github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-04-29 03:00:14 -04:00
Code Issues Packages Projects Releases Wiki Activity

new style CPULLVMRenderer (#14629)

Browse Source
This commit is contained in:
Christopher Milan
2026-02-08 18:05:01 -08:00
committed by GitHub
parent 0e505951b0
commit 4ad787ece2
2 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
tinygrad/renderer/llvmir.py
View File

@@ -205,6 +205,9 @@ class CPULLVMRenderer(LLVMRenderer):
string_rewrite = base_rewrite + PatternMatcher([(UPat(Ops.WMMA, name="wmma"), render_wmma_amx)])
def render(self, uops: list[UOp]) -> str: return "\n".join((k:=self._render_kernel(uops))[0] + (k[1], self._render_footer(uops)))
def _render_footer(self, uops: list[UOp]) -> str: return 'attributes #0 = { alwaysinline nounwind "no-builtins" "no-trapping-math"="true" }'
def __init__(self):
from tinygrad.runtime.support.compiler_cpu import CPULLVMCompiler
self.compiler = CPULLVMCompiler()
barrier = 'fence syncscope("workgroup") release\ntail call void @llvm.amdgcn.s.barrier()\nfence syncscope("workgroup") acquire\n'
code_for_workitem = {"g": lambda x: f"tail call i32 @llvm.amdgcn.workgroup.id.{chr(120+int(x))}()",

3
tinygrad/runtime/ops_cpu.py
View File

@@ -8,7 +8,6 @@ from tinygrad.runtime.support.hcq import CLikeArgsState
from tinygrad.renderer.cstyle import ClangJITRenderer
from tinygrad.renderer.llvmir import CPULLVMRenderer
from tinygrad.renderer.nir import LVPRenderer
from tinygrad.runtime.support.compiler_cpu import CPULLVMCompiler
from tinygrad.runtime.support.elf import jit_loader
from tinygrad.uop.ops import sint
@@ -134,6 +133,6 @@ class CPUDevice(HCQCompiled):
def __init__(self, device:str=""):
self.tasks:queue.Queue = queue.Queue()
CPUWorker(self, self.tasks, thread_id=0).start()
compilers = CompilerSet([CompilerPair(ClangJITRenderer, None), CompilerPair(CPULLVMRenderer, CPULLVMCompiler, ctrl_var=CPU_LLVM),
compilers = CompilerSet([CompilerPair(ClangJITRenderer, None), CompilerPair(CPULLVMRenderer, None, ctrl_var=CPU_LLVM),
CompilerPair(LVPRenderer, None, ctrl_var=CPU_LVP)], ctrl_var=CPU_CC)
super().__init__(device, CPUAllocator(self), compilers, functools.partial(CPUProgram, self), CPUSignal, CPUComputeQueue)