Fix fortran ordering bug. Re-name some interface functions.

Author: tbenthompson

README.md

@@ -18,10 +18,10 @@ See below for usage and installation instructions.
 Howdy! Usage is really simple:
 
 ```
-import cutde.fullspace
+import cutde
 
-disp = cutde.fullspace.clu_disp(pts, tris, slips, 0.25)
-strain = cutde.fullspace.clu_strain(pts, tris, slips, nu)
+disp = cutde.disp(pts, tris, slips, 0.25)
+strain = cutde.strain(pts, tris, slips, nu)
 ```
 
 * `pts` is a `np.array` with shape `(N, 3)`
@@ -39,7 +39,7 @@ IMPORTANT: N should be the same for all these arrays. There is exactly one trian
 Use:
 
 ```
-stress = cutde.fullspace.strain_to_stress(strain, sm, nu)
+stress = cutde.strain_to_stress(strain, sm, nu)
 ```
 
 to convert from stress to strain assuming isotropic linear elasticity. `sm` is the shear modulus and `nu` is the Poisson ratio.
@@ -49,10 +49,10 @@ to convert from stress to strain assuming isotropic linear elasticity. `sm` is t
 If, instead, you want to create a matrix representing the interaction between every observation point and every source triangle, there is a different interface:
 
 ```
-import cutde.fullspace
+import cutde
 
-disp = cutde.fullspace.clu_disp_all_pairs(pts, tris, slips, 0.25)
-strain = cutde.fullspace.clu_strain_all_pairs(pts, tris, slips, nu)
+disp = cutde.disp_all_pairs(pts, tris, slips, 0.25)
+strain = cutde.strain_all_pairs(pts, tris, slips, nu)
 ```
 
 * `pts` is a `np.array` with shape `(N_OBS_PTS, 3)`

VERSION

@@ -1 +1 @@
-21.03.31.1
+21.04.01

cutde/__init__.py

@@ -1,3 +1,8 @@
-import os
-
-source_dir = os.path.dirname(os.path.realpath(__file__))
+from .fullspace import (  # noqa: F401
+    disp,
+    disp_all_pairs,
+    py_disp,
+    strain,
+    strain_all_pairs,
+    strain_to_stress,
+)

cutde/fullspace.cu

@@ -26,6 +26,10 @@ typedef struct Real6 {
     Real c;
 } Real6;
 
+WITHIN_KERNEL void print(Real x) {
+    printf("%f \n", x);
+}
+
 <%def name="binop(dim, name, op, b_scalar = False)">
 <%
 b_type = 'Real' if b_scalar else 'Real' + str(dim)
@@ -56,6 +60,14 @@ ${binop(dim, 'sub_scalar','-',b_scalar = True)}
 ${binop(dim, 'mul_scalar','*',b_scalar = True)}
 ${binop(dim, 'div_scalar','/',b_scalar = True)}
 
+WITHIN_KERNEL void print${dim}(Real${dim} x) {
+    <%
+    format_str = "%f " * dim 
+    var_str = ','.join(['x.' + comp(d) for d in range(dim)])
+    %>
+    printf("${format_str} \n", ${var_str});
+}
+
 WITHIN_KERNEL Real sum${dim}(Real${dim} x) {
     Real out = 0.0;
     % for d in range(dim):
@@ -160,10 +172,6 @@ WITHIN_KERNEL Real6 tensor_transform3(Real3 a, Real3 b, Real3 c, Real6 tensor) {
     return out;
 }
 
-WITHIN_KERNEL void print_vec(Real3 x) {
-    printf("%f, %f, %f\n", x.x, x.y, x.z);
-}
-
 WITHIN_KERNEL int trimodefinder(Real3 obs, Real3 tri0, Real3 tri1, Real3 tri2) {
     // trimodefinder calculates the normalized barycentric coordinates of
     // the points with respect to the TD vertices and specifies the appropriate

cutde/fullspace.py

@@ -1,10 +1,13 @@
+import os
+
 import numpy as np
 
 import cutde.gpu as cluda
 
-from . import source_dir
 from .TDdispFS import TDdispFS
 
+source_dir = os.path.dirname(os.path.realpath(__file__))
+
 
 def py_disp(obs_pt, tri, slip, nu):
     return TDdispFS(obs_pt, tri, slip, nu)
@@ -24,6 +27,13 @@ def call_clu(obs_pts, tris, slips, nu, fnc_name, out_dim, float_type):
     gpu_config = dict(block_size=block_size, float_type=cluda.np_to_c_type(float_type))
     module = cluda.load_gpu("fullspace.cu", tmpl_args=gpu_config, tmpl_dir=source_dir)
 
