; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; NOTE: This sample test demonstrates the pretty print feature for NVPTX intrinsics
; RUN: llvm-as < %s | llvm-dis | FileCheck %s

target triple = "nvptx64-nvidia-cuda"

define void @tcgen05_mma_fp16_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d) {
  ; CHECK-LABEL: define void @tcgen05_mma_fp16_cta1(
  ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=discard */ i32 0)
  call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0)

  ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=lastuse */ i32 1)
  call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 1)

  ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=fill */ i32 2)
  call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 2)

  ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=use */ i32 3)
  call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 3)

  ret void
}

define void @tcgen05_mma_f8f6f4_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d) {
  ; CHECK-LABEL: define void @tcgen05_mma_f8f6f4_cta2(
  ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=discard */ i32 0)
  call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0)

  ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=lastuse */ i32 1)
  call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 1)

  ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=fill */ i32 2)
  call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 2)

  ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=use */ i32 3)
  call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 3)

  ret void
}

; This test verifies that printImmArg is safe to call on all constant arguments, but only prints comments for arguments that have pretty printing configured.
define void @test_mixed_constants_edge_case(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor) {
  ; CHECK-LABEL: define void @test_mixed_constants_edge_case(
  ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 42, i32 100, i1 true, /* kind=i8 */ i32 3, /* cta_group= */ i32 1, /* collector=discard */ i32 0)
  call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 42, i32 100, i1 true, i32 3, i32 1, i32 0)

  ret void
}

declare void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6), ptr addrspace(6), i64, i32, i1, i32, i32, i32)