llvm/test/CodeGen/NVPTX/convert_fp6x2_to_bf16x2.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_100f -mattr=+ptx92 | FileCheck %s
; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_110f -mattr=+ptx92 | FileCheck %s
; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_120f -mattr=+ptx92 | FileCheck %s
; RUN: %if ptxas-sm_100f && ptxas-isa-9.2 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_100f -mattr=+ptx92 | %ptxas-verify -arch=sm_100f %}
; RUN: %if ptxas-sm_110f && ptxas-isa-9.2 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_110f -mattr=+ptx92 | %ptxas-verify -arch=sm_110f %}
; RUN: %if ptxas-sm_120f && ptxas-isa-9.2 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_120f -mattr=+ptx92 | %ptxas-verify -arch=sm_120f %}

define <2 x bfloat> @test_e2m3x2_to_bf16x2_rn_scale_ue8m0(i16 %a, i16 %b) {
; CHECK-LABEL: test_e2m3x2_to_bf16x2_rn_scale_ue8m0(
; CHECK:       {
; CHECK-NEXT:    .reg .b16 %rs<3>;
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0:
; CHECK-NEXT:    ld.param.b16 %rs1, [test_e2m3x2_to_bf16x2_rn_scale_ue8m0_param_0];
; CHECK-NEXT:    ld.param.b16 %rs2, [test_e2m3x2_to_bf16x2_rn_scale_ue8m0_param_1];
; CHECK-NEXT:    cvt.rn.scaled::n2::ue8m0.bf16x2.e2m3x2 %r1, %rs1, %rs2;
; CHECK-NEXT:    st.param.b32 [func_retval0], %r1;
; CHECK-NEXT:    ret;
  %val = call <2 x bfloat> @llvm.nvvm.e2m3x2.to.bf16x2.rn.scale.n2.ue8m0(i16 %a, i16 %b)
  ret <2 x bfloat> %val
}

define <2 x bfloat> @test_e2m3x2_to_bf16x2_rn_relu_scale_ue8m0(i16 %a, i16 %b) {
; CHECK-LABEL: test_e2m3x2_to_bf16x2_rn_relu_scale_ue8m0(
; CHECK:       {
; CHECK-NEXT:    .reg .b16 %rs<3>;
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0:
; CHECK-NEXT:    ld.param.b16 %rs1, [test_e2m3x2_to_bf16x2_rn_relu_scale_ue8m0_param_0];
; CHECK-NEXT:    ld.param.b16 %rs2, [test_e2m3x2_to_bf16x2_rn_relu_scale_ue8m0_param_1];
; CHECK-NEXT:    cvt.rn.relu.scaled::n2::ue8m0.bf16x2.e2m3x2 %r1, %rs1, %rs2;
; CHECK-NEXT:    st.param.b32 [func_retval0], %r1;
; CHECK-NEXT:    ret;
  %val = call <2 x bfloat> @llvm.nvvm.e2m3x2.to.bf16x2.rn.relu.scale.n2.ue8m0(i16 %a, i16 %b)
  ret <2 x bfloat> %val
}

define <2 x bfloat> @test_e2m3x2_to_bf16x2_rn_satfinite_scale_ue8m0(i16 %a, i16 %b) {
; CHECK-LABEL: test_e2m3x2_to_bf16x2_rn_satfinite_scale_ue8m0(
; CHECK:       {
; CHECK-NEXT:    .reg .b16 %rs<3>;
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0:
; CHECK-NEXT:    ld.param.b16 %rs1, [test_e2m3x2_to_bf16x2_rn_satfinite_scale_ue8m0_param_0];
; CHECK-NEXT:    ld.param.b16 %rs2, [test_e2m3x2_to_bf16x2_rn_satfinite_scale_ue8m0_param_1];
; CHECK-NEXT:    cvt.rn.satfinite.scaled::n2::ue8m0.bf16x2.e2m3x2 %r1, %rs1, %rs2;
; CHECK-NEXT:    st.param.b32 [func_retval0], %r1;
; CHECK-NEXT:    ret;
  %val = call <2 x bfloat> @llvm.nvvm.e2m3x2.to.bf16x2.rn.satfinite.scale.n2.ue8m0(i16 %a, i16 %b)
  ret <2 x bfloat> %val
}

define <2 x bfloat> @test_e2m3x2_to_bf16x2_rn_relu_satfinite_scale_ue8m0(i16 %a, i16 %b) {
; CHECK-LABEL: test_e2m3x2_to_bf16x2_rn_relu_satfinite_scale_ue8m0(
; CHECK:       {
; CHECK-NEXT:    .reg .b16 %rs<3>;
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0:
; CHECK-NEXT:    ld.param.b16 %rs1, [test_e2m3x2_to_bf16x2_rn_relu_satfinite_scale_ue8m0_param_0];
; CHECK-NEXT:    ld.param.b16 %rs2, [test_e2m3x2_to_bf16x2_rn_relu_satfinite_scale_ue8m0_param_1];
; CHECK-NEXT:    cvt.rn.relu.satfinite.scaled::n2::ue8m0.bf16x2.e2m3x2 %r1, %rs1, %rs2;
; CHECK-NEXT:    st.param.b32 [func_retval0], %r1;
; CHECK-NEXT:    ret;
  %val = call <2 x bfloat> @llvm.nvvm.e2m3x2.to.bf16x2.rn.relu.satfinite.scale.n2.ue8m0(i16 %a, i16 %b)
  ret <2 x bfloat> %val
}

