derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
162 lines
9.4 KiB
MLIR
162 lines
9.4 KiB
MLIR
// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s
|
|
|
|
// -----
|
|
|
|
// CHECK-LABEL: @convert_f32x2_to_f8x2_e4m3
|
|
llvm.func @convert_f32x2_to_f8x2_e4m3(%srcA : f32, %srcB : f32) {
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.ff.to.e4m3x2.rn(float %{{.*}}, float %{{.*}})
|
|
%res1 = nvvm.convert.f32x2.to.f8x2 %srcA, %srcB {rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : i16 (f8E4M3FN)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.ff.to.e4m3x2.rn.relu(float %{{.*}}, float %{{.*}})
|
|
%res2 = nvvm.convert.f32x2.to.f8x2 %srcA, %srcB {relu = true, rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : i16 (f8E4M3FN)
|
|
llvm.return
|
|
}
|
|
|
|
// CHECK-LABEL: @convert_f32x2_to_f8x2_e5m2
|
|
llvm.func @convert_f32x2_to_f8x2_e5m2(%srcA : f32, %srcB : f32) {
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.ff.to.e5m2x2.rn(float %{{.*}}, float %{{.*}})
|
|
%res1 = nvvm.convert.f32x2.to.f8x2 %srcA, %srcB {rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : i16 (f8E5M2)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.ff.to.e5m2x2.rn.relu(float %{{.*}}, float %{{.*}})
|
|
%res2 = nvvm.convert.f32x2.to.f8x2 %srcA, %srcB {relu = true, rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : i16 (f8E5M2)
|
|
llvm.return
|
|
}
|
|
|
|
// CHECK-LABEL: @convert_f32x2_to_f8x2_ue8m0
|
|
llvm.func @convert_f32x2_to_f8x2_ue8m0(%srcA : f32, %srcB : f32) {
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.ff.to.ue8m0x2.rz(float %{{.*}}, float %{{.*}})
|
|
%res1 = nvvm.convert.f32x2.to.f8x2 %srcA, %srcB {rnd = #nvvm.fp_rnd_mode<rz>} : i16 (f8E8M0FNU)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.ff.to.ue8m0x2.rp(float %{{.*}}, float %{{.*}})
|
|
%res2 = nvvm.convert.f32x2.to.f8x2 %srcA, %srcB {rnd = #nvvm.fp_rnd_mode<rp>} : i16 (f8E8M0FNU)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.ff.to.ue8m0x2.rz.satfinite(float %{{.*}}, float %{{.*}})
|
|
%res3 = nvvm.convert.f32x2.to.f8x2 %srcA, %srcB {rnd = #nvvm.fp_rnd_mode<rz>, sat = #nvvm.sat_mode<satfinite>} : i16 (f8E8M0FNU)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.ff.to.ue8m0x2.rp.satfinite(float %{{.*}}, float %{{.*}})
|
|
%res4 = nvvm.convert.f32x2.to.f8x2 %srcA, %srcB {rnd = #nvvm.fp_rnd_mode<rp>, sat = #nvvm.sat_mode<satfinite>} : i16 (f8E8M0FNU)
|
|
llvm.return
|
|
}
|
|
|
|
// CHECK-LABEL: @convert_f32x2_to_f8x2_vector_return
|
|
llvm.func @convert_f32x2_to_f8x2_vector_return(%srcA : f32, %srcB : f32) {
