backend: (riscv) remove optimization level from test-lower-linalg-to-snitch (xdslproject#3385)

We introduced this for benchmarking purposes, but it was easier to just
construct the pass pipeline programmatically instead in the test setup,
so this is no longer used/useful.

Author: superlopuh

tests/filecheck/projects/riscv-backend-paper/nsnet.mlir

@@ -1,6 +1,4 @@
 // RUN: xdsl-opt -p arith-add-fastmath,test-lower-linalg-to-snitch -t riscv-asm %s | filecheck %s
-// RUN: xdsl-opt -p arith-add-fastmath,test-lower-linalg-to-snitch{optimization-level=4} -t riscv-asm %s | filecheck %s
-// RUN: xdsl-opt -p arith-add-fastmath,test-lower-linalg-to-snitch{optimization-level=0} -t riscv-asm %s | filecheck %s --check-prefix=CHECK-OPT
 
 func.func @main$async_dispatch_0_matmul_transpose_b_1x400x161_f64$xdsl_kernel1(%arg0: memref<1x161xf64>, %arg1: memref<5x161xf64, strided<[161, 1]>>, %arg2: memref<1x5xf64, strided<[40, 1]>>) {
   linalg.generic {indexing_maps = [affine_map<(d0, d1, d2) -> (d0, d2)>, affine_map<(d0, d1, d2) -> (d1, d2)>, affine_map<(d0, d1, d2) -> (d0, d1)>], iterator_types = ["parallel", "parallel", "reduction"]} ins(%arg0, %arg1 : memref<1x161xf64>, memref<5x161xf64, strided<[161, 1]>>) outs(%arg2 : memref<1x5xf64, strided<[40, 1]>>) {
@@ -59,50 +57,3 @@ func.func @main$async_dispatch_0_matmul_transpose_b_1x400x161_f64$xdsl_kernel1(%
 // CHECK-NEXT:      fsd ft3, 32(t0)                              # store double value to memref of shape (1, 5)
 // CHECK-NEXT:      csrrci zero, 1984, 1
 // CHECK-NEXT:      ret
-
-// CHECK-OPT:       .text
-// CHECK-OPT-NEXT:  .globl main$async_dispatch_0_matmul_transpose_b_1x400x161_f64$xdsl_kernel1
-// CHECK-OPT-NEXT:  .p2align 2
-// CHECK-OPT-NEXT:  # Regalloc stats: {"preallocated_float": [], "preallocated_int": ["a0", "a1", "a2", "zero"], "allocated_float": ["ft0", "ft1", "ft2"], "allocated_int": ["a0", "a1", "a2", "a3", "a4", "a5", "t0", "t1", "t2", "t3", "t4", "t5", "t6", "zero"]}
-// CHECK-OPT-NEXT:  main$async_dispatch_0_matmul_transpose_b_1x400x161_f64$xdsl_kernel1:
-// CHECK-OPT-NEXT:      mv t4, a0
-// CHECK-OPT-NEXT:      mv t3, a1
-// CHECK-OPT-NEXT:      mv t2, a2
-// CHECK-OPT-NEXT:      li t6, 5
-// CHECK-OPT-NEXT:      li t0, 161
-// CHECK-OPT-NEXT:      mv t5, zero
-// CHECK-OPT-NEXT:      # Constant folded riscv_cf.bge
-// CHECK-OPT-NEXT:  scf_body_1_for:
-// CHECK-OPT-NEXT:      mv a3, zero
-// CHECK-OPT-NEXT:      # Constant folded riscv_cf.bge
-// CHECK-OPT-NEXT:  scf_body_0_for:
-// CHECK-OPT-NEXT:      mv a4, a3
-// CHECK-OPT-NEXT:      li a5, 8
-// CHECK-OPT-NEXT:      mul a4, a4, a5                               # multiply by element size
-// CHECK-OPT-NEXT:      add a4, t4, a4
-// CHECK-OPT-NEXT:      fld ft0, 0(a4)                               # load double from memref of shape (1, 161)
-// CHECK-OPT-NEXT:      li a4, 161
-// CHECK-OPT-NEXT:      mul a4, t5, a4
-// CHECK-OPT-NEXT:      add a4, a4, a3
-// CHECK-OPT-NEXT:      li a5, 8
-// CHECK-OPT-NEXT:      mul a4, a4, a5                               # multiply by element size
-// CHECK-OPT-NEXT:      add a4, t3, a4
-// CHECK-OPT-NEXT:      fld ft1, 0(a4)                               # load double from memref of shape (5, 161)
-// CHECK-OPT-NEXT:      mv a4, t5
-// CHECK-OPT-NEXT:      li a5, 8
-// CHECK-OPT-NEXT:      mul a4, a4, a5                               # multiply by element size
-// CHECK-OPT-NEXT:      add a4, t2, a4
-// CHECK-OPT-NEXT:      fld ft2, 0(a4)                               # load double from memref of shape (1, 5)
-// CHECK-OPT-NEXT:      fmadd.d ft0, ft0, ft1, ft2
-// CHECK-OPT-NEXT:      mv a4, t5
-// CHECK-OPT-NEXT:      li a5, 8
-// CHECK-OPT-NEXT:      mul a4, a4, a5                               # multiply by element size
-// CHECK-OPT-NEXT:      add a4, t2, a4
-// CHECK-OPT-NEXT:      fsd ft0, 0(a4)                               # store double value to memref of shape (1, 5)
-// CHECK-OPT-NEXT:      addi a3, a3, 1
-// CHECK-OPT-NEXT:      blt a3, t0, scf_body_0_for
-// CHECK-OPT-NEXT:  scf_body_end_0_for:
-// CHECK-OPT-NEXT:      addi t5, t5, 1
-// CHECK-OPT-NEXT:      blt t5, t6, scf_body_1_for
-// CHECK-OPT-NEXT:  scf_body_end_1_for:
-// CHECK-OPT-NEXT:      ret

