Problem Description
fx.make_layout_tv API fails due to compilation issue. I see unit tests are missing for this API.
Operating System
Ubuntu 24.04.3 LTS (Noble Numbat)
CPU
AMD EPYC 9575F 64-Core Processor
GPU
AMD Instinct MI355, gfx950
ROCm Version
ROCm 7.1.0
ROCm Component
No response
Steps to Reproduce
import torch
import flydsl.compiler as flyc
import flydsl.expr as fx
@flyc.jit
def tv_layout(
mA: fx.Tensor,
):
thr_layout = fx.make_layout((4,64), (64,1))
val_layout = fx.make_layout((4,8), (8,1))
tiler_mn, tv_layout = fx.make_layout_tv(thr_layout, val_layout)
M, N = (2048, 2048)
A = torch.randn(M,N,dtype=torch.bfloat16).cuda()
tv_layout(A)
python3 examples/tv_layout.py
Traceback (most recent call last):
File "/workspace/FlyDSL2/examples/tv_layout.py", line 16, in <module>
tv_layout(A)
File "/workspace/FlyDSL2/python/flydsl/compiler/jit_function.py", line 816, in __call__
compiled_module = MlirCompiler.compile(module, arch=backend.target.arch, func_name=self.func.__name__)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/FlyDSL2/python/flydsl/compiler/jit_function.py", line 372, in compile
module.operation.verify()
flydsl._mlir._mlir_libs._mlir.ir.MLIRError: Verification failed:
error: "tmp = make_layout((thr_size, val_size), (1, thr_size))"("/workspace/FlyDSL2/python/flydsl/expr/derived.py":169:10): 'fly.make_shape' op operand #0 must be variadic of 32-bit signless integer or 64-bit signless integer, but got '!fly.int_tuple<256>'
note: "tmp = make_layout((thr_size, val_size), (1, thr_size))"("/workspace/FlyDSL2/python/flydsl/expr/derived.py":169:10): see current operation: %9 = "fly.make_shape"(%7, %8) : (!fly.int_tuple<256>, !fly.int_tuple<32>) -> !fly.int_tuple<(?,?)>
(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support
No response
Additional Information
No response
Problem Description
fx.make_layout_tv API fails due to compilation issue. I see unit tests are missing for this API.
Operating System
Ubuntu 24.04.3 LTS (Noble Numbat)
CPU
AMD EPYC 9575F 64-Core Processor
GPU
AMD Instinct MI355, gfx950
ROCm Version
ROCm 7.1.0
ROCm Component
No response
Steps to Reproduce
(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support
No response
Additional Information
No response