start support for adreno 830

extra/qcom_gpu_driver/opencl_ioctl.py

@@ -177,13 +177,14 @@ def ioctl(fd, request, argp):
       for i in range(s.numcmds):
         cmd = get_struct(s.cmdlist+ctypes.sizeof(msm_kgsl.struct_kgsl_command_object)*i, msm_kgsl.struct_kgsl_command_object)
         if IOCTL > 0: print(f"cmd {i}:", format_struct(cmd))
-        parse_cmd_buf(get_mem(cmd.gpuaddr, cmd.size))
+        #parse_cmd_buf(get_mem(cmd.gpuaddr, cmd.size))
       for i in range(s.numobjs):
         obj = get_struct(s.objlist+s.objsize*i, msm_kgsl.struct_kgsl_command_object)
         if IOCTL > 0:
           print(f"obj {i}:", format_struct(obj))
-          print(format_struct(msm_kgsl.struct_kgsl_cmdbatch_profiling_buffer.from_buffer_copy(get_mem(obj.gpuaddr, obj.size))))
+          #print(format_struct(msm_kgsl.struct_kgsl_cmdbatch_profiling_buffer.from_buffer_copy(get_mem(obj.gpuaddr, obj.size))))
         #hexdump(get_mem(obj.gpuaddr, obj.size))
+        pass
   else:
     #print(f"ioctl({fd=}, (dir:{idir}, size:0x{size:3X}, type:{itype:d}, nr:0x{nr:2X}), {argp=:X}) = {ret=}")
     pass

tinygrad/runtime/ops_cl.py

@@ -2,9 +2,10 @@ from __future__ import annotations
 from typing import cast
 import ctypes, functools, hashlib
 from tinygrad.runtime.autogen import opencl as cl
-from tinygrad.helpers import init_c_var, to_char_p_p, from_mv, OSX, DEBUG, mv_address, suppress_finalizing
+from tinygrad.helpers import init_c_var, to_char_p_p, from_mv, OSX, DEBUG, mv_address, suppress_finalizing, getenv
 from tinygrad.renderer.cstyle import OpenCLRenderer, IntelRenderer
 from tinygrad.device import BufferSpec, LRUAllocator, Compiled, Compiler, CompileError
+if getenv("IOCTL"): import extra.qcom_gpu_driver.opencl_ioctl  # noqa: F401 # pylint: disable=unused-import
 
 # see test/external/external_osx_profiling.py to determine this ratio. it's in like GPU clocks or something
 OSX_TIMING_RATIO = (125/3) if OSX else 1.0