Skip to content

initial set of testcase for MXFP4 #4739

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
pdhirajkumarprasad merged 3 commits into from
Mar 3, 2026
Merged

initial set of testcase for MXFP4 #4739

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
333d1a2
initial set of testcase for MXFP4
pdhirajkumarprasad Feb 20, 2026
ead479b
Adding some more test cases for MXF4
pdhirajkumarprasad Feb 27, 2026
ba180c0
updated yaml file based on feedback
pdhirajkumarprasad Mar 3, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
55 changes: 55 additions & 0 deletions ...aslt/tensilelite/Tensile/Tests/common/gemm/gfx950/MXFP4_to_FP32_batch1_K32_depth_128.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
################################################################################
# Test where K is not multiple of DepthU
################################################################################

TestParameters:
marks: [xfail-gfx950, skip-gfx900, skip-gfx906, skip-gfx908, skip-gfx90a, skip-gfx940, skip-gfx941, skip-gfx942, skip-gfx1010, skip-gfx1011, skip-gfx1012, skip-gfx1030, skip-gfx1100, skip-gfx1101, skip-gfx1102, skip-gfx1200, skip-gfx1201]

GlobalParameters:
NumElementsToValidate: -1
MinimumRequiredVersion: 5.0.0
PrintLevel: 1
PrintSolutionRejectionReason: True
Device: 0
CMakeBuildType: Release
KernelTime: True
MaxWorkspaceSize: 13421772800
DataInitTypeA: 3
DataInitTypeB: 3
DataInitTypeC: 0
DataInitTypeMXSA: 3
DataInitTypeMXSB: 3
DataInitTypeAlpha: 1
DataInitTypeBeta: 0
BoundsCheck: 0
KeepBuildTmp: True
MaxFileName: 128
DeviceLDS: 163840
MaxLDS: 163840

BenchmarkProblems:
-
- OperationType: GEMM
DataType: F4
DataTypeB: F4
DestDataType: S
ComputeDataType: S
HighPrecisionAccumulate: True
MXBlockA: 32
MXBlockB: 32
TransposeA: 1
TransposeB: 0
UseBeta: True
Batched: True

- InitialSolutionParameters:
BenchmarkCommonParameters:
- KernelLanguage: ["Assembly"]
ForkParameters:
- MatrixInstruction:
- [16, 16, 128, 1, 1, 1, 1, 1, 1]
- DepthU: [128]
BenchmarkJoinParameters:
BenchmarkFinalParameters:
- ProblemSizes:
- Exact: [128, 128, 1, 32]
71 changes: 71 additions & 0 deletions ...pblaslt/tensilelite/Tensile/Tests/common/gemm/gfx950/MXFP4_to_FP32_batch1_ctype_bf16.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,71 @@
TestParameters:
marks: [xfail-gfx950, skip-gfx900, skip-gfx906, skip-gfx908, skip-gfx90a, skip-gfx940, skip-gfx941, skip-gfx942, skip-gfx1010, skip-gfx1011, skip-gfx1012, skip-gfx1030, skip-gfx1100, skip-gfx1101, skip-gfx1102, skip-gfx1200, skip-gfx1201]

GlobalParameters:
NumElementsToValidate: -1
MinimumRequiredVersion: 5.0.0
PrintLevel: 1
PrintSolutionRejectionReason: True
Device: 0
CMakeBuildType: Debug
KernelTime: True
MaxWorkspaceSize: 13421772800
DataInitTypeA: 3
DataInitTypeB: 3
DataInitTypeC: 0
DataInitTypeMXSA: 3
DataInitTypeMXSB: 3
DataInitTypeAlpha: 1
DataInitTypeBeta: 0
NumElementsToValidate: -1

@talumbau talumbau Mar 2, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

duplicative

@pdhirajkumarprasad pdhirajkumarprasad Mar 2, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I had added this initially as it was part of issue which I reported but these are covered under other set of yaml so will remove these yaml

image

BoundsCheck: 0
KeepBuildTmp: True
MaxFileName: 128
DeviceLDS: 163840
MaxLDS: 163840

BenchmarkProblems:
########################################
# FP8HS with ScaleABCD
########################################
-
- # ProblemType
OperationType: GEMM
DataType: F4
DestDataType: B
ComputeDataType: S
HighPrecisionAccumulate: True
MXBlockA: 32
MXBlockB: 32
TransposeA: 1
TransposeB: 0
UseBeta: True
Batched: True
Activation: True
ActivationType: hipblaslt_all
UseScaleAlphaVec: 1
UseBias: 1
BiasDataTypeList: [s]

