[midend/lib/Conversion/LowerLinalgToGemmini] Add pass support for gemmini to run E2E LeNet and some tests. #463

shirohasuki · 2025-02-18T21:40:42Z

[midend/lib/Conversion/LowerLinalgToGemmini] Add conv2d_nhwc_fhwc pass for gemmini and add relevant examples and tests.
[midend/lib/Conversion/LowerGemmini] Add scalar print Op for convenient testing and add relevant examples and tests.
[midend/lib/Dialect/Gemmini] Split gemmini conv's RISC and CISC instruction lowering into two separate passes. (Now it's default lowering to RISC type different from native-gemmini, maybe we need a option to config this. Both can pass the LENet test.)

Below are the RISC version and CISC version running LeNet E2E on firesim.

…mini to run E2E LeNet and some tests.

shirohasuki · 2025-02-18T21:47:43Z

The commits seem to be missing the endline? Although I can see them locally.....

linuxlonelyeagle · 2025-02-19T08:08:29Z

examples/GemminiDialect/conv_2d_nhwc_fhwc_5x5_i8.mlir

+  %output = memref.alloc() : memref<1x3x3x1xi8>
+
+  // CHECK: gemmini.tile_conv %{{[0-9]+}} %alloc_{{[0-9]+}} %alloc_{{[0-9]+}} %alloc_{{[0-9]+}} %{{.+}} %{{.+}} :
+  // CHECK-SAME: memref<1x7x7x1xi8> memref<25x1xi8> memref<1xi32> memref<9x1xi8> i64 i64


Maybe you don't need check-same, and feel that the content of check-smae is not important. The same is true for the following example.

gemmini.tile_conv ({{.*}} ({{.*}} ...， Maybe better since you're not capturing the variable and then using it.

linuxlonelyeagle · 2025-02-19T08:15:15Z

midend/lib/Conversion/LowerGemmini/LowerGemminiPass.cpp

+
+    ModuleOp parentModule = op->getParentOfType<ModuleOp>();
+
+    auto printfRef = getOrInsertPrintf(rewriter, parentModule);


Do not use auto,The data types here can be confusing for people who are not familiar with this code.

linuxlonelyeagle · 2025-02-19T08:18:29Z

midend/lib/Conversion/LowerGemmini/LowerGemminiPass.cpp

+      valueToPrint = rewriter.create<LLVM::SExtOp>(loc, rewriter.getI32Type(),
+                                                   valueToPrint);
+    }
+


For unsupported data types, you should return failure.

linuxlonelyeagle

Thanks for the work, I just did a quick review.

linuxlonelyeagle · 2025-02-19T08:20:34Z

examples/GemminiDialect/print.mlir

+  // CHECK: gemmini.print_scalar %{{.*}} : i8
+  gemmini.print_scalar %scalar : i8
+
+  %vector = memref.alloc() : memref<4xi8>        // 1D向量


use English.

linuxlonelyeagle · 2025-02-19T08:21:59Z

midend/lib/Conversion/LowerGemmini/LowerGemminiPass.cpp

@@ -222,6 +311,7 @@ void LowerGemminiToLLVMPass::runOnOperation() {
  cf::populateControlFlowToLLVMConversionPatterns(converter, patterns);
  populateFuncToLLVMConversionPatterns(converter, patterns);
  patterns.add<PrintOpLowering>(&getContext());
+  patterns.add<PrintScalarOpLowering>(&getContext());


... context = &getContext();
patterns.add<...>(context);

linuxlonelyeagle · 2025-02-19T08:25:24Z

midend/lib/Conversion/LowerLinalgToGemmini/LowerLinalgToGemmini.cpp

+    if (stridesAttr)
+      strides = (*stridesAttr.begin()).getLimitedValue();
+
+    if (inputShape[1] != inputShape[2]) // h, w


The comment should be more explicit and should be placed above the if statement.

linuxlonelyeagle · 2025-02-19T08:31:50Z

Ping @shirohasuki I don't seem to have approval authority. If no one approves this PR (CI cannot run, please @zhanghb97 in time.

shirohasuki · 2025-02-19T08:37:26Z

Thank you so much for the REVIEW! I'll modify it and @zhanghb97 for CI.

