forked from llvm/llvm-project
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Use the new LLVM dot intrinsics to build SPIRV instructions. This involves generating multiply and add operations for integers and the existing OpDot operation for floating point. This includes adding some generic opcodes for signed, unsigned and floats. These require updating an existing test for all such opcodes. New tests for generating SPIRV float and integer dot intrinsics are added as well. Fixes llvm#88056
- Loading branch information
Showing
8 changed files
with
289 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,75 @@ | ||
; RUN: llc -O0 -mtriple=spirv-unknown-unknown %s -o - | FileCheck %s | ||
; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o - -filetype=obj | spirv-val %} | ||
|
||
; Make sure dxil operation function calls for dot are generated for float type vectors. | ||
|
||
; CHECK-DAG: %[[#float_16:]] = OpTypeFloat 16 | ||
; CHECK-DAG: %[[#vec2_float_16:]] = OpTypeVector %[[#float_16]] 2 | ||
; CHECK-DAG: %[[#vec3_float_16:]] = OpTypeVector %[[#float_16]] 3 | ||
; CHECK-DAG: %[[#vec4_float_16:]] = OpTypeVector %[[#float_16]] 4 | ||
; CHECK-DAG: %[[#float_32:]] = OpTypeFloat 32 | ||
; CHECK-DAG: %[[#vec2_float_32:]] = OpTypeVector %[[#float_32]] 2 | ||
; CHECK-DAG: %[[#vec3_float_32:]] = OpTypeVector %[[#float_32]] 3 | ||
; CHECK-DAG: %[[#vec4_float_32:]] = OpTypeVector %[[#float_32]] 4 | ||
|
||
|
||
define noundef half @dot_half2(<2 x half> noundef %a, <2 x half> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec2_float_16]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec2_float_16]] | ||
; CHECK: OpDot %[[#float_16]] %[[#arg0:]] %[[#arg1:]] | ||
%dx.dot = call half @llvm.fdot.v2f16(<2 x half> %a, <2 x half> %b) | ||
ret half %dx.dot | ||
} | ||
|
||
define noundef half @dot_half3(<3 x half> noundef %a, <3 x half> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec3_float_16]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec3_float_16]] | ||
; CHECK: OpDot %[[#float_16]] %[[#arg0:]] %[[#arg1:]] | ||
%dx.dot = call half @llvm.fdot.v3f16(<3 x half> %a, <3 x half> %b) | ||
ret half %dx.dot | ||
} | ||
|
||
define noundef half @dot_half4(<4 x half> noundef %a, <4 x half> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_16]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_16]] | ||
; CHECK: OpDot %[[#float_16]] %[[#arg0:]] %[[#arg1:]] | ||
%dx.dot = call half @llvm.fdot.v4f16(<4 x half> %a, <4 x half> %b) | ||
ret half %dx.dot | ||
} | ||
|
||
define noundef float @dot_float2(<2 x float> noundef %a, <2 x float> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec2_float_32]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec2_float_32]] | ||
; CHECK: OpDot %[[#float_32]] %[[#arg0:]] %[[#arg1:]] | ||
%dx.dot = call float @llvm.fdot.v2f32(<2 x float> %a, <2 x float> %b) | ||
ret float %dx.dot | ||
} | ||
|
||
define noundef float @dot_float3(<3 x float> noundef %a, <3 x float> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec3_float_32]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec3_float_32]] | ||
; CHECK: OpDot %[[#float_32]] %[[#arg0:]] %[[#arg1:]] | ||
%dx.dot = call float @llvm.fdot.v3f32(<3 x float> %a, <3 x float> %b) | ||
ret float %dx.dot | ||
} | ||
|
||
define noundef float @dot_float4(<4 x float> noundef %a, <4 x float> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_float_32]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_float_32]] | ||
; CHECK: OpDot %[[#float_32]] %[[#arg0:]] %[[#arg1:]] | ||
%dx.dot = call float @llvm.fdot.v4f32(<4 x float> %a, <4 x float> %b) | ||
ret float %dx.dot | ||
} | ||
|
||
declare half @llvm.fdot.v2f16(<2 x half> , <2 x half> ) | ||
declare half @llvm.fdot.v3f16(<3 x half> , <3 x half> ) | ||
declare half @llvm.fdot.v4f16(<4 x half> , <4 x half> ) | ||
declare float @llvm.fdot.v2f32(<2 x float>, <2 x float>) | ||
declare float @llvm.fdot.v3f32(<3 x float>, <3 x float>) | ||
declare float @llvm.fdot.v4f32(<4 x float>, <4 x float>) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,88 @@ | ||
; RUN: llc -O0 -mtriple=spirv-unknown-unknown %s -o - | FileCheck %s | ||
; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o - -filetype=obj | spirv-val %} | ||
|
||
; Make sure dxil operation function calls for dot are generated for int/uint vectors. | ||
|
||
; CHECK-DAG: %[[#int_16:]] = OpTypeInt 16 | ||
