148 lines · plain
1; RUN: llc -O0 -mtriple=spirv1.5-unknown-unknown %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-EXP2; RUN: llc -O0 -mtriple=spirv1.6-unknown-unknown %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-DOT3; RUN: llc -O0 -mtriple=spirv1.5-unknown-unknown -spirv-ext=+SPV_KHR_integer_dot_product %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-DOT,CHECK-EXT4; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv1.5-unknown-unknown %s -o - -filetype=obj | spirv-val %}5; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv1.6-unknown-unknown %s -o - -filetype=obj | spirv-val %}6; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv1.5-unknown-unknown -spirv-ext=+SPV_KHR_integer_dot_product %s -o - -filetype=obj | spirv-val %}7 8; Make sure dxil operation function calls for dot are generated for int/uint vectors.9 10; CHECK-DAG: OpCapability Int811; CHECK-DOT-DAG: OpCapability DotProduct12; CHECK-DOT-DAG: OpCapability DotProductInputAll13; CHECK-DOT-DAG: OpCapability DotProductInput4x8Bit14; CHECK-EXT-DAG: OpExtension "SPV_KHR_integer_dot_product"15 16; CHECK-DAG: %[[#int_8:]] = OpTypeInt 817; CHECK-DAG: %[[#vec4_int_8:]] = OpTypeVector %[[#int_8]] 418; CHECK-DAG: %[[#int_16:]] = OpTypeInt 1619; CHECK-DAG: %[[#vec2_int_16:]] = OpTypeVector %[[#int_16]] 220; CHECK-DAG: %[[#vec3_int_16:]] = OpTypeVector %[[#int_16]] 321; CHECK-DAG: %[[#int_32:]] = OpTypeInt 3222; CHECK-DAG: %[[#vec4_int_32:]] = OpTypeVector %[[#int_32]] 423; CHECK-DAG: %[[#int_64:]] = OpTypeInt 6424; CHECK-DAG: %[[#vec2_int_64:]] = OpTypeVector %[[#int_64]] 225 26define noundef i8 @dot_int8_t4(<4 x i8> noundef %a, <4 x i8> noundef %b) {27entry:28; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_int_8]]29; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_int_8]]30 31; CHECK-DOT: %[[#dot:]] = OpSDot %[[#int_8]] %[[#arg0]] %[[#arg1]]32 33; CHECK-EXP: %[[#mul_vec:]] = OpIMul %[[#vec4_int_8]] %[[#arg0]] %[[#arg1]]34; CHECK-EXP: %[[#elt0:]] = OpCompositeExtract %[[#int_8]] %[[#mul_vec]] 035; CHECK-EXP: %[[#elt1:]] = OpCompositeExtract %[[#int_8]] %[[#mul_vec]] 136; CHECK-EXP: %[[#sum:]] = OpIAdd %[[#int_8]] %[[#elt0]] %[[#elt1]]37 %dot = call i8 @llvm.spv.sdot.v4i8(<4 x i8> %a, <4 x i8> %b)38 ret i8 %dot39}40 41define noundef i16 @dot_int16_t2(<2 x i16> noundef %a, <2 x i16> noundef %b) {42entry:43; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec2_int_16]]44; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec2_int_16]]45 46; CHECK-DOT: %[[#dot:]] = OpSDot %[[#int_16]] %[[#arg0]] %[[#arg1]]47 48; CHECK-EXP: %[[#mul_vec:]] = OpIMul %[[#vec2_int_16]] %[[#arg0]] %[[#arg1]]49; CHECK-EXP: %[[#elt0:]] = OpCompositeExtract %[[#int_16]] %[[#mul_vec]] 050; CHECK-EXP: %[[#elt1:]] = OpCompositeExtract %[[#int_16]] %[[#mul_vec]] 151; CHECK-EXP: %[[#sum:]] = OpIAdd %[[#int_16]] %[[#elt0]] %[[#elt1]]52 %dot = call i16 @llvm.spv.sdot.v3i16(<2 x i16> %a, <2 x i16> %b)53 ret i16 %dot54}55 56define noundef i32 @dot_int4(<4 x i32> noundef %a, <4 x i32> noundef %b) {57entry:58; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_int_32]]59; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_int_32]]60 61; CHECK-DOT: %[[#dot:]] = OpSDot %[[#int_32]] %[[#arg0]] %[[#arg1]]62 63; CHECK-EXP: %[[#mul_vec:]] = OpIMul %[[#vec4_int_32]] %[[#arg0]] %[[#arg1]]64; CHECK-EXP: %[[#elt0:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 065; CHECK-EXP: %[[#elt1:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 166; CHECK-EXP: %[[#sum0:]] = OpIAdd %[[#int_32]] %[[#elt0]] %[[#elt1]]67; CHECK-EXP: %[[#elt2:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 268; CHECK-EXP: %[[#sum1:]] = OpIAdd %[[#int_32]] %[[#sum0]] %[[#elt2]]69; CHECK-EXP: %[[#elt3:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 