Fix MAGMA svd and eig (#9082)

ssnl · facebook-github-bot · commit 179807a8c7c2 · 2018-07-01T22:24:17.000-07:00
Summary: Fixes #9079 There is room for speed-up for both functions (see #9083), but let's get this in to unblock #9052 . Closes #9082 Reviewed By: ezyang Differential Revision: D8711687 Pulled By: SsnL fbshipit-source-id: f043a9bf55cb6aec5126c3331d35761f7aa3f8e3
diff --git a/aten/src/THC/generic/THCTensorMathMagma.cu b/aten/src/THC/generic/THCTensorMathMagma.cu
@@ -30,7 +30,7 @@ static void THCTensor_(copyTensor2d)(THCState *state, real *dst, THCTensor *self
   size_t len = THCTensor_(nElement)(state, self)*sizeof(real);
   THCTensor *temp = THCTensor_(newTranspose)(state, self, 0, 1);
   THCTensor *selfc = THCTensor_(newContiguous)(state, temp);
-  THCudaCheck(cudaMemcpy(dst, THCStorage_(data)(state, self->storage) + selfc->storageOffset, len, cudaMemcpyDeviceToHost));
+  THCudaCheck(cudaMemcpy(dst, THCStorage_(data)(state, selfc->storage) + selfc->storageOffset, len, cudaMemcpyDeviceToHost));
   THCTensor_(free)(state, temp);
   THCTensor_(free)(state, selfc);
 }
diff --git a/test/test_cuda.py b/test/test_cuda.py
@@ -435,6 +435,23 @@ def tmp(t):
     ('inverse', new_t(20, 20), lambda t: [], None, float_types, False),
     ('geqrf', new_t(20, 20), lambda t: [], None, float_types, False,
         unittest.skipIf(not TEST_MAGMA, "no MAGMA library detected")),
+    ('svd', new_t(10, 10), lambda t: [], 'square', float_types_no_half, False,
+        unittest.skipIf(not TEST_MAGMA, "no MAGMA library detected")),
+    ('svd', lambda t: new_t(10, 10)(t).t(), lambda t: [True], 'square_col_maj',
+        float_types_no_half, False,
+        unittest.skipIf(not TEST_MAGMA, "no MAGMA library detected")),
+    ('svd', new_t(20, 5), lambda t: [True], 'tall_some', float_types_no_half, False,
+        unittest.skipIf(not TEST_MAGMA, "no MAGMA library detected")),
+    ('svd', new_t(20, 5), lambda t: [False], 'tall_all', float_types_no_half, False,
+        unittest.skipIf(not TEST_MAGMA, "no MAGMA library detected")),
+    ('svd', lambda t: new_t(5, 20)(t).t(), lambda t: [True],
+        'tall_some_col_maj', float_types_no_half, False,
+        unittest.skipIf(not TEST_MAGMA, "no MAGMA library detected")),
+    ('svd', lambda t: new_t(5, 20)(t).t(), lambda t: [False],
+        'tall_all_col_maj', float_types_no_half, False,
+        unittest.skipIf(not TEST_MAGMA, "no MAGMA library detected")),
+    ('eig', new_t(10, 10), lambda t: [True], 'with_eigvec', float_types_no_half, False,
+        unittest.skipIf(not TEST_MAGMA, "no MAGMA library detected")),
 ]
 
 # TODO: random functions, cat, gather, scatter, index*, masked*,

Original file line number	Diff line number	Diff line change
`@@ -30,7 +30,7 @@ static void THCTensor_(copyTensor2d)(THCState state, real dst, THCTensor *self`
`30`	`30`	`size_t len = THCTensor_(nElement)(state, self)*sizeof(real);`
`31`	`31`	`THCTensor *temp = THCTensor_(newTranspose)(state, self, 0, 1);`
`32`	`32`	`THCTensor *selfc = THCTensor_(newContiguous)(state, temp);`
`33`		`- THCudaCheck(cudaMemcpy(dst, THCStorage_(data)(state, self->storage) + selfc->storageOffset, len, cudaMemcpyDeviceToHost));`
	`33`	`+ THCudaCheck(cudaMemcpy(dst, THCStorage_(data)(state, selfc->storage) + selfc->storageOffset, len, cudaMemcpyDeviceToHost));`
`34`	`34`	`THCTensor_(free)(state, temp);`
`35`	`35`	`THCTensor_(free)(state, selfc);`
`36`	`36`	`}`