add restrict qualifier to inputs in c backend (#593)

* add restrict qualifier for clang backend convolution inputs/ outputs
see https://godbolt.org/z/Tb9jMxWfx for generated assembly

* enable more checks

* inline fmax to motivate the compiler to inline some more

* fix if else binding power

examples/compile_efficientnet.py

@@ -5,7 +5,7 @@ import ast
 
 def compile_net(run, special_names):
   # c header
-  cprog = ["#include <stdio.h>", "#include <math.h>","#define max(x,y) fmax(x,y)"] 
+  cprog = ["#include <stdio.h>", "#include <math.h>", "#define max(x,y) ((x>y)?x:y)"]
 
   # functions that run the net
   bufs = {}

tinygrad/llops/ops_gpu.py

@@ -324,7 +324,7 @@ class CLASTKernel(ASTKernel):
 
     # kernel function definition
     function_name = ("re_S" if self.reduceop else "ew_S") + '_'.join([str(x) for x in self.bufs[0].shape if x != 1])
-    buftypes = [f"{'read_only' if i > 0 else 'write_only'} image2d_t" if isinstance(x._buf, CLImage) else (CLProgram.buffer_prefix+self.buftokens[i].decltype()) for i,x in enumerate(self.bufs)]
+    buftypes = [f"{'read_only' if i > 0 else 'write_only'} image2d_t" if isinstance(x._buf, CLImage) else (CLProgram.buffer_prefix+self.buftokens[i].decltype() + ("restrict" if CLANG else "")) for i,x in enumerate(self.bufs)]
     self.kernel = list(self.prekernel) + [f"{CLProgram.kernel_prefix} void {function_name}(",] + \
       [', '.join([f'{t} data{i}' for i,t in enumerate(buftypes) if i not in self.bufs_to_delete] + CLProgram.extra_args)] + \
       [") {\n"] + self.kernel

tinygrad/runtime/clang.py

@@ -26,12 +26,12 @@ class CLProgram:
     self.name = f"{name}{('_N'+str(CLProgram.kernel_cnt[name])) if CLProgram.kernel_cnt[name] else str()}" if rename else name
     CLProgram.kernel_cnt[name] += 1
     self.prg = prg.replace(f"{name}(", f"{self.name}(")
-    prg = "#include <math.h>\n#define max(x,y) fmax(x,y)\n" + prg
+    prg = "#include <math.h>\n#define max(x,y) ((x>y)?x:y)\n" + prg
     if DEBUG >= 4: print(prg)  # TODO: outside runtime!
     # TODO: is there a way to not write this to disk?
     fn = f"/tmp/clang_{hashlib.md5(prg.encode('utf-8')).hexdigest()}.{'dylib' if OSX else 'so'}"
     if not os.path.exists(fn):
-      subprocess.check_output(['clang', '-shared', '-O2', '-lm', '-fPIC', '-x', 'c', '-', '-o', fn+".tmp"], input=prg.encode('utf-8'))
+      subprocess.check_output(['clang', '-shared', '-O2', '-Wall','-Werror', '-lm', '-fPIC', '-x', 'c', '-', '-o', fn+".tmp"], input=prg.encode('utf-8'))
       os.rename(fn+".tmp", fn)
     self.lib = ctypes.CDLL(fn)
     self.fxn = self.lib[name]