Dialect assembly format - make offsets optional for create_nd_tdesc #1

Jianhui-Li · 2025-07-02T03:14:40Z

This PR allows xegpu to take optional offsets when create_nd_tdesc

For memref with static shape, user may or may not set up the offset.

  %3 = xegpu.create_nd_tdesc %src2 : memref<24x32xf32> -> !xegpu.tensor_desc<8x16xf32>
  %3 = xegpu.create_nd_tdesc %src2[0, 0] : memref<24x32xf32> ->  !xegpu.tensor_desc<8x16xf32>

For memref with dynamic shape, user may or may not set up the offset. User must set shape and strides using the new assembly.

  %2 = xegpu.create_nd_tdesc %src shape:[%h, %w] strides:[%w, %c1]  : memref<?x?xf16> -> !xegpu.tensor_desc<8x16xf16>
  %2 = xegpu.create_nd_tdesc %src[%x, %y] shape:[%h, %w] strides:[%w, %c1]  : memref<?x?xf16> -> !xegpu.tensor_desc<8x16xf16>

This PR deprecate the current assembly of setting shape and strides .

  %2 = xegpu.create_nd_tdesc %src[%x, %y], [%h, %w], [%w, %c1]  : memref<?x?xf16> -> !xegpu.tensor_desc<8x16xf16>_**

For initialization with u64, the current PR has a limitation - it can't make the offsets option. So user must give an offset, say [0, 0], even user doesn't intend to use the offset.

 %2 = xegpu.create_nd_tdesc %src[0, 0] shape : [%h, %w] strides : [%w, %c1]  : ui64 -> !xegpu.tensor_desc<8x16xf32>

chencha3 · 2025-07-03T15:14:48Z

mlir/include/mlir/Dialect/XeGPU/IR/XeGPUOps.td

+        return attr;
+
+      // The offsets are allowed to be empty. The Traits verification of OffsetSizeAndStrideOpInterface interface assumes offsets being present. So it is set to be MAX to indicate user not passed any value (kDynamic means offsets passed as variable).
+      setConstOffsets(llvm::SmallVector<int64_t, 4>(getTensorDescShape().size(), std::numeric_limits<int64_t>::max()));


Here using TensorDescShape as anchor is not appropriate, since TensorDesc could have smaller rank than MemRef. Use Shape for MemRef is better.

chencha3 · 2025-07-03T15:16:01Z

mlir/include/mlir/Dialect/XeGPU/IR/XeGPUOps.td

@@ -163,9 +176,19 @@ def XeGPU_CreateNdDescOp: XeGPU_Op<"create_nd_tdesc", [Pure, ViewLikeOpInterface
    }

    ArrayRef<int64_t> getStaticOffsets(){
-      return getConstOffsets();
+      auto attr = getConstOffsetsAttr();
+      if (llvm::isa<IntegerType>(getSourceType()) || attr)


Why does the Source Type matter here?

For createNdDesc, we have the same check in getStaticStrides and getStaticSizes .
The op description says the following for the optional strides and shape:

For the case of dynamic memrefs or pointer, the shape and layout information of the
memory region should be explicitly passed via shape and strides parameters

Both getStaticStrides and getStaticSizes rely on a source memref to extract strides and shape, and on a user to provide them alongside ui64.

For optional offsets, you propose to use memref shape in one of the comments below, in that case, the check starts to matter. Without a memref, the above quoted op description would also extend to, now optional, offsets.

However, since we still require shape for ui64 and we only want to match the rank, we could indirectly depend on the user input via shape :

Suggested change

if (llvm::isa<IntegerType>(getSourceType()) || attr)

if (attr)

return attr;

setConstOffsets(llvm::SmallVector<int64_t, 4>(getStaticSizes().size(), std::numeric_limits<int64_t>::max()));

This should work for both memref and ui64 sources

Did you try it out on your side? I still run into the same error for ui64 case (if we remove offset[0,0]).

