; CHECK-NEXT:    [[LANE:%.*]] = add i32 [[BASE]], 2
; CHECK-NEXT:    br i1 [[COND]], label %[[THEN:.*]], label %[[END:.*]]
; CHECK:       [[THEN]]:
; CHECK-NEXT:    [[RL:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[ARG]], i32 [[LANE]])
; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[RL]], 16777215
; CHECK-NEXT:    br label %[[END]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[RES:%.*]] = phi i32 [ [[TMP0]], %[[THEN]] ], [ [[LANE]], %[[BB]] ]
; CHECK-NEXT:    ret i32 [[RES]]
;
bb:
  %lane = add i32 %base, 2
  br i1 %cond, label %then, label %end

then:
  %val = add i32 %arg, 16777215
  %rl = call i32 @llvm.amdgcn.readlane.i32(i32 %val, i32 %lane)
  br label %end

end:
  %res = phi i32 [%rl, %then], [%lane, %bb]
  ret i32 %res
}

; Check cases where we can't move the readlane higher

define float @cannot_move_readlane(float %arg, i32 %base) {
; CHECK-LABEL: define float @cannot_move_readlane(
; CHECK-SAME: float [[ARG:%.*]], i32 [[BASE:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[VAL:%.*]] = fsub float 1.280000e+02, [[ARG]]
; CHECK-NEXT:    [[LANE:%.*]] = add i32 [[BASE]], 2
; CHECK-NEXT:    [[RFL:%.*]] = call float @llvm.amdgcn.readlane.f32(float [[VAL]], i32 [[LANE]])
; CHECK-NEXT:    ret float [[RFL]]
;
bb:
  %val = fsub float 128.0, %arg
  %lane = add i32 %base, 2
  %rl = call float @llvm.amdgcn.readlane.f32(float %val, i32 %lane)
  ret float %rl
}


; test that convergence tokens are preserved

define i32 @hoist_preserves_convergence_token(i1 %cond, i32 %arg, i32 %lane) convergent {
; CHECK-LABEL: define i32 @hoist_preserves_convergence_token(
; CHECK-SAME: i1 [[COND:%.*]], i32 [[ARG:%.*]], i32 [[LANE:%.*]]) #[[ATTR1:[0-9]+]] {
; CHECK-NEXT:  [[BB:.*]]:
; CHECK-NEXT:    [[ENTRY:%.*]] = call token @llvm.experimental.convergence.entry()
; CHECK-NEXT:    br i1 [[COND]], label %[[THEN:.*]], label %[[END:.*]]
; CHECK:       [[THEN]]:
; CHECK-NEXT:    [[RL:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[ARG]], i32 [[LANE]]) [ "convergencectrl"(token [[ENTRY]]) ]
; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[RL]], 16777215
; CHECK-NEXT:    br label %[[END]]
; CHECK:       [[END]]:
; CHECK-NEXT:    [[RES:%.*]] = phi i32 [ [[TMP0]], %[[THEN]] ], [ [[ARG]], %[[BB]] ]
; CHECK-NEXT:    ret i32 [[RES]]
;
bb:
  %entry = call token @llvm.experimental.convergence.entry()
  br i1 %cond, label %then, label %end

then:
  %val = add i32 %arg, 16777215
  %rl = call i32 @llvm.amdgcn.readlane.i32(i32 %val, i32 %lane) [ "convergencectrl"(token %entry)]
  br label %end

end:
  %res = phi i32 [%rl, %then], [%arg, %bb]
  ret i32 %res
}