From cdb78954cbf4d3b2012ec8633ac07ddc0c2a19bd Mon Sep 17 00:00:00 2001 From: Christopher Milan Date: Mon, 9 Feb 2026 22:03:46 -0800 Subject: [PATCH] better cl compiler name (#14660) cl_compiler instead of compiler because overriding Compiled.compiler seems more confusing --- tinygrad/runtime/ops_cl.py | 5 +++-- tinygrad/runtime/ops_qcom.py | 2 +- 2 files changed, 4 insertions(+), 3 deletions(-) diff --git a/tinygrad/runtime/ops_cl.py b/tinygrad/runtime/ops_cl.py index cdc8119403..265f35c682 100644 --- a/tinygrad/runtime/ops_cl.py +++ b/tinygrad/runtime/ops_cl.py @@ -39,7 +39,7 @@ class CLCompiler(Compiler): class CLProgram: def __init__(self, device:CLDevice, name:str, lib:bytes, buf_dtypes=[], **kwargs): - self.dev, self.name, self.lib, self.buf_dtypes = device, name, device.clc.compile_cached(lib.decode()), buf_dtypes + self.dev, self.name, self.lib, self.buf_dtypes = device, name, device.cl_compiler.compile_cached(lib.decode()), buf_dtypes self.program = checked(cl.clCreateProgramWithBinary(device.context, 1, device.device_id, (ctypes.c_size_t * 1)(len(self.lib)), to_char_p_p([self.lib], ctypes.c_ubyte), binary_status := ctypes.c_int32(), errcode_ret := ctypes.c_int32()), errcode_ret) @@ -125,8 +125,9 @@ class CLDevice(Compiled): ctypes.string_at(buf, size=total.value).decode())[1] renderer = IntelRenderer if "cl_intel_subgroup_matrix_multiply_accumulate" in self.device_exts else OpenCLRenderer - self.clc = CLCompiler(self, f"{hashlib.md5(self.device_name.encode() + self.driver_version.encode()).hexdigest()}") + self.cl_compiler = CLCompiler(self, f"{hashlib.md5(self.device_name.encode() + self.driver_version.encode()).hexdigest()}") super().__init__(device, CLAllocator(self), CompilerSet([(renderer, None)]), functools.partial(CLProgram, self)) + def synchronize(self): check(cl.clFinish(self.queue)) self.pending_copyin.clear() diff --git a/tinygrad/runtime/ops_qcom.py b/tinygrad/runtime/ops_qcom.py index 4bcdc5be55..99c8c1e29f 100644 --- a/tinygrad/runtime/ops_qcom.py +++ b/tinygrad/runtime/ops_qcom.py @@ -249,7 +249,7 @@ class QCOMProgram(HCQProgram): self.fregs, self.hregs = v.info.max_reg + 1, v.info.max_half_reg + 1 self.consts_info:list[tuple] = [] else: - self._parse_lib(lib:=self.dev.cl_dev.clc.compile_cached(lib.decode())) + self._parse_lib(lib:=self.dev.cl_dev.cl_compiler.compile_cached(lib.decode())) if DEBUG >= 7: fromimport('tinygrad.runtime.support.compiler_mesa', 'disas_adreno')(lib[(ofs:=_read_lib(lib, 0xc0)):ofs+_read_lib(lib, 0x100)]) self.lib_gpu: HCQBuffer = self.dev.allocator.alloc(self.image_size, buf_spec:=BufferSpec(cpu_access=True, nolru=True))