Ahh. Now with some more trying, it works. Thanks!

chencha3 · 2025-07-03T15:16:10Z

mlir/include/mlir/Dialect/XeGPU/IR/XeGPUOps.td

+
+      // The offsets are allowed to be empty. The Traits verification of OffsetSizeAndStrideOpInterface interface assumes offsets being present. So it is set to be MAX to indicate user not passed any value (kDynamic means offsets passed as variable).
+      setConstOffsets(llvm::SmallVector<int64_t, 4>(getTensorDescShape().size(), std::numeric_limits<int64_t>::max()));
+      //setConstOffsets(llvm::SmallVector<int64_t, 4>(getTensorDescShape().size(), mlir::ShapedType::kDynamic));


Fix unnecessary conversion of C-String to StringRef in the `Cmp` lambda inside `lookupLLVMIntrinsicByName`. This both fixes an ASAN error in the code that happens when the `Name` StringRef passed in is not a Null terminated StringRef, and additionally can potentially speed up the code as well by eliminating the unnecessary computation of string length every time a C String is converted to StringRef in this code (It seems practically this computation is eliminated in optimized builds, but this will avoid it in O0 builds as well). Added a unit test that demonstrates this issue by building LLVM with these options: ``` CMAKE_BUILD_TYPE=Debug LLVM_USE_SANITIZER=Address LLVM_OPTIMIZE_SANITIZED_BUILDS=OFF ``` The error reported is as follows: ``` ==462665==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x5030000391a2 at pc 0x56525cc30bbf bp 0x7fff9e4ccc60 sp 0x7fff9e4cc428 READ of size 19 at 0x5030000391a2 thread T0 #0 0x56525cc30bbe in strlen (upstream-llvm-second/llvm-project/build/unittests/IR/IRTests+0x713bbe) (BuildId: 0651acf1e582a4d2) #1 0x7f8ff22ad334 in std::char_traits<char>::length(char const*) /usr/bin/../lib/gcc/x86_64-linux-gnu/13/../../../../include/c++/13/bits/char_traits.h:399:9 llvm#2 0x7f8ff22a34a0 in llvm::StringRef::StringRef(char const*) /home/rjoshi/upstream-llvm-second/llvm-project/llvm/include/llvm/ADT/StringRef.h:96:33 llvm#3 0x7f8ff28ca184 in _ZZL25lookupLLVMIntrinsicByNameN4llvm8ArrayRefIjEENS_9StringRefES2_ENK3$_0clIjPKcEEDaT_T0_ upstream-llvm-second/llvm-project/llvm/lib/IR/Intrinsics.cpp:673:18 ```

…lvm#148205) In the original motivating test case, [FoldList](https://github.com/llvm/llvm-project/blob/d8a2141ff98ee35cd1886f536ccc3548b012820b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp#L1764) had entries: ``` #0: UseMI: %224:sreg_32 = S_OR_B32 %219.sub0:sreg_64, %219.sub1:sreg_64, implicit-def dead $scc UseOpNo: 1 #1: UseMI: %224:sreg_32 = S_OR_B32 %219.sub0:sreg_64, %219.sub1:sreg_64, implicit-def dead $scc UseOpNo: 2 ``` After calling [updateOperand(#0)](https://github.com/llvm/llvm-project/blob/d8a2141ff98ee35cd1886f536ccc3548b012820b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp#L1773), [tryConstantFoldOp(#0.UseMI)](https://github.com/llvm/llvm-project/blob/d8a2141ff98ee35cd1886f536ccc3548b012820b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp#L1786) removed operand 1, and entry #&llvm#8203;1.UseOpNo was no longer valid, resulting in an [assert](https://github.com/llvm/llvm-project/blob/4a35214bddbb67f9597a500d48ab8c4fb25af150/llvm/include/llvm/ADT/ArrayRef.h#L452). This change defers constant folding until all operands have been updated so that UseOpNo values remain stable.