define <2 x bfloat> @test_e3m2x2_to_bf16x2_rn_scale_ue8m0(i16 %a, i16 %b) {
; CHECK-LABEL: test_e3m2x2_to_bf16x2_rn_scale_ue8m0(
; CHECK:       {
; CHECK-NEXT:    .reg .b16 %rs<3>;
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0:
; CHECK-NEXT:    ld.param.b16 %rs1, [test_e3m2x2_to_bf16x2_rn_scale_ue8m0_param_0];
; CHECK-NEXT:    ld.param.b16 %rs2, [test_e3m2x2_to_bf16x2_rn_scale_ue8m0_param_1];
; CHECK-NEXT:    cvt.rn.scaled::n2::ue8m0.bf16x2.e3m2x2 %r1, %rs1, %rs2;
; CHECK-NEXT:    st.param.b32 [func_retval0], %r1;
; CHECK-NEXT:    ret;
  %val = call <2 x bfloat> @llvm.nvvm.e3m2x2.to.bf16x2.rn.scale.n2.ue8m0(i16 %a, i16 %b)
  ret <2 x bfloat> %val
}

define <2 x bfloat> @test_e3m2x2_to_bf16x2_rn_relu_scale_ue8m0(i16 %a, i16 %b) {
; CHECK-LABEL: test_e3m2x2_to_bf16x2_rn_relu_scale_ue8m0(
; CHECK:       {
; CHECK-NEXT:    .reg .b16 %rs<3>;
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0:
; CHECK-NEXT:    ld.param.b16 %rs1, [test_e3m2x2_to_bf16x2_rn_relu_scale_ue8m0_param_0];
; CHECK-NEXT:    ld.param.b16 %rs2, [test_e3m2x2_to_bf16x2_rn_relu_scale_ue8m0_param_1];
; CHECK-NEXT:    cvt.rn.relu.scaled::n2::ue8m0.bf16x2.e3m2x2 %r1, %rs1, %rs2;
; CHECK-NEXT:    st.param.b32 [func_retval0], %r1;
; CHECK-NEXT:    ret;
  %val = call <2 x bfloat> @llvm.nvvm.e3m2x2.to.bf16x2.rn.relu.scale.n2.ue8m0(i16 %a, i16 %b)
  ret <2 x bfloat> %val
}

define <2 x bfloat> @test_e3m2x2_to_bf16x2_rn_satfinite_scale_ue8m0(i16 %a, i16 %b) {
; CHECK-LABEL: test_e3m2x2_to_bf16x2_rn_satfinite_scale_ue8m0(
; CHECK:       {
; CHECK-NEXT:    .reg .b16 %rs<3>;
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0:
; CHECK-NEXT:    ld.param.b16 %rs1, [test_e3m2x2_to_bf16x2_rn_satfinite_scale_ue8m0_param_0];
; CHECK-NEXT:    ld.param.b16 %rs2, [test_e3m2x2_to_bf16x2_rn_satfinite_scale_ue8m0_param_1];
; CHECK-NEXT:    cvt.rn.satfinite.scaled::n2::ue8m0.bf16x2.e3m2x2 %r1, %rs1, %rs2;
; CHECK-NEXT:    st.param.b32 [func_retval0], %r1;
; CHECK-NEXT:    ret;
  %val = call <2 x bfloat> @llvm.nvvm.e3m2x2.to.bf16x2.rn.satfinite.scale.n2.ue8m0(i16 %a, i16 %b)
  ret <2 x bfloat> %val
}

define <2 x bfloat> @test_e3m2x2_to_bf16x2_rn_relu_satfinite_scale_ue8m0(i16 %a, i16 %b) {
; CHECK-LABEL: test_e3m2x2_to_bf16x2_rn_relu_satfinite_scale_ue8m0(
; CHECK:       {
; CHECK-NEXT:    .reg .b16 %rs<3>;
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0:
; CHECK-NEXT:    ld.param.b16 %rs1, [test_e3m2x2_to_bf16x2_rn_relu_satfinite_scale_ue8m0_param_0];
; CHECK-NEXT:    ld.param.b16 %rs2, [test_e3m2x2_to_bf16x2_rn_relu_satfinite_scale_ue8m0_param_1];
; CHECK-NEXT:    cvt.rn.relu.satfinite.scaled::n2::ue8m0.bf16x2.e3m2x2 %r1, %rs1, %rs2;
; CHECK-NEXT:    st.param.b32 [func_retval0], %r1;
; CHECK-NEXT:    ret;
  %val = call <2 x bfloat> @llvm.nvvm.e3m2x2.to.bf16x2.rn.relu.satfinite.scale.n2.ue8m0(i16 %a, i16 %b)
  ret <2 x bfloat> %val
}