set minimum sizes and update tests

Author: andrej

programming_examples/basic/combined_transpose/combined_transpose.py

@@ -22,6 +22,12 @@ def shuffle_transpose(dev, M, N, m, n, s, dtype):
     assert m % s == 0
     assert n % s == 0
 
+    # Minimum tile sizes required by the two kernels
+    if s == 8:
+        assert m > 8 and n > 8
+    elif s == 16:
+        assert m > 32 and n > 32
+
     # Define tensor types
     matrix_ty = np.ndarray[(M, N), dtype]
     tile_ty = np.ndarray[(m, n,), dtype]
@@ -41,6 +47,12 @@ def shuffle_transpose(dev, M, N, m, n, s, dtype):
         sizes=[M // m, N // n, m, n],
         strides=[m * N, n, N, 1]
     )
+    tap_in_L3L2_straight = TensorAccessPattern(
+        tensor_dims=(M, N),
+        offset=0,
+        sizes=[M // m, N // n, m, n],
+        strides=[m * N, n, N, 1]
+    )
     tap_in_L2L1 = TensorAccessPattern(
         tensor_dims=(M, N),
         offset=0,
@@ -53,6 +65,12 @@ def shuffle_transpose(dev, M, N, m, n, s, dtype):
         sizes=[M // m, N // n, n, m],
         strides=[m, n * M, M, 1]
     )
+    tap_out_L1L3_straight = TensorAccessPattern(
+        tensor_dims=(N, M),
+        offset=0,
+        sizes=[1, 1, 1, M*N],
+        strides=[0, 0, 0, 1]
+    )
 
     in_L3L2_fifo = ObjectFifo(tile_ty, name="in_L3L2_fifo")
     in_L2L1_fifo = in_L3L2_fifo.cons(

programming_examples/basic/combined_transpose/tests/1_run_makefile_i8.lit

@@ -6,6 +6,6 @@
 // RUN: mkdir -p test_1_run_makefile_i8
 // RUN: cd test_1_run_makefile_i8
 // RUN: make -f %S/../Makefile clean
-// RUN: make -f %S/../Makefile M=320 N=288 m=64 n=32 dtype_in=i8
-// RUN: %run_on_npu1% make -f %S/../Makefile run M=320 N=288 m=64 n=32 dtype_in=i8
+// RUN: make -f %S/../Makefile M=320 N=288 m=64 n=32 dtype=i8
+// RUN: %run_on_npu1% make -f %S/../Makefile run M=320 N=288 m=64 n=32 dtype=i8
 // CHECK: PASS!

programming_examples/basic/combined_transpose/tests/2_run_makefile_i16.lit renamed to programming_examples/basic/combined_transpose/tests/2_run_makefile_i16_s4.lit

@@ -3,9 +3,9 @@
 //
 // REQUIRES: ryzen_ai_npu1, peano
 //
-// RUN: mkdir -p test_2_run_makefile_i16
+// RUN: mkdir -p test_2_run_makefile_i16_s4
 // RUN: cd test_2_run_makefile_i16
 // RUN: make -f %S/../Makefile clean
-// RUN: make -f %S/../Makefile M=256 N=128 m=32 n=16 dtype_in=i16
-// RUN: %run_on_npu1% make -f %S/../Makefile run M=256 N=128 m=32 n=16 dtype_in=i16
+// RUN: make -f %S/../Makefile M=256 N=128 m=32 n=16 s=4 dtype=i16
+// RUN: %run_on_npu1% make -f %S/../Makefile run M=256 N=128 m=32 n=16 s=4 dtype=i16
 // CHECK: PASS!

programming_examples/basic/combined_transpose/tests/3_run_makefile_i32.lit renamed to programming_examples/basic/combined_transpose/tests/3_run_makefile_i32_s4.lit

@@ -3,9 +3,9 @@
 //
 // REQUIRES: ryzen_ai_npu1, peano
 //
-// RUN: mkdir -p test_3_run_makefile_i32
+// RUN: mkdir -p test_3_run_makefile_i32_s4
 // RUN: cd test_3_run_makefile_i32
 // RUN: make -f %S/../Makefile clean
-// RUN: make -f %S/../Makefile M=128 N=1024 m=16 n=32 dtype_in=i32
-// RUN: %run_on_npu1% make -f %S/../Makefile run M=128 N=1024 m=16 n=32 dtype_in=i32
+// RUN: make -f %S/../Makefile M=128 N=1024 m=16 n=32 dtype=i32
+// RUN: %run_on_npu1% make -f %S/../Makefile run M=128 N=1024 m=16 n=32 dtype=i32
 // CHECK: PASS!

programming_examples/basic/combined_transpose/tests/4_run_makefile_i32_s8.lit

@@ -0,0 +1,11 @@
+// (c) Copyright 2024 Advanced Micro Devices, Inc.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+// REQUIRES: ryzen_ai_npu1, peano
+//
+// RUN: mkdir -p test_4_run_makefile_i32_s8
+// RUN: cd test_3_run_makefile_i32
+// RUN: make -f %S/../Makefile clean
+// RUN: make -f %S/../Makefile M=128 N=1024 m=64 n=32 s=8 dtype=i32
+// RUN: %run_on_npu1% make -f %S/../Makefile run M=128 N=1024 m=64 n=32 s=8 dtype=i32
+// CHECK: PASS!