Jianhui-Li · 2025-07-24T23:29:00Z

upstreamed llvm#148335

Extend support in LLDB for WebAssembly. This PR adds a new Process plugin (ProcessWasm) that extends ProcessGDBRemote for WebAssembly targets. It adds support for WebAssembly's memory model with separate address spaces, and the ability to fetch the call stack from the WebAssembly runtime. I have tested this change with the WebAssembly Micro Runtime (WAMR, https://github.com/bytecodealliance/wasm-micro-runtime) which implements a GDB debug stub and supports the qWasmCallStack packet. ``` (lldb) process connect --plugin wasm connect://localhost:4567 Process 1 stopped * thread #1, name = 'nobody', stop reason = trace frame #0: 0x40000000000001ad wasm32_args.wasm`main: -> 0x40000000000001ad <+3>: global.get 0 0x40000000000001b3 <+9>: i32.const 16 0x40000000000001b5 <+11>: i32.sub 0x40000000000001b6 <+12>: local.set 0 (lldb) b add Breakpoint 1: where = wasm32_args.wasm`add + 28 at test.c:4:12, address = 0x400000000000019c (lldb) c Process 1 resuming Process 1 stopped * thread #1, name = 'nobody', stop reason = breakpoint 1.1 frame #0: 0x400000000000019c wasm32_args.wasm`add(a=<unavailable>, b=<unavailable>) at test.c:4:12 1 int 2 add(int a, int b) 3 { -> 4 return a + b; 5 } 6 7 int (lldb) bt * thread #1, name = 'nobody', stop reason = breakpoint 1.1 * frame #0: 0x400000000000019c wasm32_args.wasm`add(a=<unavailable>, b=<unavailable>) at test.c:4:12 frame #1: 0x40000000000001e5 wasm32_args.wasm`main at test.c:12:12 frame llvm#2: 0x40000000000001fe wasm32_args.wasm ``` This PR is based on an unmerged patch from Paolo Severini: https://reviews.llvm.org/D78801. I intentionally stuck to the foundations to keep this PR small. I have more PRs in the pipeline to support the other features/packets. My motivation for supporting Wasm is to support debugging Swift compiled to WebAssembly: https://www.swift.org/documentation/articles/wasm-getting-started.html