+    if obs_pts.flags.f_contiguous:
+        obs_pts = obs_pts.copy()
+    if tris.flags.f_contiguous:
+        tris = tris.copy()
+    if slips.flags.f_contiguous:
+        slips = slips.copy()
+
     gpu_results = cluda.empty_gpu(n * out_dim, float_type)
     gpu_obs_pts = cluda.to_gpu(obs_pts, float_type)
     gpu_tris = cluda.to_gpu(tris, float_type)
@@ -43,32 +53,39 @@ def call_clu(obs_pts, tris, slips, nu, fnc_name, out_dim, float_type):
     return out
 
 
-def call_clu_all_pairs(obs_pts, src_tris, slips, nu, fnc_name, out_dim, float_type):
+def call_clu_all_pairs(obs_pts, tris, slips, nu, fnc_name, out_dim, float_type):
     assert obs_pts.shape[1] == 3
-    assert src_tris.shape[1] == 3
-    assert src_tris.shape[2] == 3
-    assert slips.shape[0] == src_tris.shape[0]
+    assert tris.shape[1] == 3
+    assert tris.shape[2] == 3
+    assert slips.shape[0] == tris.shape[0]
     assert slips.shape[1] == 3
 
     n_obs = obs_pts.shape[0]
-    n_src = src_tris.shape[0]
+    n_src = tris.shape[0]
     block_size = 8
     n_obs_blocks = int(np.ceil(n_obs / block_size))
     n_src_blocks = int(np.ceil(n_src / block_size))
     gpu_config = dict(block_size=block_size, float_type=cluda.np_to_c_type(float_type))
     module = cluda.load_gpu("fullspace.cu", tmpl_args=gpu_config, tmpl_dir=source_dir)
 
+    if obs_pts.flags.f_contiguous:
+        obs_pts = obs_pts.copy()
+    if tris.flags.f_contiguous:
+        tris = tris.copy()
+    if slips.flags.f_contiguous:
+        slips = slips.copy()
+
     gpu_results = cluda.empty_gpu(n_obs * n_src * out_dim, float_type)
     gpu_obs_pts = cluda.to_gpu(obs_pts, float_type)
-    gpu_src_tris = cluda.to_gpu(src_tris, float_type)
+    gpu_tris = cluda.to_gpu(tris, float_type)
     gpu_slips = cluda.to_gpu(slips, float_type)
 
     getattr(module, fnc_name + "_all_pairs")(
         gpu_results,
         np.int32(n_obs),
         np.int32(n_src),
         gpu_obs_pts,
-        gpu_src_tris,
+        gpu_tris,
         gpu_slips,
         float_type(nu),
         grid=(n_obs_blocks, n_src_blocks, 1),
@@ -78,21 +95,21 @@ def call_clu_all_pairs(obs_pts, src_tris, slips, nu, fnc_name, out_dim, float_ty
     return out
 
 
-def clu_disp(obs_pts, tris, slips, nu, float_dtype=np.float32):
+def disp(obs_pts, tris, slips, nu, float_dtype=np.float32):
     return call_clu(obs_pts, tris, slips, nu, "disp_fullspace", 3, float_dtype)
 
 
-def clu_strain(obs_pts, tris, slips, nu, float_dtype=np.float32):
+def strain(obs_pts, tris, slips, nu, float_dtype=np.float32):
     return call_clu(obs_pts, tris, slips, nu, "strain_fullspace", 6, float_dtype)
 