tests/transforms/test_test_lower_linalg_to_snitch.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 
 from xdsl.backend.riscv.lowering import (
     convert_arith_to_riscv,
@@ -96,73 +96,16 @@
     memref_streamify.MemrefStreamifyPass(),
 )
 
-MAX_OPT_LEVEL = len(LINALG_SNITCH_OPTIMIZATION_PASSES)
-
-
-def get_excluded_passes(
-    optimization_level: int = MAX_OPT_LEVEL,
-) -> tuple[ModulePass, ...]:
-    """
-    This function determines which optimization passes should be excluded from the
-    lowering pipeline based on the specified optimization level. A higher optimization
-    level includes more passes.
-
-    Args:
-        optimization_level (int): The desired optimization level, ranging from 0 to
-            4 (inclusive). Defaults to 4.
-
-    Returns:
-        tuple[ModulePass, ...]: A tuple containing the ModulePass objects to be excluded
-        from the lowering pipeline.
-    """
-
-    if optimization_level == MAX_OPT_LEVEL:
-        return ()
-
-    return (
-        LINALG_SNITCH_OPTIMIZATION_PASSES[:-optimization_level]
-        if optimization_level
-        else LINALG_SNITCH_OPTIMIZATION_PASSES
-    )
-
-
-def get_passes(optimization_level: int = MAX_OPT_LEVEL) -> tuple[ModulePass, ...]:
-    """
-    This function returns a tuple of ModulePass objects to be applied in the lowering
-    pipeline, based on the specified optimization level.
-
-    Args:
-        optimization_level (int): The desired optimization level, ranging from 0 to
-            4 (inclusive). Defaults to 4.
-
-    Returns:
-        tuple[ModulePass, ...]: A tuple containing the ModulePass objects to be applied
-        in the lowering pipeline.
-    """
-
-    excluded_passes = get_excluded_passes(optimization_level)
-    return tuple(
-        p for p in TEST_LOWER_LINALG_TO_SNITCH_PASSES if p not in excluded_passes
-    )
-
 
 @dataclass(frozen=True)
 class TestLowerLinalgToSnitchPass(ModulePass):
     """
     A compiler pass used for testing lowering microkernels from linalg generic to snitch
     assembly.
-
-    Args:
-        optimization_level (int): The desired optimization level, ranging from 0 to
-            4 (inclusive). Defaults to 4.
     """
 
     name = "test-lower-linalg-to-snitch"
 
-    optimization_level: int = field(default=MAX_OPT_LEVEL)
-
     def apply(self, ctx: MLContext, op: builtin.ModuleOp) -> None:
-        passes = get_passes(self.optimization_level)
-
-        for p in passes:
+        for p in TEST_LOWER_LINALG_TO_SNITCH_PASSES:
             p.apply(ctx, op)