…erver (llvm#148774) Summary: There was a deadlock was introduced by [PR llvm#146441](llvm#146441) which changed `CurrentThreadIsPrivateStateThread()` to `CurrentThreadPosesAsPrivateStateThread()`. This change caused the execution path in [`ExecutionContextRef::SetTargetPtr()`](https://github.com/llvm/llvm-project/blob/10b5558b61baab59c7d3dff37ffdf0861c0cc67a/lldb/source/Target/ExecutionContext.cpp#L513) to now enter a code block that was previously skipped, triggering [`GetSelectedFrame()`](https://github.com/llvm/llvm-project/blob/10b5558b61baab59c7d3dff37ffdf0861c0cc67a/lldb/source/Target/ExecutionContext.cpp#L522) which leads to a deadlock. Thread 1 gets m_modules_mutex in [`ModuleList::AppendImpl`](https://github.com/llvm/llvm-project/blob/96148f92146e5211685246722664e51ec730e7ba/lldb/source/Core/ModuleList.cpp#L218), Thread 3 gets m_language_runtimes_mutex in [`GetLanguageRuntime`](https://github.com/llvm/llvm-project/blob/96148f92146e5211685246722664e51ec730e7ba/lldb/source/Target/Process.cpp#L1501), but then Thread 1 waits for m_language_runtimes_mutex in [`GetLanguageRuntime`](https://github.com/llvm/llvm-project/blob/96148f92146e5211685246722664e51ec730e7ba/lldb/source/Target/Process.cpp#L1501) while Thread 3 waits for m_modules_mutex in [`ScanForGNUstepObjCLibraryCandidate`](https://github.com/llvm/llvm-project/blob/96148f92146e5211685246722664e51ec730e7ba/lldb/source/Plugins/LanguageRuntime/ObjC/GNUstepObjCRuntime/GNUstepObjCRuntime.cpp#L57). This fixes the deadlock by adding a scoped block around the mutex lock before the call to the notifier, and moved the notifier call outside of the mutex-guarded section. The notifier call [`NotifyModuleAdded`](https://github.com/llvm/llvm-project/blob/96148f92146e5211685246722664e51ec730e7ba/lldb/source/Target/Target.cpp#L1810) should be thread-safe, since the module should be added to the `ModuleList` before the mutex is released, and the notifier doesn't modify the module list further, and the call is operates on local state and the `Target` instance. ### Deadlocked Thread backtraces: ``` * thread llvm#3, name = 'dbg.evt-handler', stop reason = signal SIGSTOP * frame #0: 0x00007f2f1e2973dc libc.so.6`futex_wait(private=0, expected=2, futex_word=0x0000563786bd5f40) at futex-internal.h:146:13 /*... a bunch of mutex related bt ... */ liblldb.so.21.0git`std::lock_guard<std::recursive_mutex>::lock_guard(this=0x00007f2f0f1927b0, __m=0x0000563786bd5f40) at std_mutex.h:229:19 frame llvm#8: 0x00007f2f27946eb7 liblldb.so.21.0git`ScanForGNUstepObjCLibraryCandidate(modules=0x0000563786bd5f28, TT=0x0000563786bd5eb8) at GNUstepObjCRuntime.cpp:60:41 frame llvm#9: 0x00007f2f27946c80 liblldb.so.21.0git`lldb_private::GNUstepObjCRuntime::CreateInstance(process=0x0000563785e1d360, language=eLanguageTypeObjC) at GNUstepObjCRuntime.cpp:87:8 frame llvm#10: 0x00007f2f2746fca5 liblldb.so.21.0git`lldb_private::LanguageRuntime::FindPlugin(process=0x0000563785e1d360, language=eLanguageTypeObjC) at LanguageRuntime.cpp:210:36 frame llvm#11: 0x00007f2f2742c9e3 liblldb.so.21.0git`lldb_private::Process::GetLanguageRuntime(this=0x0000563785e1d360, language=eLanguageTypeObjC) at Process.cpp:1516:9 ... frame llvm#21: 0x00007f2f2750b5cc liblldb.so.21.0git`lldb_private::Thread::GetSelectedFrame(this=0x0000563785e064d0, select_most_relevant=DoNoSelectMostRelevantFrame) at Thread.cpp:274:48 frame llvm#22: 0x00007f2f273f9957 liblldb.so.21.0git`lldb_private::ExecutionContextRef::SetTargetPtr(this=0x00007f2f0f193778, target=0x0000563786bd5be0, adopt_selected=true) at ExecutionContext.cpp:525:32 frame llvm#23: 0x00007f2f273f9714 liblldb.so.21.0git`lldb_private::ExecutionContextRef::ExecutionContextRef(this=0x00007f2f0f193778, target=0x0000563786bd5be0, adopt_selected=true) at ExecutionContext.cpp:413:3 frame llvm#24: 0x00007f2f270e80af liblldb.so.21.0git`lldb_private::Debugger::GetSelectedExecutionContext(this=0x0000563785d83bc0) at Debugger.cpp:1225:23 frame llvm#25: 0x00007f2f271bb7fd liblldb.so.21.0git`lldb_private::Statusline::Redraw(this=0x0000563785d83f30, update=true) at Statusline.cpp:136:41 ... * thread #1, name = 'lldb', stop reason = signal SIGSTOP * frame #0: 0x00007f2f1e2973dc libc.so.6`futex_wait(private=0, expected=2, futex_word=0x0000563785e1dd98) at futex-internal.h:146:13 /*... a bunch of mutex related bt ... */ liblldb.so.21.0git`std::lock_guard<std::recursive_mutex>::lock_guard(this=0x00007ffe62be0488, __m=0x0000563785e1dd98) at std_mutex.h:229:19 frame llvm#8: 0x00007f2f2742c8d1 liblldb.so.21.0git`lldb_private::Process::GetLanguageRuntime(this=0x0000563785e1d360, language=eLanguageTypeC_plus_plus) at Process.cpp:1510:41 frame llvm#9: 0x00007f2f2743c46f liblldb.so.21.0git`lldb_private::Process::ModulesDidLoad(this=0x0000563785e1d360, module_list=0x00007ffe62be06a0) at Process.cpp:6082:36 ... frame llvm#13: 0x00007f2f2715cf03 liblldb.so.21.0git`lldb_private::ModuleList::AppendImpl(this=0x0000563786bd5f28, module_sp=ptr = 0x563785cec560, use_notifier=true) at ModuleList.cpp:246:19 frame llvm#14: 0x00007f2f2715cf4c liblldb.so.21.0git`lldb_private::ModuleList::Append(this=0x0000563786bd5f28, module_sp=ptr = 0x563785cec560, notify=true) at ModuleList.cpp:251:3 ... frame llvm#19: 0x00007f2f274349b3 liblldb.so.21.0git`lldb_private::Process::ConnectRemote(this=0x0000563785e1d360, remote_url=(Data = "connect://localhost:1234", Length = 24)) at Process.cpp:3250:9 frame llvm#20: 0x00007f2f27411e0e liblldb.so.21.0git`lldb_private::Platform::DoConnectProcess(this=0x0000563785c59990, connect_url=(Data = "connect://localhost:1234", Length = 24), plugin_name=(Data = "gdb-remote", Length = 10), debugger=0x0000563785d83bc0, stream=0x00007ffe62be3128, target=0x0000563786bd5be0, error=0x00007ffe62be1ca0) at Platform.cpp:1926:23 ``` ## Test Plan: Built a hello world a.out Run server in one terminal: ``` ~/llvm/build/Debug/bin/lldb-server g :1234 a.out ``` Run client in another terminal ``` ~/llvm/build/Debug/bin/lldb -o "gdb-remote 1234" -o "b hello.cc:3" ``` Before: Client hangs indefinitely ``` ~/llvm/build/Debug/bin/lldb -o "gdb-remote 1234" -o "b main" (lldb) gdb-remote 1234 ^C^C ``` After: ``` ~/llvm/build/Debug/bin/lldb -o "gdb-remote 1234" -o "b hello.cc:3" (lldb) gdb-remote 1234 Process 837068 stopped * thread #1, name = 'a.out', stop reason = signal SIGSTOP frame #0: 0x00007ffff7fe4a60 ld-linux-x86-64.so.2`_start: -> 0x7ffff7fe4a60 <+0>: movq %rsp, %rdi 0x7ffff7fe4a63 <+3>: callq 0x7ffff7fe5780 ; _dl_start at rtld.c:522:1 ld-linux-x86-64.so.2`_dl_start_user: 0x7ffff7fe4a68 <+0>: movq %rax, %r12 0x7ffff7fe4a6b <+3>: movl 0x18067(%rip), %eax ; _dl_skip_args (lldb) b hello.cc:3 Breakpoint 1: where = a.out`main + 15 at hello.cc:4:13, address = 0x00005555555551bf (lldb) c Process 837068 resuming Process 837068 stopped * thread #1, name = 'a.out', stop reason = breakpoint 1.1 frame #0: 0x00005555555551bf a.out`main at hello.cc:4:13 1 #include <iostream> 2 3 int main() { -> 4 std::cout << "Hello World" << std::endl; 5 return 0; 6 } ```

