rename folder

BallisticLA · Jun 26, 2024 · 9278446 · 9278446
1 parent 7f0b17a
commit 9278446
Show file tree

Hide file tree

Showing 11 changed files with 66 additions and 19 deletions.
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
@@ -10,11 +10,11 @@ if (GTest_FOUND)
         test_datastructures/test_denseskop.cc
         test_datastructures/test_sparseskop.cc
 
-        test_matmul_impls/linop_common.hh
-        test_matmul_impls/test_lskge3.cc
-        test_matmul_impls/test_rskge3.cc
-        test_matmul_impls/test_lskges.cc
-        test_matmul_impls/test_rskges.cc
+        test_matmul_cores/linop_common.hh
+        test_matmul_cores/test_lskge3.cc
+        test_matmul_cores/test_rskge3.cc
+        test_matmul_cores/test_lskges.cc
+        test_matmul_cores/test_rskges.cc
 
         test_matmul_wrappers/test_sketch_vector.cc
     )
@@ -34,10 +34,10 @@ if (GTest_FOUND)
         test_datastructures/test_spmats/test_coo.cc
         test_datastructures/test_spmats/test_conversions.cc
 
-        test_matmul_impls/test_spmm/spmm_test_helpers.hh
-        test_matmul_impls/test_spmm/test_spmm_csc.cc
-        test_matmul_impls/test_spmm/test_spmm_csr.cc
-        test_matmul_impls/test_spmm/test_spmm_coo.cc
+        test_matmul_cores/test_spmm/spmm_test_helpers.hh
+        test_matmul_cores/test_spmm/test_spmm_csc.cc
+        test_matmul_cores/test_spmm/test_spmm_csr.cc
+        test_matmul_cores/test_spmm/test_spmm_coo.cc
     )
     target_link_libraries(SparseRandBLAS_tests
         RandBLAS

diff --git a/test/DevNotes.md b/test/DevNotes.md
@@ -1,2 +1,51 @@
 Developer notes for RandBLAS' testing infrastructure
 ====================================================
+
+Meta note:
+
+    Don't defend previous design decisions. Just explain how they work. 
+    That's easier and more useful. Plus, a good explanation will make
+    pros and cons of the decision (more) self-evident.
+
+
+
+
+Right-multplication by a structured linear operator in a GEMM-like API can
+always be reduced to left-multiplication by flipping transposition flags and
+layout parameters. So, why have equally fleshed-out tests(/test tooling) for
+both cases?
+
+Short answer: maybe it was a bad idea.
+
+Big picture defensive answer:
+
+    Different linear operators vary in the extent to which code for their 
+    action on the left can be reduced to their action on the right. Action
+    on the right is equivalent to adjoint-action from the left.
+
+    Someone who's adding a new linear operator might prefer to think mostly
+    in terms of right-multiplication, and just have left-multiplication reduce
+    to adjoint-action from the right. 
+
+    We want someone who adds new functionality to benefit from our testing infrastructure. So we made infrastructure to test GEMM-like APIs where
+    one operand is structured, and it's easy to get started using this 
+    infrastructure because it's equally valid to start with tests that
+    multiply only for one side and only another.
+
+Specifics:
+
+    RSKGE3 doesn't actually reduce to LSKGE3. It could, but it was
+    easy enough to have it reduce directly to GEMM, and reducing 
+    directly to GEMM had the advantage of improved readibility. We don't
+    test all possible combinations of flags (we omit when both arguments
+    are transposed) but the combinationed we leave untested are unrelated
+    to flow-of-control.
+
+    RSKGES reduces to right_spmm, which does indeed fall back on
+    left_spmm. But left_spmm has a large number codepaths (twelve!).
+    It would have been awkward to write tests that hit all of those codepaths
+    directly. Instead, we write a smaller set of tests for left_spmm
+    and right_spmm, and count on the right_spmm tests to hit complementary
+    codepaths compared to the paths hit in the left_spmm tests.
+    (Admission: we don't know for sure if all codepaths are hit. Verifying
+     that is on our TODO list.)
diff --git a/test/test_matmul_impls/linop_common.hh → test/test_matmul_cores/linop_common.hh b/test/test_matmul_impls/linop_common.hh → test/test_matmul_cores/linop_common.hh
diff --git a/test/test_matmul_impls/test_lskge3.cc → test/test_matmul_cores/test_lskge3.cc b/test/test_matmul_impls/test_lskge3.cc → test/test_matmul_cores/test_lskge3.cc
@@ -27,7 +27,7 @@
 // POSSIBILITY OF SUCH DAMAGE.
 //
 
