fix linalg fill (#66)

Author: makslevental

README.md

@@ -133,7 +133,7 @@ Practically speaking that means you need to have *some* package installed that i
 So
 
 ```shell
-$ YOUR_HOST_MLIR_PYTHON_PACKAGE_PREFIX=<YOUR_HOST_MLIR_PYTHON_PACKAGE_PREFIX> pip install git+https://github.com/makslevental/mlir-python-extras
+$ HOST_MLIR_PYTHON_PACKAGE_PREFIX=<YOUR_HOST_MLIR_PYTHON_PACKAGE_PREFIX> pip install git+https://github.com/makslevental/mlir-python-extras
 ```
 
 where `YOUR_HOST_MLIR_PYTHON_PACKAGE_PREFIX` is (as it says) the package prefix for your chosen host bindings.

mlir/extras/dialects/ext/linalg.py

@@ -1,8 +1,9 @@
+from . import arith
 from ...util import get_user_code_loc
-
+from ....dialects import linalg
 # noinspection PyUnresolvedReferences
 from ....dialects.linalg import *
-from ....dialects import linalg
+from ....extras import types as T
 
 
 def abs(I, O, *, loc=None, ip=None):
@@ -263,16 +264,25 @@ def exp(I, O, *, loc=None, ip=None):
     return linalg.exp(I, loc=loc, ip=ip, outs=[O])
 
 
-def fill(O, *, loc=None, ip=None):
+def fill(v, O, *, loc=None, ip=None):
+    if isinstance(v, (float, int, bool)):
+        v = arith.constant(v)
     if loc is None:
         loc = get_user_code_loc()
-    return linalg.fill(loc=loc, ip=ip, outs=[O])
+    return linalg.fill(v, loc=loc, ip=ip, outs=[O])
 
 
-def fill_rng_2d(O, *, loc=None, ip=None):
+def fill_rng_2d(min, max, seed, O, *, loc=None, ip=None):
+    params = [min, max]
+    for i, m in enumerate(params):
+        if isinstance(m, (float, int)):
+            params[i] = arith.constant(m, type=T.f64())
+    min, max = params
+    if isinstance(seed, int):
+        seed = arith.constant(seed, T.i32())
     if loc is None:
         loc = get_user_code_loc()
-    return linalg.fill_rng_2d(loc=loc, ip=ip, outs=[O])
+    return linalg.fill_rng_2d(min, max, seed, loc=loc, ip=ip, outs=[O])
 
 
 def floor(I, O, *, loc=None, ip=None):

tests/test_linalg.py

@@ -0,0 +1,44 @@
+from textwrap import dedent
+
+import pytest
+
+import mlir.extras.types as T
+from mlir.extras.dialects.ext import linalg, memref, tensor
+
+# noinspection PyUnresolvedReferences
+from mlir.extras.testing import MLIRContext, filecheck, mlir_ctx as ctx
+
+# needed since the fix isn't defined here nor conftest.py
+pytest.mark.usefixtures("ctx")
+
+
+def test_np_constructor(ctx: MLIRContext):
+    x = memref.alloc(10, 10, T.i32())
+    linalg.fill(5, x)
+    linalg.fill_rng_2d(0.0, 10.0, 1, x)
+
+    x = tensor.empty(10, 10, T.i32())
+    y = linalg.fill_rng_2d(0.0, 10.0, 1, x)
+    z = linalg.fill(5, x)
+
+    correct = dedent(
+        """\
+    module {
+      %alloc = memref.alloc() : memref<10x10xi32>
+      %c5_i32 = arith.constant 5 : i32
+      linalg.fill ins(%c5_i32 : i32) outs(%alloc : memref<10x10xi32>)
+      %cst = arith.constant 0.000000e+00 : f64
+      %cst_0 = arith.constant 1.000000e+01 : f64
+      %c1_i32 = arith.constant 1 : i32
+      linalg.fill_rng_2d ins(%cst, %cst_0, %c1_i32 : f64, f64, i32) outs(%alloc : memref<10x10xi32>)
+      %0 = tensor.empty() : tensor<10x10xi32>
+      %cst_1 = arith.constant 0.000000e+00 : f64
+      %cst_2 = arith.constant 1.000000e+01 : f64
+      %c1_i32_3 = arith.constant 1 : i32
+      %1 = linalg.fill_rng_2d ins(%cst_1, %cst_2, %c1_i32_3 : f64, f64, i32) outs(%0 : tensor<10x10xi32>) -> tensor<10x10xi32>
+      %c5_i32_4 = arith.constant 5 : i32
+      %2 = linalg.fill ins(%c5_i32_4 : i32) outs(%0 : tensor<10x10xi32>) -> tensor<10x10xi32>
+    }
+    """
+    )
+    filecheck(correct, ctx.module)