…lvm#152156) With this new A320 in-order core, we follow adding the FeatureUseFixedOverScalableIfEqualCost feature to A510 and A520 (llvm#132246), which reaps the same code generation benefits of preferring fixed over scalable when the cost is equal. So when we have: ``` void foo(float* a, float* b, float* dst, unsigned n) { for (unsigned i = 0; i < n; ++i) dst[i] = a[i] + b[i]; } ``` When compiling without the feature enabled, we get: ``` ... ld1b { z0.b }, p0/z, [x0, x10] ld1b { z2.b }, p0/z, [x1, x10] add x12, x0, x10 ldr z1, [x12, #1, mul vl] add x12, x1, x10 ldr z3, [x12, #1, mul vl] fadd z0.s, z2.s, z0.s add x12, x2, x10 fadd z1.s, z3.s, z1.s dech x11 st1b { z0.b }, p0, [x2, x10] incb x10, all, mul llvm#2 str z1, [x12, #1, mul vl] ... ``` When compiling with, we get: ``` ... ldp q0, q1, [x12, #-16] ldp q2, q3, [x11, #-16] subs x13, x13, llvm#8 fadd v0.4s, v2.4s, v0.4s fadd v1.4s, v3.4s, v1.4s add x11, x11, llvm#32 add x12, x12, llvm#32 stp q0, q1, [x10, #-16] add x10, x10, llvm#32 ... ```