|
|
// CHECK: %[[res1:.*]] = call i16 @llvm.nvvm.ff.to.e4m3x2.rn(float %{{.*}}, float %{{.*}})
|
|
// CHECK-NEXT: %{{.*}} = bitcast i16 %[[res1]] to <2 x i8>
|
|
%res1 = nvvm.convert.f32x2.to.f8x2 %srcA, %srcB {rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : vector<2xi8> (f8E4M3FN)
|
|
// CHECK: %[[res2:.*]] = call i16 @llvm.nvvm.ff.to.e4m3x2.rn.relu(float %{{.*}}, float %{{.*}})
|
|
// CHECK-NEXT: %{{.*}} = bitcast i16 %[[res2]] to <2 x i8>
|
|
%res2 = nvvm.convert.f32x2.to.f8x2 %srcA, %srcB {relu = true, rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : vector<2xi8> (f8E4M3FN)
|
|
llvm.return
|
|
}
|
|
|
|
// -----
|
|
|
|
// CHECK-LABEL: @convert_f16x2_to_f8x2_e4m3
|
|
llvm.func @convert_f16x2_to_f8x2_e4m3(%src : vector<2xf16>) {
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.f16x2.to.e4m3x2.rn(<2 x half> %{{.*}})
|
|
%res1 = nvvm.convert.f16x2.to.f8x2 %src : vector<2xf16> -> i16 (f8E4M3FN)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.f16x2.to.e4m3x2.rn.relu(<2 x half> %{{.*}})
|
|
%res2 = nvvm.convert.f16x2.to.f8x2 %src {relu = true} : vector<2xf16> -> i16 (f8E4M3FN)
|
|
llvm.return
|
|
}
|
|
|
|
// CHECK-LABEL: @convert_f16x2_to_f8x2_e5m2
|
|
llvm.func @convert_f16x2_to_f8x2_e5m2(%src : vector<2xf16>) {
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.f16x2.to.e5m2x2.rn(<2 x half> %{{.*}})
|
|
%res1 = nvvm.convert.f16x2.to.f8x2 %src : vector<2xf16> -> i16 (f8E5M2)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.f16x2.to.e5m2x2.rn.relu(<2 x half> %{{.*}})
|
|
%res2 = nvvm.convert.f16x2.to.f8x2 %src {relu = true} : vector<2xf16> -> i16 (f8E5M2)
|
|
llvm.return
|
|
}
|
|
|
|
// CHECK-LABEL: @convert_f16x2_to_f8x2_vector_return
|
|
llvm.func @convert_f16x2_to_f8x2_vector_return(%src : vector<2xf16>) {
|
|
// CHECK: %[[res1:.*]] = call i16 @llvm.nvvm.f16x2.to.e4m3x2.rn(<2 x half> %{{.*}})
|
|
// CHECK-NEXT: %{{.*}} = bitcast i16 %[[res1]] to <2 x i8>
|
|
%res1 = nvvm.convert.f16x2.to.f8x2 %src : vector<2xf16> -> vector<2xi8> (f8E4M3FN)
|
|
// CHECK: %[[res2:.*]] = call i16 @llvm.nvvm.f16x2.to.e5m2x2.rn(<2 x half> %{{.*}})
|
|
// CHECK-NEXT: %{{.*}} = bitcast i16 %[[res2]] to <2 x i8>
|
|
%res2 = nvvm.convert.f16x2.to.f8x2 %src : vector<2xf16> -> vector<2xi8> (f8E5M2)
|
|
llvm.return
|
|
}
|
|
|
|
// -----
|
|
|
|
// CHECK-LABEL: @convert_bf16x2_to_f8x2_ue8m0
|
|
llvm.func @convert_bf16x2_to_f8x2_ue8m0(%src : vector<2xbf16>) {
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rz(<2 x bfloat> %{{.*}})
|
|
