objdump intel syntax (#7605)

* objdump intel syntax

* test for objdump intel syntax

* add disassemble to ClangCompiler and LLVMCompiler. Use just llvm-objdump

* linter

test/test_objdump.py

@@ -0,0 +1,16 @@
+from tinygrad import Device
+from tinygrad.runtime.ops_clang import ClangCompiler
+import io, contextlib, unittest
+
+@unittest.skipUnless(Device.DEFAULT == "CLANG", "compile with clang")
+class TestCpuObjDump(unittest.TestCase):
+  def test_intel_syntax(self):
+    src = "int main() { return 0; }"
+    lib = ClangCompiler().compile(src)
+    out = io.StringIO()
+    with contextlib.redirect_stdout(out):
+      ClangCompiler().disassemble(lib)
+    assert "%" not in out.getvalue(), "objdump disassembly should be intel syntax"
+
+if __name__ == '__main__':
+  unittest.main()

tinygrad/helpers.py

@@ -259,10 +259,12 @@ def cpu_time_execution(cb, enable):
   cb()
   if enable: return time.perf_counter()-st
 
-def cpu_objdump(lib, objdump_tool='objdump'):
-  with tempfile.NamedTemporaryFile(delete=True) as f:
-    pathlib.Path(f.name).write_bytes(lib)
-    print(subprocess.check_output([objdump_tool, '-d', f.name]).decode('utf-8'))
+def cpu_objdump(lib):
+  try:
+    with tempfile.NamedTemporaryFile(delete=True) as f:
+      pathlib.Path(f.name).write_bytes(lib)
+      print(subprocess.check_output(['llvm-objdump', '-d', '-M', 'intel', f.name]).decode('utf-8'))
+  except Exception as e: print(str(e))
 
 # *** ctypes helpers
 

tinygrad/runtime/ops_clang.py

@@ -1,7 +1,7 @@
 from typing import Optional, List
 import ctypes, subprocess, pathlib, tempfile
 from tinygrad.device import Compiled, Compiler, MallocAllocator
-from tinygrad.helpers import cpu_time_execution, DEBUG, cpu_objdump
+from tinygrad.helpers import cpu_time_execution, cpu_objdump
 from tinygrad.renderer.cstyle import ClangRenderer
 
 class ClangCompiler(Compiler):
@@ -16,9 +16,10 @@ class ClangCompiler(Compiler):
                                '-', '-o', str(output_file.name)], input=src.encode('utf-8'))
       return pathlib.Path(output_file.name).read_bytes()
 
+  def disassemble(self, lib:bytes): cpu_objdump(lib)
+
 class ClangProgram:
   def __init__(self, name:str, lib:bytes):
-    if DEBUG >= 6: cpu_objdump(lib)
     self.name, self.lib = name, lib
     # write to disk so we can load it
     with tempfile.NamedTemporaryFile(delete=True) as cached_file_path:

tinygrad/runtime/ops_dsp.py

@@ -3,7 +3,7 @@ from typing import Tuple, Any
 import ctypes, os, mmap, tempfile, pathlib, array, functools, threading, contextlib, sys
 assert sys.platform != 'win32'
 from tinygrad.device import BufferOptions, Compiled, Allocator
-from tinygrad.helpers import from_mv, getenv, DEBUG, round_up, mv_address, to_mv, cpu_objdump
+from tinygrad.helpers import from_mv, getenv, round_up, mv_address, to_mv
 from tinygrad.runtime.ops_clang import ClangCompiler
 from tinygrad.renderer.cstyle import DSPRenderer
 from tinygrad.runtime.autogen import libc, qcom_dsp
@@ -23,7 +23,6 @@ def rpc_prep_args(ins=None, outs=None, in_fds=None):
 class DSPProgram:
   def __init__(self, device:DSPDevice, name:str, lib:bytes):
     self.device, self.lib = device, lib
-    if DEBUG >= 6: cpu_objdump(lib, objdump_tool='llvm-objdump')
 
   def __call__(self, *bufs, vals:Tuple[int, ...]=(), wait=False):
     if len(bufs) >= 16: raise RuntimeError(f"Too many buffers to execute: {len(bufs)}")

tinygrad/runtime/ops_llvm.py

@@ -22,10 +22,10 @@ class LLVMCompiler(Compiler):
     self.optimizer.run(mod)
     if DEBUG >= 5: print(self.device.target_machine.emit_assembly(mod))
     return self.device.target_machine.emit_object(mod)
+  def disassemble(self, lib: bytes): cpu_objdump(lib)
 
 class LLVMProgram:
   def __init__(self, device:LLVMDevice, name:str, lib:bytes):
-    if DEBUG >= 6: cpu_objdump(lib)
     self.name, self.lib = name, lib
     device.engine.add_object_file(llvm.object_file.ObjectFileRef.from_data(lib))
     self.fxn = device.engine.get_function_address(name)