M68k's SETCC instruction (`scc`) distinctly fills the destination byte with all 1s. If boolean contents are set to `ZeroOrOneBooleanContent`, LLVM can mistakenly think the destination holds `0x01` instead of `0xff` and emit broken code as a result. This change corrects the boolean content type to `ZeroOrNegativeOneBooleanContent`. For example, this IR: ```llvm define dso_local signext range(i8 0, 2) i8 @testBool(i32 noundef %a) local_unnamed_addr #0 { entry: %cmp = icmp eq i32 %a, 4660 %. = zext i1 %cmp to i8 ret i8 %. } ``` would previously build as: ```asm testBool: ; @testBool cmpi.l llvm#4660, (4,%sp) seq %d0 and.l llvm#255, %d0 rts ``` Notice the `zext` is erroneously not clearing the low bits, and thus the register returns with 255 instead of 1. This patch fixes the issue: ```asm testBool: ; @testBool cmpi.l llvm#4660, (4,%sp) seq %d0 and.l #1, %d0 rts ``` Most of the tests containing `scc` suffered from the same value error as described above, so those tests have been updated to match the new output (which also logically corrects them).

## Problem When the new setting ``` set target.parallel-module-load true ``` was added, lldb began fetching modules from the devices from multiple threads simultaneously. This caused crashes of lldb when debugging on android devices. The top of the stack in the crash look something like this: ``` #0 0x0000555aaf2b27fe llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/opt/llvm/bin/lldb-dap+0xb87fe) #1 0x0000555aaf2b0a99 llvm::sys::RunSignalHandlers() (/opt/llvm/bin/lldb-dap+0xb6a99) llvm#2 0x0000555aaf2b2fda SignalHandler(int, siginfo_t*, void*) (/opt/llvm/bin/lldb-dap+0xb8fda) llvm#3 0x00007f9c02444560 __restore_rt /home/engshare/third-party2/glibc/2.34/src/glibc-2.34/signal/../sysdeps/unix/sysv/linux/libc_sigaction.c:13:0 llvm#4 0x00007f9c04ea7707 lldb_private::ConnectionFileDescriptor::Disconnect(lldb_private::Status*) (usr/bin/../lib/liblldb.so.15+0x22a7707) llvm#5 0x00007f9c04ea5b41 lldb_private::ConnectionFileDescriptor::~ConnectionFileDescriptor() (usr/bin/../lib/liblldb.so.15+0x22a5b41) llvm#6 0x00007f9c04ea5c1e lldb_private::ConnectionFileDescriptor::~ConnectionFileDescriptor() (usr/bin/../lib/liblldb.so.15+0x22a5c1e) llvm#7 0x00007f9c052916ff lldb_private::platform_android::AdbClient::SyncService::Stat(lldb_private::FileSpec const&, unsigned int&, unsigned int&, unsigned int&) (usr/bin/../lib/liblldb.so.15+0x26916ff) llvm#8 0x00007f9c0528b9dc lldb_private::platform_android::PlatformAndroid::GetFile(lldb_private::FileSpec const&, lldb_private::FileSpec const&) (usr/bin/../lib/liblldb.so.15+0x268b9dc) ``` Our workaround was to set `set target.parallel-module-load ` to `false` to avoid the crash. ## Background PlatformAndroid creates two different classes with one stateful adb connection shared between the two -- one through AdbClient and another through AdbClient::SyncService. The connection management and state is complex, and seems to be responsible for the segfault we are seeing. The AdbClient code resets these connections at times, and re-establishes connections if they are not active. Similarly, PlatformAndroid caches its SyncService, which uses an AdbClient class, but the SyncService puts its connection into a different 'sync' state that is incompatible with a standard connection. ## Changes in this diff * This diff refactors the code to (hopefully) have clearer ownership of the connection, clearer separation of AdbClient and SyncService by making a new class for clearer separations of concerns, called AdbSyncService. * New unit tests are added * Additional logs were added (see llvm#145382 (comment) for details)