%res1 = nvvm.convert.bf16x2.to.f8x2 %src {rnd = #nvvm.fp_rnd_mode<rz>} : vector<2xbf16> -> i16 (f8E8M0FNU)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rp(<2 x bfloat> %{{.*}})
|
|
%res2 = nvvm.convert.bf16x2.to.f8x2 %src {rnd = #nvvm.fp_rnd_mode<rp>} : vector<2xbf16> -> i16 (f8E8M0FNU)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rz.satfinite(<2 x bfloat> %{{.*}})
|
|
%res3 = nvvm.convert.bf16x2.to.f8x2 %src {rnd = #nvvm.fp_rnd_mode<rz>, sat = #nvvm.sat_mode<satfinite>} : vector<2xbf16> -> i16 (f8E8M0FNU)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rp.satfinite(<2 x bfloat> %{{.*}})
|
|
%res4 = nvvm.convert.bf16x2.to.f8x2 %src {rnd = #nvvm.fp_rnd_mode<rp>, sat = #nvvm.sat_mode<satfinite>} : vector<2xbf16> -> i16 (f8E8M0FNU)
|
|
llvm.return
|
|
}
|
|
|
|
|
|
// CHECK-LABEL: @convert_bf16x2_to_f8x2_e4m3
|
|
llvm.func @convert_bf16x2_to_f8x2_e4m3(%srcA : vector<2xbf16>) {
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.bf16x2.to.e4m3x2.rn.satfinite(<2 x bfloat> %{{.*}})
|
|
%res1 = nvvm.convert.bf16x2.to.f8x2 %srcA {rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : vector<2xbf16> -> i16 (f8E4M3FN)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.bf16x2.to.e4m3x2.rn.relu.satfinite(<2 x bfloat> %{{.*}})
|
|
%res2 = nvvm.convert.bf16x2.to.f8x2 %srcA {relu = true, rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : vector<2xbf16> -> i16 (f8E4M3FN)
|
|
llvm.return
|
|
}
|
|
|
|
// CHECK-LABEL: @convert_bf16x2_to_f8x2_e5m2
|
|
llvm.func @convert_bf16x2_to_f8x2_e5m2(%srcA : vector<2xbf16>) {
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.bf16x2.to.e5m2x2.rn.satfinite(<2 x bfloat> %{{.*}})
|
|
%res1 = nvvm.convert.bf16x2.to.f8x2 %srcA {rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : vector<2xbf16> -> vector<2xi8> (f8E5M2)
|
|
// CHECK: %{{.*}} = call i16 @llvm.nvvm.bf16x2.to.e5m2x2.rn.relu.satfinite(<2 x bfloat> %{{.*}})
|
|
%res2 = nvvm.convert.bf16x2.to.f8x2 %srcA {relu = true, rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : vector<2xbf16> -> vector<2xi8> (f8E5M2)
|
|
llvm.return
|
|
}
|
|
|
|
// CHECK-LABEL: @convert_bf16x2_to_f8x2_vector_return
|
|
llvm.func @convert_bf16x2_to_f8x2_vector_return(%src : vector<2xbf16>) {
|
|
// CHECK: %[[res1:.*]] = call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rz(<2 x bfloat> %{{.*}})
|
|
// CHECK-NEXT: %{{.*}} = bitcast i16 %[[res1]] to <2 x i8>
|
|
%res1 = nvvm.convert.bf16x2.to.f8x2 %src {rnd = #nvvm.fp_rnd_mode<rz>} : vector<2xbf16> -> vector<2xi8> (f8E8M0FNU)
|
|
// CHECK: %[[res2:.*]] = call i16 @llvm.nvvm.bf16x2.to.ue8m0x2.rp.satfinite(<2 x bfloat> %{{.*}})