 
-def clu_disp_all_pairs(obs_pts, tris, slips, nu, float_dtype=np.float32):
+def disp_all_pairs(obs_pts, tris, slips, nu, float_dtype=np.float32):
     return call_clu_all_pairs(
         obs_pts, tris, slips, nu, "disp_fullspace", 3, float_dtype
     )
 
 
-def clu_strain_all_pairs(obs_pts, tris, slips, nu, float_dtype=np.float32):
+def strain_all_pairs(obs_pts, tris, slips, nu, float_dtype=np.float32):
     return call_clu_all_pairs(
         obs_pts, tris, slips, nu, "strain_fullspace", 6, float_dtype
     )

tests/test_tde.py

@@ -3,7 +3,7 @@
 import numpy as np
 import scipy.io
 
-import cutde.fullspace
+import cutde
 
 
 def get_pt_grid():
@@ -40,7 +40,7 @@ def py_tde_tester(setup_fnc, N_test=-1):
     start = time.time()
     for i in range(N_test):
         pt = test_pts[i, :]
-        results[i, :] = cutde.fullspace.py_disp(pt, tri, slip, 0.25)
+        results[i, :] = cutde.py_disp(pt, tri, slip, 0.25)
         np.testing.assert_almost_equal(results[i, 0], correct["UEf"][i, 0])
         np.testing.assert_almost_equal(results[i, 1], correct["UNf"][i, 0])
         np.testing.assert_almost_equal(results[i, 2], correct["UVf"][i, 0])
@@ -70,16 +70,22 @@ def cluda_tde_tester(setup_fnc):
     tris = np.array([tri] * N_test)
     slips = np.array([slip] * N_test)
 
-    disp = cutde.fullspace.clu_disp(test_pts[:N_test], tris, slips, 0.25, np.float64)
-    strain = cutde.fullspace.clu_strain(test_pts[:N_test], tris, slips, nu, np.float64)
-    stress = cutde.fullspace.strain_to_stress(strain, sm, nu)
+    disp = cutde.disp(test_pts[:N_test], tris, slips, 0.25, np.float64)
+    strain = cutde.strain(test_pts[:N_test], tris, slips, nu, np.float64)
+    stress = cutde.strain_to_stress(strain, sm, nu)
 
     np.testing.assert_almost_equal(disp[:, 0], correct["UEf"][:N_test, 0])
     np.testing.assert_almost_equal(disp[:, 1], correct["UNf"][:N_test, 0])
     np.testing.assert_almost_equal(disp[:, 2], correct["UVf"][:N_test, 0])
     np.testing.assert_almost_equal(strain, correct["Strain"][:N_test])
     np.testing.assert_almost_equal(stress, correct["Stress"][:N_test])
 
+    test_ptsF = np.asfortranarray(test_pts[:N_test])
+    trisF = np.asfortranarray(tris)
+    slipsF = np.asfortranarray(slips)
+    dispF = cutde.disp(test_ptsF, trisF, slipsF, 0.25, np.float64)
+    np.testing.assert_almost_equal(disp, dispF)
+
 
 def test_cluda_simple():
     cluda_tde_tester(get_simple_test)
@@ -96,6 +102,13 @@ def test_all_pairs():
         strain1 = np.empty((n_obs, n_src, 6))
         for i in range(n_obs):
             tiled_pt = np.tile(pts[i, np.newaxis, :], (tris.shape[0], 1))
-            strain1[i] = cutde.fullspace.clu_strain(tiled_pt, tris, slips, 0.25)
-        strain2 = cutde.fullspace.clu_strain_all_pairs(pts, tris, slips, 0.25)
-    np.testing.assert_almost_equal(strain1, strain2)
+            strain1[i] = cutde.strain(tiled_pt, tris, slips, 0.25)
+        strain2 = cutde.strain_all_pairs(pts, tris, slips, 0.25)
+        strain3 = cutde.strain_all_pairs(
+            np.asfortranarray(pts),
+            np.asfortranarray(tris),
+            np.asfortranarray(slips),
+            0.25,
+        )
+        np.testing.assert_almost_equal(strain1, strain2)
+        np.testing.assert_almost_equal(strain2, strain3)