linuxlonelyeagle · 2025-02-19T08:47:57Z

Print Ops in Gemmini has always been coupled.Its logic should actually be memref -> affine -> scf -> llvm,In fact, Gemmini is just gemmini -> llvm,So maybe you can separate the printOp patterns into a new pass.You can see that the populate pattern functions from the upstream are introduced in the lower gemmini process.

shirohasuki · 2025-02-19T09:19:31Z

I think it's right, Separate the printOp pattern into a decoupled pass and it will be like this?

populateGemminiPrintLegalizeForLLVMExportPatterns()
populateGemminiLegalizeForLLVMExportPatterns()
then, affine -> scf -> llvm................

linuxlonelyeagle · 2025-02-19T09:40:45Z

I think it's right, Separate the printOp pattern into a decoupled pass and it will be like this?
populateGemminiPrintLegalizeForLLVMExportPatterns()
populateGemminiLegalizeForLLVMExportPatterns()
then, affine -> scf -> llvm................

populateGemminiLegalizeForLLVMExportPatterns() should only include patterns that are lower to the gemmini instructions. And populateGemminiPrintLegalizeForLLVMExportPatterns() should only include patterns for print ops, which lower print ops to MLIR upstream ops.and then use the upstream pass, which should be lower to llvm dialect.

gemmini.print => affine/scf
// then use spsteam pass
affine/scf => llvm dialect

shirohasuki · 2025-02-19T10:00:12Z

I think it's right, Separate the printOp pattern into a decoupled pass and it will be like this?
populateGemminiPrintLegalizeForLLVMExportPatterns()
populateGemminiLegalizeForLLVMExportPatterns()
then, affine -> scf -> llvm................
populateGemminiLegalizeForLLVMExportPatterns() should only include patterns that are lower to the gemmini instructions. And populateGemminiPrintLegalizeForLLVMExportPatterns() should only include patterns for print ops, which lower print ops to MLIR upstream ops.and then use the upstream pass, which should be lower to llvm dialect.
gemmini.print => affine/scf
// then use spsteam pass
affine/scf => llvm dialect

I got it!

…ddy-compiler#466)

* [Test] Add testcases for python binding Signed-off-by: Junyi Mei <[email protected]> * [Test] Add testcase for conv-vectorization Signed-off-by: Junyi Mei <[email protected]> * [Test] More detailed check for conv-vectorization Signed-off-by: Junyi Mei <[email protected]> * [Test] Add testcase for matmul-vectorization Signed-off-by: Junyi Mei <[email protected]> * [Test] Verify module ops in python binding Signed-off-by: Junyi Mei <[email protected]> --------- Signed-off-by: Junyi Mei <[email protected]>

…iler#465) Current vectorization pass of matmul-transpose-b reduce the vector in each iteration and accumulate it to the result element. This commit modify it into elementwise addition and do the reduction after the inner loop with reassoc enabled. Signed-off-by: Junyi Mei <[email protected]>

) * [examples] Add MLIR vector dialect to RVV code generation makefile target Signed-off-by: asdf1113 <[email protected]> * [docs] Add RVV instruction support docs Mapping MLIR Vector Operations to RVV Instructions --------- Signed-off-by: asdf1113 <[email protected]>

* [feat] accelerate rotate2d, add rotate4d affine part. * [feat] finish rotate_4d(nhwc and nchw format) * [modify] add test, fix code format. * clean up code format using pre-commit * [fix] force inline in DIP header file to avoid link failure * [fix] format code with pre-commit

zhanghb97 · 2025-03-05T15:58:07Z

Please rebase the code. Let's check if it passes the CI tests after rebase.

…uddy-compiler#471) * [midend] Support scalable vector for matmul-transpos-b vectorization * [midend] Fix matmul-transpose-b scalable vector problem * [Test] Disable scalable in matmul-transpose-b vectorization testcase

…mini to run E2E LeNet and some tests.

…o dev

shirohasuki · 2025-03-12T19:03:01Z

I sync from the remote repo. Maybe we can try CI again?

linuxlonelyeagle · 2025-03-13T02:15:26Z

You can see how much code you've changed in the upper right corner.Looks like you introduced someone else's PR.You should go back to the very first commit and then use git rebase your code to eliminate code conflicts.Or pulling the contents of the main branch should do the trick.

shirohasuki · 2025-03-13T02:26:22Z

OKOK.