…namic (llvm#153420) Canonicalizing the following IR: ``` func.func @mul_zero_dynamic_nofold(%arg0: tensor<?x17xf32>) -> tensor<?x17xf32> { %0 = "tosa.const"() <{values = dense<0.000000e+00> : tensor<1x1xf32>}> : () -> tensor<1x1xf32> %1 = "tosa.const"() <{values = dense<0> : tensor<1xi8>}> : () -> tensor<1xi8> %2 = tosa.mul %arg0, %0, %1 : (tensor<?x17xf32>, tensor<1x1xf32>, tensor<1xi8>) -> tensor<?x17xf32> return %2 : tensor<?x17xf32> } ``` resulted in a crash ``` #0 0x000056513187e8db backtrace (./build-release/bin/mlir-opt+0x9d698db) #1 0x0000565131b17737 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) /local-ssd/sayans/Softwares/llvm-repo/llvm-project-latest/llvm/lib/Support/Unix/Signals.inc:838:8 llvm#2 0x0000565131b187f3 PrintStackTraceSignalHandler(void*) /local-ssd/sayans/Softwares/llvm-repo/llvm-project-latest/llvm/lib/Support/Unix/Signals.inc:918:1 llvm#3 0x0000565131b18c30 llvm::sys::RunSignalHandlers() /local-ssd/sayans/Softwares/llvm-repo/llvm-project-latest/llvm/lib/Support/Signals.cpp:105:18 llvm#4 0x0000565131b18c30 SignalHandler(int, siginfo_t*, void*) /local-ssd/sayans/Softwares/llvm-repo/llvm-project-latest/llvm/lib/Support/Unix/Signals.inc:409:3 llvm#5 0x00007f2e4165b050 (/lib/x86_64-linux-gnu/libc.so.6+0x3c050) llvm#6 0x00007f2e416a9eec __pthread_kill_implementation ./nptl/pthread_kill.c:44:76 llvm#7 0x00007f2e4165afb2 raise ./signal/../sysdeps/posix/raise.c:27:6 llvm#8 0x00007f2e41645472 abort ./stdlib/abort.c:81:7 llvm#9 0x00007f2e41645395 _nl_load_domain ./intl/loadmsgcat.c:1177:9 llvm#10 0x00007f2e41653ec2 (/lib/x86_64-linux-gnu/libc.so.6+0x34ec2) llvm#11 0x00005651443ec4ba mlir::DenseIntOrFPElementsAttr::getRaw(mlir::ShapedType, llvm::ArrayRef<char>) /local-ssd/sayans/Softwares/llvm-repo/llvm-project-latest/mlir/lib/IR/BuiltinAttributes.cpp:1361:3 llvm#12 0x00005651443f1209 mlir::DenseElementsAttr::resizeSplat(mlir::ShapedType) /local-ssd/sayans/Softwares/llvm-repo/llvm-project-latest/mlir/lib/IR/BuiltinAttributes.cpp:0:10 llvm#13 0x000056513f76f2b6 mlir::tosa::MulOp::fold(mlir::tosa::MulOpGenericAdaptor<llvm::ArrayRef<mlir::Attribute>>) /local-ssd/sayans/Softwares/llvm-repo/llvm-project-latest/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp:0:0 ``` from the folder for `tosa::mul` since the zero value was being reshaped to `?x17` size which isn't supported. AFAIK, `tosa.const` requires all dimensions to be static. So in this case, the fix is to not to fold the op.

