assembly gemm clean ups, instructions for cli (#13892)

CLAUDE.md

@@ -208,3 +208,9 @@ Key patterns to watch (from ResNet50 benchmark):
 - `vmin==vmax folding`: ~55ms, 0.33% match rate - checks 52K ops but rarely matches
 
 Patterns with 0% match rate are workload-specific overhead. They may be useful in other workloads, so don't remove them without understanding their purpose.
+
+## AMD Performance Counter Profiling
+
+Set VIZ to `-2` to save performance counters traces for the AMD backend.
+
+Use the CLI in `./extra/sqtt/roc.py` to explore the trace.

extra/gemm/asm/test.py

@@ -1,7 +1,7 @@
 # Run assembly on the AMD runtime and check correctness
 # VIZ=2 to profile
 import pathlib
-from tinygrad import Tensor, Device, dtypes
+from tinygrad import Tensor, Device, dtypes, Context
 from tinygrad.engine.realize import ExecItem, CompiledRunner
 from tinygrad.renderer import ProgramSpec
 from tinygrad.uop.ops import track_rewrites, UOp
@@ -55,9 +55,10 @@ def get_asm_prg() -> ProgramSpec:
 eis.append(ExecItem(ast, [C_asm.uop.buffer, from_torch(B).uop.buffer, from_torch(A).uop.buffer], fixedvars={"SZ":N, "NUM_WG":NUM_WG},
                     prg=CompiledRunner(get_asm_prg())))
 
-for ei in eis:
+with Context(DEBUG=2):
-  et = ei.run(wait=True)
+  for ei in eis:
-  print(f"{(N*N*N*2 / et)*1e-12:.2f} REAL TFLOPS")
+    et = ei.run(wait=True)
+    print(f"{(N*N*N*2 / et)*1e-12:.2f} REAL TFLOPS")
 
 # ** correctness