- # BenchmarkProblemSizeGroup - Standard
InitialSolutionParameters:
BenchmarkCommonParameters:
- KernelLanguage: ["Assembly"]
ForkParameters:
- MatrixInstruction:
- [16, 16, 128, 1, 1, 1,1, 1,1] # 64x64
- DepthU: [128]
- LocalReadVectorWidthA: [32]
- LocalReadVectorWidthB: [32]
- PrefetchGlobalRead: [2]
- PrefetchLocalRead: [1]
- DirectToLds: [1]
- GlobalReadVectorWidthA: [32]
- GlobalReadVectorWidthB: [32]
BenchmarkJoinParameters:
BenchmarkFinalParameters:
- ProblemSizes:
- Exact: [32, 16, 1, 1024]
- BiasTypeArgs: ['s']
- ActivationArgs:
- [Enum: none]
80 changes: 80 additions & 0 deletions ...cts/hipblaslt/tensilelite/Tensile/Tests/common/gemm/gfx950/MXFP4_to_FP32_batch1_dtl3.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,80 @@
TestParameters:
marks: [xfail-gfx950, skip-gfx900, skip-gfx906, skip-gfx908, skip-gfx90a, skip-gfx940, skip-gfx941, skip-gfx942, skip-gfx1010, skip-gfx1011, skip-gfx1012, skip-gfx1030, skip-gfx1100, skip-gfx1101, skip-gfx1102, skip-gfx1200, skip-gfx1201]

GlobalParameters:
NumElementsToValidate: -1
MinimumRequiredVersion: 5.0.0
PrintLevel: 1
PrintSolutionRejectionReason: True
Device: 0
CMakeBuildType: Debug
KernelTime: True
MaxWorkspaceSize: 13421772800
DataInitTypeA: 3
DataInitTypeB: 3
DataInitTypeC: 0
DataInitTypeMXSA: 3
DataInitTypeMXSB: 3
DataInitTypeAlpha: 1
DataInitTypeBeta: 0
NumElementsToValidate: -1

@talumbau talumbau Mar 2, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

duplicative

BoundsCheck: 0
KeepBuildTmp: True
MaxFileName: 128
DeviceLDS: 163840
MaxLDS: 163840

BenchmarkProblems:
########################################
# FP8HS with ScaleABCD
########################################
-
- # ProblemType
OperationType: GEMM
DataType: F4
DestDataType: S
ComputeDataType: S
HighPrecisionAccumulate: True
MXBlockA: 32
MXBlockB: 32
TransposeA: 1
TransposeB: 0
UseBeta: True
Batched: True
Activation: True
ActivationType: hipblaslt_all
UseScaleAlphaVec: 1
UseBias: 1
BiasDataTypeList: [s]

- # BenchmarkProblemSizeGroup - Standard
InitialSolutionParameters:
BenchmarkCommonParameters:
- KernelLanguage: ["Assembly"]
ForkParameters:
- MatrixInstruction:
- [16, 16, 128, 1, 1, 1,1, 1,1] # 64x64
- DepthU: [128]
- LocalReadVectorWidthA: [32]
- LocalReadVectorWidthB: [32]
- PrefetchGlobalRead: [2]
- PrefetchLocalRead: [1]
- DirectToLds: [3]
- GlobalReadVectorWidthA: [32]
- GlobalReadVectorWidthB: [32]
- TransposeLDS: [1]
#- WaveSeparateGlobalReadA: [0]
#- WaveSeparateGlobalReadB: [0]
#- 1LDSBuffer: [0]
#- GlobalSplitU: [1,2]
#- GlobalReadPerMfma: [1]
#- LocalWritePerMfma: [-1]
#- StoreVectorWidth: [-1]
#- SourceSwap: [1]
BenchmarkJoinParameters:
BenchmarkFinalParameters:
- ProblemSizes:
- Exact: [32, 16, 1, 1024]
- BiasTypeArgs: ['s']
- ActivationArgs:
- [Enum: none]
71 changes: 71 additions & 0 deletions .../hipblaslt/tensilelite/Tensile/Tests/common/gemm/gfx950/MXFP4_to_FP32_batch1_numeric.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,71 @@
TestParameters:
marks: [xfail-gfx950, skip-gfx900, skip-gfx906, skip-gfx908, skip-gfx90a, skip-gfx940, skip-gfx941, skip-gfx942, skip-gfx1010, skip-gfx1011, skip-gfx1012, skip-gfx1030, skip-gfx1100, skip-gfx1101, skip-gfx1102, skip-gfx1200, skip-gfx1201]

GlobalParameters:
NumElementsToValidate: -1
MinimumRequiredVersion: 5.0.0
PrintLevel: 1
PrintSolutionRejectionReason: True
Device: 0
CMakeBuildType: Debug
KernelTime: True
MaxWorkspaceSize: 13421772800
DataInitTypeA: 3
DataInitTypeB: 3
DataInitTypeC: 0
DataInitTypeMXSA: 3
DataInitTypeMXSB: 3
DataInitTypeAlpha: 1
DataInitTypeBeta: 0
NumElementsToValidate: -1

@talumbau talumbau Mar 2, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

duplicative

BoundsCheck: 0
KeepBuildTmp: True
MaxFileName: 128
DeviceLDS: 163840
MaxLDS: 163840

