remove LOAD where valid is an empty set

356 -> 354 valids
chenyuxyz · Sep 18, 2024 · 1fa44b7 · 1fa44b7
1 parent d4b662c
commit 1fa44b7
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 1 deletion.
diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
@@ -197,7 +197,7 @@ jobs:
       - if: ${{ matrix.task == 'optimage' }}
         name: Test openpilot model compile and size
         run: |
-          PYTHONPATH="." DEBUG=2 ALLOWED_KERNEL_COUNT=208 ALLOWED_GATED_READ_IMAGE=356 FLOAT16=1 DEBUGCL=1 GPU=1 IMAGE=2 python examples/openpilot/compile2.py
+          PYTHONPATH="." DEBUG=2 ALLOWED_KERNEL_COUNT=208 ALLOWED_GATED_READ_IMAGE=354 FLOAT16=1 DEBUGCL=1 GPU=1 IMAGE=2 python examples/openpilot/compile2.py
           python -c 'import os; assert os.path.getsize("/tmp/output.thneed") < 100_000_000'
       - if: ${{ matrix.task == 'optimage' }}
         name: Test openpilot model correctness (float32)

diff --git a/tinygrad/codegen/uopgraph.py b/tinygrad/codegen/uopgraph.py
@@ -195,6 +195,9 @@ def simplify_valid_image_load(load:UOp, buf:UOp):
         lower = -stmt.src[1].arg+1
         drop_stmt.append(stmt)
 
+  # valid is an empty set
+  if upper < lower: return UOp(UOps.LOAD, load.dtype, (buf, idx, invalid_val, valid.const_like(False)))
+
   new_indx0, new_indx1 = None, None
   if (L:=(lower * c + d)) // m == (U:=(upper * c + d)) // m:  # in the same row
     if (L % m - c < 0) and (U % m + c >= m):  # spans the whole row