-#include "test/test_matmul_impls/linop_common.hh"
+#include "test/test_matmul_cores/linop_common.hh"
 #include <gtest/gtest.h>
 
 using RandBLAS::DenseDist;

diff --git a/test/test_matmul_impls/test_lskges.cc → test/test_matmul_cores/test_lskges.cc b/test/test_matmul_impls/test_lskges.cc → test/test_matmul_cores/test_lskges.cc
@@ -27,7 +27,7 @@
 // POSSIBILITY OF SUCH DAMAGE.
 //
 
-#include "test/test_matmul_impls/linop_common.hh"
+#include "test/test_matmul_cores/linop_common.hh"
 #include <gtest/gtest.h>
 
 using RandBLAS::SparseDist;
@@ -185,7 +185,6 @@ TEST_F(TestLSKGES, sketch_saso_rowMajor_oneThread)
     }
 }
 
-
 TEST_F(TestLSKGES, sketch_laso_rowMajor_oneThread)
 {
     for (int64_t k_idx : {0, 1, 2}) {

diff --git a/test/test_matmul_impls/test_rskge3.cc → test/test_matmul_cores/test_rskge3.cc b/test/test_matmul_impls/test_rskge3.cc → test/test_matmul_cores/test_rskge3.cc
@@ -27,7 +27,7 @@
 // POSSIBILITY OF SUCH DAMAGE.
 //
 
-#include "test/test_matmul_impls/linop_common.hh"
+#include "test/test_matmul_cores/linop_common.hh"
 #include <gtest/gtest.h>
 
 using namespace test::linop_common;

diff --git a/test/test_matmul_impls/test_rskges.cc → test/test_matmul_cores/test_rskges.cc b/test/test_matmul_impls/test_rskges.cc → test/test_matmul_cores/test_rskges.cc
@@ -27,7 +27,7 @@
 // POSSIBILITY OF SUCH DAMAGE.
 //
 
-#include "test/test_matmul_impls/linop_common.hh"
+#include "test/test_matmul_cores/linop_common.hh"
 #include <gtest/gtest.h>
 
 using namespace test::linop_common;
@@ -193,7 +193,6 @@ TEST_F(TestRSKGES, sketch_saso_rowMajor_oneThread)
     }
 }
 
-
 TEST_F(TestRSKGES, sketch_laso_rowMajor_oneThread)
 {
     for (int64_t k_idx : {0, 1, 2}) {

diff --git a/...tmul_impls/test_spmm/spmm_test_helpers.hh → ...tmul_cores/test_spmm/spmm_test_helpers.hh b/...tmul_impls/test_spmm/spmm_test_helpers.hh → ...tmul_cores/test_spmm/spmm_test_helpers.hh
@@ -28,7 +28,7 @@
 //
 
 #include "test/test_datastructures/test_spmats/common.hh"
-#include "test/test_matmul_impls/linop_common.hh"
+#include "test/test_matmul_cores/linop_common.hh"
 #include <gtest/gtest.h>
 
 using namespace test::linop_common;

diff --git a/...t_matmul_impls/test_spmm/test_spmm_coo.cc → ...t_matmul_cores/test_spmm/test_spmm_coo.cc b/...t_matmul_impls/test_spmm/test_spmm_coo.cc → ...t_matmul_cores/test_spmm/test_spmm_coo.cc
@@ -27,7 +27,7 @@
 // POSSIBILITY OF SUCH DAMAGE.
 //
 
-#include "test/test_matmul_impls/test_spmm/spmm_test_helpers.hh"
+#include "test/test_matmul_cores/test_spmm/spmm_test_helpers.hh"
 #include <vector>
 
 using namespace RandBLAS::sparse_data;

diff --git a/...t_matmul_impls/test_spmm/test_spmm_csc.cc → ...t_matmul_cores/test_spmm/test_spmm_csc.cc b/...t_matmul_impls/test_spmm/test_spmm_csc.cc → ...t_matmul_cores/test_spmm/test_spmm_csc.cc
@@ -27,7 +27,7 @@
 // POSSIBILITY OF SUCH DAMAGE.
 //
 
-#include "test/test_matmul_impls/test_spmm/spmm_test_helpers.hh"
+#include "test/test_matmul_cores/test_spmm/spmm_test_helpers.hh"
 #include <algorithm>
 #include <vector>
 

diff --git a/...t_matmul_impls/test_spmm/test_spmm_csr.cc → ...t_matmul_cores/test_spmm/test_spmm_csr.cc b/...t_matmul_impls/test_spmm/test_spmm_csr.cc → ...t_matmul_cores/test_spmm/test_spmm_csr.cc
@@ -27,7 +27,7 @@
 // POSSIBILITY OF SUCH DAMAGE.
 //
 
-#include "test/test_matmul_impls/test_spmm/spmm_test_helpers.hh"
+#include "test/test_matmul_cores/test_spmm/spmm_test_helpers.hh"
 #include <algorithm>
 #include <vector>