BenchmarkProblems:
########################################
# FP8HS with ScaleABCD
########################################
-
- # ProblemType
OperationType: GEMM
DataType: F4
DataTypeA: F4
DestDataType: S
ComputeDataType: S
HighPrecisionAccumulate: True
MXBlockA: 32
MXBlockB: 32
TransposeA: 1
TransposeB: 0
UseBeta: True
Batched: True
Activation: True
ActivationType: hipblaslt_all
UseScaleAlphaVec: 1
UseBias: 1
BiasDataTypeList: [s]

- # BenchmarkProblemSizeGroup - Standard
InitialSolutionParameters:
BenchmarkCommonParameters:
- KernelLanguage: ["Assembly"]
ForkParameters:
- MatrixInstruction:
- [16, 16, 128, 1, 1, 4,4, 2,2] # 128x128
- DepthU: [128]
- LocalReadVectorWidthA: [32]
- LocalReadVectorWidthB: [32]
- PrefetchGlobalRead: [2]
- PrefetchLocalRead: [1]
- GlobalReadVectorWidthA: [32]
- GlobalReadVectorWidthB: [32]
BenchmarkJoinParameters:
BenchmarkFinalParameters:
- ProblemSizes:
- Exact: [32, 16, 1, 1024]
- BiasTypeArgs: ['s']
- ActivationArgs:
- [Enum: none]
90 changes: 90 additions & 0 deletions ...te/Tensile/Tests/common/gemm/gfx950/MXFP4_to_FP32_batch2_depth256_dtl0_bias_act_gelu.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,90 @@
TestParameters:
marks: [skip-gfx900, skip-gfx906, skip-gfx908, skip-gfx90a, skip-gfx940, skip-gfx941, skip-gfx942, skip-gfx1010, skip-gfx1011, skip-gfx1012, skip-gfx1030, skip-gfx1100, skip-gfx1101, skip-gfx1102, skip-gfx1200, skip-gfx1201, xfail-gfx950]

GlobalParameters:
NumElementsToValidate: -1
MinimumRequiredVersion: 5.0.0
PrintLevel: 1
PrintSolutionRejectionReason: True
Device: 0
CMakeBuildType: Release
KernelTime: True
MaxWorkspaceSize: 13421772800
DataInitTypeA: 3
DataInitTypeB: 3
DataInitTypeC: 0
DataInitTypeMXSA: 3
DataInitTypeMXSB: 3
DataInitTypeAlpha: 1
DataInitTypeBeta: 0
NumElementsToValidate: -1

@talumbau talumbau Mar 2, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

duplicative

BoundsCheck: 0
KeepBuildTmp: True
MaxFileName: 128
DeviceLDS: 163840
MaxLDS: 163840

BenchmarkProblems:
########################################
# MXFP4 to FP32 - TN
########################################
-
- # ProblemType
OperationType: GEMM
DataType: F4
DestDataType: s
ComputeDataType: S
HighPrecisionAccumulate: True
MXBlockA: 32
MXBlockB: 32
TransposeA: 1
TransposeB: 0
UseBeta: True
Batched: True
Activation: True
ActivationType: hipblaslt_all
UseBias: 1
BiasDataTypeList: [s]

- # BenchmarkProblemSizeGroup - Standard
InitialSolutionParameters:
BenchmarkCommonParameters:
- KernelLanguage: ["Assembly"]
ForkParameters:
- MatrixInstruction:
- [16, 16, 128, 1, 1, 2,2, 2,2]
- ForceDisableShadowInit: [True, False]
- UseSgprForGRO: [0,1]
- DepthU: [256]
- DirectToLds: [0]
- AssertFree0ElementMultiple: [1]
- AssertFree1ElementMultiple: [1]
- AssertSummationElementMultiple: [32]
- LocalReadVectorWidth: [32]
- PrefetchGlobalRead: [2]
- PrefetchLocalRead: [1]
- PreloadKernArgs: [0,1]
- ClusterLocalRead: [0,1]
- VectorWidthA: [1]
- VectorWidthB: [1]
- GlobalReadVectorWidthA: [32]
- GlobalReadVectorWidthB: [32]
- ScheduleIterAlg: [1,2,3]
- InnerUnroll: [1]
- ExpandPointerSwap: [0,1]
- TransposeLDS: [1]
- WaveSeparateGlobalReadA: [0]
- WaveSeparateGlobalReadB: [0]
- 1LDSBuffer: [0]
- GlobalSplitU: [1,2]
- GlobalReadPerMfma: [1]
- LocalWritePerMfma: [-1]
- StoreVectorWidth: [-1]
- SourceSwap: [1]
BenchmarkJoinParameters:
BenchmarkFinalParameters:
- ProblemSizes:
- Exact: [64, 64, 1, 128]
- BiasTypeArgs: ['s']
- ActivationArgs:
- [Enum: gelu]
Loading
Loading