; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt < %s -S -mtriple=nvptx-nvidia-cuda -mcpu=sm_100 -passes=nvvm-intr-range | FileCheck %s

; When .reqnctapercluster specifies 3D dimensions, cluster_nctaid.x/y/z and
; cluster_nctarank should be replaced with constants, and cluster_ctaid.x/y/z
; should get per-dimension ranges.
define ptx_kernel i32 @test_cluster_dim_3d() "nvvm.cluster_dim"="4,2,1" {
; CHECK-LABEL: define ptx_kernel i32 @test_cluster_dim_3d(
; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT:    [[CTAID_X:%.*]] = call range(i32 0, 4) i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.x()
; CHECK-NEXT:    [[CTAID_Y:%.*]] = call range(i32 0, 2) i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.y()
; CHECK-NEXT:    [[CTAID_Z:%.*]] = call range(i32 0, 1) i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.z()
; CHECK-NEXT:    [[NCTAID_X:%.*]] = call range(i32 4, 5) i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.x()
; CHECK-NEXT:    [[NCTAID_Y:%.*]] = call range(i32 2, 3) i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.y()
; CHECK-NEXT:    [[NCTAID_Z:%.*]] = call range(i32 1, 2) i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.z()
; CHECK-NEXT:    [[RANK:%.*]] = call range(i32 0, 8) i32 @llvm.nvvm.read.ptx.sreg.cluster.ctarank()
; CHECK-NEXT:    [[NRANK:%.*]] = call range(i32 8, 9) i32 @llvm.nvvm.read.ptx.sreg.cluster.nctarank()
; CHECK-NEXT:    [[A:%.*]] = add i32 [[CTAID_X]], [[CTAID_Y]]
; CHECK-NEXT:    [[B:%.*]] = add i32 [[A]], [[CTAID_Z]]
; CHECK-NEXT:    [[C:%.*]] = add i32 [[B]], [[NCTAID_X]]
; CHECK-NEXT:    [[D:%.*]] = add i32 [[C]], [[NCTAID_Y]]
; CHECK-NEXT:    [[E:%.*]] = add i32 [[D]], [[NCTAID_Z]]
; CHECK-NEXT:    [[F:%.*]] = add i32 [[E]], [[RANK]]
; CHECK-NEXT:    [[G:%.*]] = add i32 [[F]], [[NRANK]]
; CHECK-NEXT:    ret i32 [[G]]
;
  %ctaid.x = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.x()
  %ctaid.y = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.y()
  %ctaid.z = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.z()
  %nctaid.x = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.x()
  %nctaid.y = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.y()
  %nctaid.z = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.z()
  %rank = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctarank()
  %nrank = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctarank()
  %a = add i32 %ctaid.x, %ctaid.y
  %b = add i32 %a, %ctaid.z
  %c = add i32 %b, %nctaid.x
  %d = add i32 %c, %nctaid.y
  %e = add i32 %d, %nctaid.z
  %f = add i32 %e, %rank
  %g = add i32 %f, %nrank
  ret i32 %g
}

; When .reqnctapercluster specifies only 1D, y and z default to 1.
define ptx_kernel i32 @test_cluster_dim_1d() "nvvm.cluster_dim"="4" {
; CHECK-LABEL: define ptx_kernel i32 @test_cluster_dim_1d(
; CHECK-SAME: ) #[[ATTR1:[0-9]+]] {
; CHECK-NEXT:    [[CTAID_X:%.*]] = call range(i32 0, 4) i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.x()
; CHECK-NEXT:    [[CTAID_Y:%.*]] = call range(i32 0, 1) i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.y()
; CHECK-NEXT:    [[CTAID_Z:%.*]] = call range(i32 0, 1) i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.z()
; CHECK-NEXT:    [[NCTAID_X:%.*]] = call range(i32 4, 5) i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.x()
; CHECK-NEXT:    [[NCTAID_Y:%.*]] = call range(i32 1, 2) i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.y()
; CHECK-NEXT:    [[NCTAID_Z:%.*]] = call range(i32 1, 2) i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.z()
; CHECK-NEXT:    [[A:%.*]] = add i32 [[CTAID_X]], [[CTAID_Y]]
; CHECK-NEXT:    [[B:%.*]] = add i32 [[A]], [[CTAID_Z]]
; CHECK-NEXT:    [[C:%.*]] = add i32 [[B]], [[NCTAID_X]]
; CHECK-NEXT:    [[D:%.*]] = add i32 [[C]], [[NCTAID_Y]]
; CHECK-NEXT:    [[E:%.*]] = add i32 [[D]], [[NCTAID_Z]]
; CHECK-NEXT:    ret i32 [[E]]
;
  %ctaid.x = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.x()
  %ctaid.y = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.y()
  %ctaid.z = call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.z()
  %nctaid.x = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.x()
  %nctaid.y = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.y()
  %nctaid.z = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.z()
  %a = add i32 %ctaid.x, %ctaid.y
  %b = add i32 %a, %ctaid.z
  %c = add i32 %b, %nctaid.x
  %d = add i32 %c, %nctaid.y
  %e = add i32 %d, %nctaid.z
  ret i32 %e
}

; maxclusterrank is still an upper bound, not an exact cluster shape.
define ptx_kernel i32 @test_maxclusterrank() "nvvm.maxclusterrank"="8" {
; CHECK-LABEL: define ptx_kernel i32 @test_maxclusterrank(
; CHECK-SAME: ) #[[ATTR2:[0-9]+]] {
; CHECK-NEXT:    [[NCTAID_X:%.*]] = call range(i32 1, 9) i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.x()
; CHECK-NEXT:    [[NRANK:%.*]] = call range(i32 1, 9) i32 @llvm.nvvm.read.ptx.sreg.cluster.nctarank()
; CHECK-NEXT:    [[A:%.*]] = add i32 [[NCTAID_X]], [[NRANK]]
; CHECK-NEXT:    ret i32 [[A]]
;
  %nctaid.x = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.x()
  %nrank = call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctarank()
  %a = add i32 %nctaid.x, %nrank
  ret i32 %a
}

declare i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.x()
declare i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.y()
declare i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.z()
declare i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.x()
declare i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.y()
declare i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.z()
declare i32 @llvm.nvvm.read.ptx.sreg.cluster.ctarank()
declare i32 @llvm.nvvm.read.ptx.sreg.cluster.nctarank()