Jianhui-Li added 5 commits July 1, 2025 23:00

init code

b9a6d98

add tests

2465050

git-clang-format

1077871

add more tests

42baa22

git-clang-format

204d347

chencha3 reviewed Jul 3, 2025

View reviewed changes

Jianhui-Li added 18 commits July 12, 2025 05:07

add ui64 case support

2793c81

modify ui64 test

f23ea03

Merge branch 'main' into dialect-assembly-format

0bb958b

remove unnecessary comments

6793689

fix VectorToXeGPU tests

4a96c71

tweak default offset value

689a8a5

git-clang-format

02d3795

add builders

01718f4

git-clang-format

5ef6ca9

Merge branch 'main' into dialect-assembly-format

26a222d

simplify custom parser

882313f

add comma before shape and strides

456534a

tie the offsets rank to input tensor shape instead of tdesc

b6f016e

git-clang-format

cd518d2

addverifier for invalid cases

546a3f7

git-clang-format

7846955

add comments

ded9552

simplify custom print

97b6e39

Jianhui-Li added 3 commits July 17, 2025 15:54

git-clang-format

ed1d48e

Merge branch 'main' into dialect-assembly-format

b3edff6

use simpler interface for DenseI64ArrayAttr

d3e935b

address feedback

205fea7

Jianhui-Li closed this Jul 24, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Dialect assembly format - make offsets optional for create_nd_tdesc #1

Dialect assembly format - make offsets optional for create_nd_tdesc #1

Uh oh!

Jianhui-Li commented Jul 2, 2025 •

edited

Loading

Uh oh!

chencha3 Jul 3, 2025

Uh oh!

chencha3 Jul 3, 2025

Uh oh!

akroviakov Jul 8, 2025 •

edited

Loading

Uh oh!

Jianhui-Li Jul 10, 2025

Uh oh!

Jianhui-Li Jul 10, 2025

Uh oh!

chencha3 Jul 3, 2025

Uh oh!

Jianhui-Li commented Jul 24, 2025

Uh oh!

Uh oh!

Dialect assembly format - make offsets optional for create_nd_tdesc #1

Dialect assembly format - make offsets optional for create_nd_tdesc #1

Uh oh!

Conversation

Jianhui-Li commented Jul 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chencha3 Jul 3, 2025

Choose a reason for hiding this comment

Uh oh!

chencha3 Jul 3, 2025

Choose a reason for hiding this comment

Uh oh!

akroviakov Jul 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Jianhui-Li Jul 10, 2025

Choose a reason for hiding this comment

Uh oh!

Jianhui-Li Jul 10, 2025

Choose a reason for hiding this comment

Uh oh!

chencha3 Jul 3, 2025

Choose a reason for hiding this comment

Uh oh!

Jianhui-Li commented Jul 24, 2025

Uh oh!

Uh oh!

Jianhui-Li commented Jul 2, 2025 •

edited

Loading

akroviakov Jul 8, 2025 •

edited

Loading