; CHECK-DAG: %[[#vec2_int_16:]] = OpTypeVector %[[#int_16]] 2 | ||
; CHECK-DAG: %[[#vec3_int_16:]] = OpTypeVector %[[#int_16]] 3 | ||
; CHECK-DAG: %[[#int_32:]] = OpTypeInt 32 | ||
; CHECK-DAG: %[[#vec4_int_32:]] = OpTypeVector %[[#int_32]] 4 | ||
; CHECK-DAG: %[[#int_64:]] = OpTypeInt 64 | ||
; CHECK-DAG: %[[#vec2_int_64:]] = OpTypeVector %[[#int_64]] 2 | ||
|
||
define noundef i16 @dot_int16_t2(<2 x i16> noundef %a, <2 x i16> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec2_int_16]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec2_int_16]] | ||
; CHECK: %[[#mul_vec:]] = OpIMul %[[#vec2_int_16]] %[[#arg0]] %[[#arg1]] | ||
; CHECK: %[[#elt0:]] = OpCompositeExtract %[[#int_16]] %[[#mul_vec]] 0 | ||
; CHECK: %[[#elt1:]] = OpCompositeExtract %[[#int_16]] %[[#mul_vec]] 1 | ||
; CHECK: %[[#sum:]] = OpIAdd %[[#int_16]] %[[#elt0]] %[[#elt1]] | ||
%dot = call i16 @llvm.sdot.v3i16(<2 x i16> %a, <2 x i16> %b) | ||
ret i16 %dot | ||
} | ||
|
||
define noundef i32 @dot_int4(<4 x i32> noundef %a, <4 x i32> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_int_32]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_int_32]] | ||
; CHECK: %[[#mul_vec:]] = OpIMul %[[#vec4_int_32]] %[[#arg0]] %[[#arg1]] | ||
; CHECK: %[[#elt0:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 0 | ||
; CHECK: %[[#elt1:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 1 | ||
; CHECK: %[[#sum0:]] = OpIAdd %[[#int_32]] %[[#elt0]] %[[#elt1]] | ||
; CHECK: %[[#elt2:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 2 | ||
; CHECK: %[[#sum1:]] = OpIAdd %[[#int_32]] %[[#sum0]] %[[#elt2]] | ||
; CHECK: %[[#elt3:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 3 | ||
; CHECK: %[[#sum2:]] = OpIAdd %[[#int_32]] %[[#sum1]] %[[#elt3]] | ||
%dot = call i32 @llvm.sdot.v4i32(<4 x i32> %a, <4 x i32> %b) | ||
ret i32 %dot | ||
} | ||
|
||
define noundef i16 @dot_uint16_t3(<3 x i16> noundef %a, <3 x i16> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec3_int_16]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec3_int_16]] | ||
; CHECK: %[[#mul_vec:]] = OpIMul %[[#vec3_int_16]] %[[#arg0]] %[[#arg1]] | ||
; CHECK: %[[#elt0:]] = OpCompositeExtract %[[#int_16]] %[[#mul_vec]] 0 | ||
; CHECK: %[[#elt1:]] = OpCompositeExtract %[[#int_16]] %[[#mul_vec]] 1 | ||
; CHECK: %[[#sum0:]] = OpIAdd %[[#int_16]] %[[#elt0]] %[[#elt1]] | ||
; CHECK: %[[#elt2:]] = OpCompositeExtract %[[#int_16]] %[[#mul_vec]] 2 | ||
; CHECK: %[[#sum1:]] = OpIAdd %[[#int_16]] %[[#sum0]] %[[#elt2]] | ||
%dot = call i16 @llvm.udot.v3i16(<3 x i16> %a, <3 x i16> %b) | ||
ret i16 %dot | ||
} | ||
|
||
define noundef i32 @dot_uint4(<4 x i32> noundef %a, <4 x i32> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec4_int_32]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec4_int_32]] | ||
; CHECK: %[[#mul_vec:]] = OpIMul %[[#vec4_int_32]] %[[#arg0]] %[[#arg1]] | ||
; CHECK: %[[#elt0:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 0 | ||
; CHECK: %[[#elt1:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 1 | ||
; CHECK: %[[#sum0:]] = OpIAdd %[[#int_32]] %[[#elt0]] %[[#elt1]] | ||
; CHECK: %[[#elt2:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 2 | ||
; CHECK: %[[#sum1:]] = OpIAdd %[[#int_32]] %[[#sum0]] %[[#elt2]] | ||
; CHECK: %[[#elt3:]] = OpCompositeExtract %[[#int_32]] %[[#mul_vec]] 3 | ||
; CHECK: %[[#sum2:]] = OpIAdd %[[#int_32]] %[[#sum1]] %[[#elt3]] | ||
%dot = call i32 @llvm.udot.v4i32(<4 x i32> %a, <4 x i32> %b) | ||
ret i32 %dot | ||
} | ||
|
||
define noundef i64 @dot_uint64_t4(<2 x i64> noundef %a, <2 x i64> noundef %b) { | ||
entry: | ||
; CHECK: %[[#arg0:]] = OpFunctionParameter %[[#vec2_int_64]] | ||
; CHECK: %[[#arg1:]] = OpFunctionParameter %[[#vec2_int_64]] | ||
; CHECK: %[[#mul_vec:]] = OpIMul %[[#vec2_int_64]] %[[#arg0]] %[[#arg1]] | ||
; CHECK: %[[#elt0:]] = OpCompositeExtract %[[#int_64]] %[[#mul_vec]] 0 | ||
; CHECK: %[[#elt1:]] = OpCompositeExtract %[[#int_64]] %[[#mul_vec]] 1 | ||
; CHECK: %[[#sum0:]] = OpIAdd %[[#int_64]] %[[#elt0]] %[[#elt1]] | ||
%dot = call i64 @llvm.udot.v2i64(<2 x i64> %a, <2 x i64> %b) | ||
ret i64 %dot | ||
} | ||
|
||
declare i16 @llvm.sdot.v2i16(<2 x i16>, <2 x i16>) | ||
declare i32 @llvm.sdot.v4i32(<4 x i32>, <4 x i32>) | ||
declare i16 @llvm.udot.v3i32(<3 x i16>, <3 x i16>) | ||
declare i32 @llvm.udot.v4i32(<4 x i32>, <4 x i32>) | ||
declare i64 @llvm.udot.v2i64(<2 x i64>, <2 x i64>) |