370; CHECK-EXP: %[[#sum2:]] = OpIAdd %[[#int_32]] %[[#sum1]] %[[#elt3]]71 %dot = call i32 @llvm.spv.sdot.v4i32(<4 x i32> %a, <4 x i32> %b)72 ret i32 %dot73}74 75define noundef i8 @dot_uint8_t4(<4 x i8> noundef %a, <4 x i8> noundef %b) {76entry:77; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_int_8]]78; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_int_8]]79 80; CHECK-DOT: %[[#dot:]] = OpUDot %[[#int_8]] %[[#arg0]] %[[#arg1]]81 82; CHECK-EXP: %[[#mul_vec:]] = OpIMul %[[#vec4_int_8]] %[[#arg0]] %[[#arg1]]83; CHECK-EXP: %[[#elt0:]] = OpCompositeExtract %[[#int_8]] %[[#mul_vec]] 084; CHECK-EXP: %[[#elt1:]] = OpCompositeExtract %[[#int_8]] %[[#mul_vec]] 185; CHECK-EXP: %[[#sum:]] = OpIAdd %[[#int_8]] %[[#elt0]] %[[#elt1]]86 %dot = call i8 @llvm.spv.udot.v4i8(<4 x i8> %a, <4 x i8> %b)87 ret i8 %dot88}89 90define noundef i16 @dot_uint16_t3(<3 x i16> noundef %a, <3 x i16> noundef %b) {91entry:92; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec3_int_16]]93; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec3_int_16]]94 95; CHECK-DOT: %[[#dot:]] = OpUDot %[[#int_16]] %[[#arg0]] %[[#arg1]]96 97; CHECK-EXP: %[[#mul_vec:]] = OpIMul %[[#vec3_int_16]] %[[#arg0]] %[[#arg1]]98; CHECK-EXP: %[[#elt0:]] = OpCompositeExtract %[[#int_16]] %[[#mul_vec]] 099; CHECK-EXP: %[[#elt1:]] = OpCompositeExtract %[[#int_16]] %[[#mul_vec]] 1100; CHECK-EXP: %[[#sum0:]] = OpIAdd %[[#int_16]] %[[#elt0]] %[[#elt1]]101; CHECK-EXP: %[[#elt2:]] = OpCompositeExtract %[[#int_16]] %[[#mul_vec]] 2102; CHECK-EXP: %[[#sum1:]] = OpIAdd %[[#int_16]] %[[#sum0]] %[[#elt2]]103 %dot = call i16 @llvm.spv.udot.v3i16(<3 x i16> %a, <3 x i16> %b)104 ret i16 %dot105}106 107define noundef i32 @dot_uint4(<4 x i32> noundef %a, <4 x i32> noundef %b) {108entry:109; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_int_32]]110; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_int_32]]111 112; CHECK-DOT: %[[#dot:]] = OpUDot %[[#int_32]] %[[#arg0]] %[[#arg1]]113 114; CHECK-EXP: %[[#mul_vec:]] = OpIMul %[[#vec4_int_32]] %[[#arg0]] %[[#arg1]]115; CHECK-EXP: %[[#elt0:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 0116; CHECK-EXP: %[[#elt1:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 1117; CHECK-EXP: %[[#sum0:]] = OpIAdd %[[#int_32]] %[[#elt0]] %[[#elt1]]118; CHECK-EXP: %[[#elt2:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 2119; CHECK-EXP: %[[#sum1:]] = OpIAdd %[[#int_32]] %[[#sum0]] %[[#elt2]]120; CHECK-EXP: %[[#elt3:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 3121; CHECK-EXP: %[[#sum2:]] = OpIAdd %[[#int_32]] %[[#sum1]] %[[#elt3]]122 %dot = call i32 @llvm.spv.udot.v4i32(<4 x i32> %a, <4 x i32> %b)123 ret i32 %dot124}125 126define noundef i64 @dot_uint64_t4(<2 x i64> noundef %a, <2 x i64> noundef %b) {127entry:128; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec2_int_64]]129; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec2_int_64]]130 131; CHECK-DOT: %[[#dot:]] = OpUDot %[[#int_64]] %[[#arg0]] %[[#arg1]]132 133; CHECK-EXP: %[[#mul_vec:]] = OpIMul %[[#vec2_int_64]] %[[#arg0]] %[[#arg1]]134; CHECK-EXP: %[[#elt0:]] = OpCompositeExtract %[[#int_64]] %[[#mul_vec]] 0135; CHECK-EXP: %[[#elt1:]] = OpCompositeExtract %[[#int_64]] %[[#mul_vec]] 1136; CHECK-EXP: %[[#sum0:]] = OpIAdd %[[#int_64]] %[[#elt0]] %[[#elt1]]137 %dot = call i64 @llvm.spv.udot.v2i64(<2 x i64> %a, <2 x i64> %b)138 ret i64 %dot139}140 141declare i8 @llvm.spv.sdot.v4i8(<4 x i8>, <4 x i8>)142declare i16 @llvm.spv.sdot.v2i16(<2 x i16>, <2 x i16>)143declare i32 @llvm.spv.sdot.v4i32(<4 x i32>, <4 x i32>)144declare i8 @llvm.spv.udot.v4i8(<4 x i8>, <4 x i8>)145declare i16 @llvm.spv.udot.v3i32(<3 x i16>, <3 x i16>)146declare i32 @llvm.spv.udot.v4i32(<4 x i32>, <4 x i32>)147declare i64 @llvm.spv.udot.v2i64(<2 x i64>, <2 x i64>)148