[midend/lib/Conversion/LowerLinalgToGemmini] Add pass support for gem…

963580e

…mini to run E2E LeNet and some tests.

linuxlonelyeagle reviewed Feb 19, 2025

View reviewed changes

zhanghb97 and others added 8 commits February 22, 2025 10:39

[examples] Add index cast example with vector type.

23a6260

[examples] Fix the issue with subgraph names in Stable Diffusion. (bu…

a52ec6a

…ddy-compiler#466)

[frontend] Fix tokenizeDeepSeekR1 in TextContainer

277c4b8

[CI] Fix LLVM version bumping issue in CI pipeline. (buddy-compiler#475)

a6542f2

JuniMay and others added 5 commits March 6, 2025 00:01

Update vocab.txt for DeepSeek-R1 (buddy-compiler#468)

dc45fba

[examples] Add compass case.

526b5c5

[midend/lib/Conversion/LowerLinalgToGemmini] Add pass support for gem…

9e990a3

…mini to run E2E LeNet and some tests.

Merge branch 'dev' of https://github.com/shirohasuki/buddy-mlir-1 int…

f5d4bd6

…o dev

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[midend/lib/Conversion/LowerLinalgToGemmini] Add pass support for gemmini to run E2E LeNet and some tests. #463

[midend/lib/Conversion/LowerLinalgToGemmini] Add pass support for gemmini to run E2E LeNet and some tests. #463

shirohasuki commented Feb 18, 2025

shirohasuki commented Feb 18, 2025

linuxlonelyeagle Feb 19, 2025

linuxlonelyeagle Feb 19, 2025 •

edited

Loading

linuxlonelyeagle Feb 19, 2025

linuxlonelyeagle Feb 19, 2025

linuxlonelyeagle left a comment

linuxlonelyeagle Feb 19, 2025

linuxlonelyeagle Feb 19, 2025

linuxlonelyeagle Feb 19, 2025

linuxlonelyeagle commented Feb 19, 2025

shirohasuki commented Feb 19, 2025

linuxlonelyeagle commented Feb 19, 2025

shirohasuki commented Feb 19, 2025

linuxlonelyeagle commented Feb 19, 2025

shirohasuki commented Feb 19, 2025

zhanghb97 commented Mar 5, 2025

shirohasuki commented Mar 12, 2025

linuxlonelyeagle commented Mar 13, 2025

shirohasuki commented Mar 13, 2025


		ModuleOp parentModule = op->getParentOfType<ModuleOp>();

		auto printfRef = getOrInsertPrintf(rewriter, parentModule);

[midend/lib/Conversion/LowerLinalgToGemmini] Add pass support for gemmini to run E2E LeNet and some tests. #463

Are you sure you want to change the base?

[midend/lib/Conversion/LowerLinalgToGemmini] Add pass support for gemmini to run E2E LeNet and some tests. #463

Conversation

shirohasuki commented Feb 18, 2025

shirohasuki commented Feb 18, 2025

linuxlonelyeagle Feb 19, 2025

Choose a reason for hiding this comment

linuxlonelyeagle Feb 19, 2025 • edited Loading

Choose a reason for hiding this comment

linuxlonelyeagle Feb 19, 2025

Choose a reason for hiding this comment

linuxlonelyeagle Feb 19, 2025

Choose a reason for hiding this comment

linuxlonelyeagle left a comment

Choose a reason for hiding this comment

linuxlonelyeagle Feb 19, 2025

Choose a reason for hiding this comment

linuxlonelyeagle Feb 19, 2025

Choose a reason for hiding this comment

linuxlonelyeagle Feb 19, 2025

Choose a reason for hiding this comment

linuxlonelyeagle commented Feb 19, 2025

shirohasuki commented Feb 19, 2025

linuxlonelyeagle commented Feb 19, 2025

shirohasuki commented Feb 19, 2025

linuxlonelyeagle commented Feb 19, 2025

shirohasuki commented Feb 19, 2025

zhanghb97 commented Mar 5, 2025

shirohasuki commented Mar 12, 2025

linuxlonelyeagle commented Mar 13, 2025

shirohasuki commented Mar 13, 2025

linuxlonelyeagle Feb 19, 2025 •

edited

Loading