|
|
// CHECK-NEXT: %{{.*}} = bitcast i16 %[[res2]] to <2 x i8>
|
|
%res2 = nvvm.convert.bf16x2.to.f8x2 %src {rnd = #nvvm.fp_rnd_mode<rp>, sat = #nvvm.sat_mode<satfinite>} : vector<2xbf16> -> vector<2xi8> (f8E8M0FNU)
|
|
// CHECK: %[[res3:.*]] = call i16 @llvm.nvvm.bf16x2.to.e4m3x2.rn.satfinite(<2 x bfloat> %{{.*}})
|
|
// CHECK-NEXT: %{{.*}} = bitcast i16 %[[res3]] to <2 x i8>
|
|
%res3 = nvvm.convert.bf16x2.to.f8x2 %src {rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : vector<2xbf16> -> vector<2xi8> (f8E4M3FN)
|
|
// CHECK: %[[res4:.*]] = call i16 @llvm.nvvm.bf16x2.to.e5m2x2.rn.relu.satfinite(<2 x bfloat> %{{.*}})
|
|
// CHECK-NEXT: %{{.*}} = bitcast i16 %[[res4]] to <2 x i8>
|
|
%res4 = nvvm.convert.bf16x2.to.f8x2 %src {relu = true, rnd = #nvvm.fp_rnd_mode<rn>, sat = #nvvm.sat_mode<satfinite>} : vector<2xbf16> -> vector<2xi8> (f8E5M2)
|
|
llvm.return
|
|
}
|
|
|
|
// -----
|
|
|
|
// CHECK-LABEL: @convert_f8x2_to_f16x2
|
|
llvm.func @convert_f8x2_to_f16x2_e4m3(%src : vector<2xi8>) {
|
|
// CHECK: %[[res1:.*]] = bitcast <2 x i8> %{{.*}} to i16
|
|
// CHECK-NEXT: %{{.*}} = call <2 x half> @llvm.nvvm.e4m3x2.to.f16x2.rn(i16 %[[res1]])
|
|
%res1 = nvvm.convert.f8x2.to.f16x2 %src : vector<2xi8> (f8E4M3FN)-> vector<2xf16>
|
|
// CHECK: %[[res2:.*]] = bitcast <2 x i8> %{{.*}} to i16
|
|
// CHECK-NEXT: %{{.*}} = call <2 x half> @llvm.nvvm.e4m3x2.to.f16x2.rn.relu(i16 %[[res2]])
|
|
%res2 = nvvm.convert.f8x2.to.f16x2 %src {relu = true} : vector<2xi8> (f8E4M3FN)-> vector<2xf16>
|
|
llvm.return
|
|
}
|
|
|
|
// CHECK-LABEL: @convert_f8x2_to_f16x2_e5m2
|
|
llvm.func @convert_f8x2_to_f16x2_e5m2(%src : vector<2xi8>) {
|
|
// CHECK: %[[res1:.*]] = bitcast <2 x i8> %{{.*}} to i16
|
|
// CHECK-NEXT: %{{.*}} = call <2 x half> @llvm.nvvm.e5m2x2.to.f16x2.rn(i16 %[[res1]])
|
|
%res1 = nvvm.convert.f8x2.to.f16x2 %src : vector<2xi8> (f8E5M2)-> vector<2xf16>
|
|
// CHECK: %[[res2:.*]] = bitcast <2 x i8> %{{.*}} to i16
|
|
// CHECK-NEXT: %{{.*}} = call <2 x half> @llvm.nvvm.e5m2x2.to.f16x2.rn.relu(i16 %[[res2]])
|
|
%res2 = nvvm.convert.f8x2.to.f16x2 %src {relu = true} : vector<2xi8> (f8E5M2)-> vector<2xf16>
|
|
llvm.return
|
|
}
|
|
|
|
// -----
|
|
|
|
// CHECK-LABEL: @convert_f8x2_to_bf16x2_ue8m0
|
|
llvm.func @convert_f8x2_to_bf16x2_ue8m0(%src : vector<2xi8>) {
|
|
// CHECK: %[[res1:.*]] = bitcast <2 x i8> %{{.*}} to i16
|
|
// CHECK-NEXT: %{{.*}} = call <2 x bfloat> @llvm.nvvm.ue8m0x2.to.bf16x2(i16 %[[res1]])
|
|
%res1 = nvvm.convert.f8x2.to.bf16x2 %src : vector<2xi8> (f8E8M0FNU)-> vector<2xbf16>
|
|
llvm.return
|
|
}
|