fix large 1d strided incorrect results when stride != batch

* tree_node_1D: correctly handle stride != batch in strided L1D_CC

* rocfft-test: add strided large 1D case where batch != dist

* bump version to 1.0.20

CHANGELOG.md

@@ -2,6 +2,11 @@
 
 Full documentation for rocFFT is available at [rocfft.readthedocs.io](https://rocfft.readthedocs.io/en/latest/).
 
+## rocFFT 1.0.20 for ROCm 5.4.1
+
+### Fixed
+- Fixed incorrect results on strided large 1D FFTs where batch size does not equal the stride.
+
 ## rocFFT 1.0.19 for ROCm 5.4.0
 
 ### Optimizations

CMakeLists.txt

@@ -98,7 +98,7 @@ include( ROCMClients )
 include( ROCMHeaderWrapper )
 
 # Using standardized versioning from rocm-cmake
-set ( VERSION_STRING "1.0.19" )
+set ( VERSION_STRING "1.0.20" )
 rocm_setup_version( VERSION ${VERSION_STRING} )
 
 # Append our library helper cmake path and the cmake path for hip (for

clients/tests/accuracy_test_adhoc.cpp

@@ -209,6 +209,8 @@ auto adhoc_tokens = {
     "1_ioffset_0_0_ooffset_0_0",
     "real_forward_len_1024_1024_1024_single_op_batch_1_istride_1048576_1024_1_R_ostride_525312_513_"
     "1_HI_idist_1073741824_odist_537919488_ioffset_0_0_ooffset_0_0",
+    "complex_forward_len_6144_single_ip_batch_34_istride_35_CI_ostride_35_CI_idist_1_odist_1_"
+    "ioffset_0_0_ooffset_0_0",
 };
 
 INSTANTIATE_TEST_SUITE_P(adhoc_token,

docs/Doxyfile

@@ -38,7 +38,7 @@ PROJECT_NAME           = "rocFFT"
 # could be handy for archiving the generated documentation or if some version
 # control system is used.
 
-PROJECT_NUMBER         = v1.0.19
+PROJECT_NUMBER         = v1.0.20
 
 # Using the PROJECT_BRIEF tag one can provide an optional one line description
 # for a project that appears at the top of each page and should give viewer a

docs/source/conf.py

@@ -64,9 +64,9 @@
 # built documents.
 #
 # The short X.Y version.
-version = u'1.0.19'
+version = u'1.0.20'
 # The full version, including alpha/beta/rc tags.
-release = u'1.0.19'
+release = u'1.0.20'
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.

library/src/tree_node_1D.cpp

@@ -476,8 +476,8 @@ void CC1DNode::AssignParams_internal()
         // faster than the actual second dimension
         std::swap(col2colPlan->length.back(), col2colPlan->batch);
         col2colPlan->outputLength = {col2colPlan->length.back(), col2colPlan->length.front()};
-        col2colPlan->inStride     = {col2colPlan->length.back() * col2colPlan->batch, 1};
-        col2colPlan->iDist        = col2colPlan->length.back();
+        col2colPlan->iDist        = col2colPlan->inStride.back();
+        col2colPlan->inStride     = {col2colPlan->inStride.back() * col2colPlan->batch, 1};
         // make output the same shape as input (even though it's going to
         // a temp buffer), so both read+write are coalesced the same
         col2colPlan->outStride                     = col2colPlan->inStride;
@@ -487,10 +487,10 @@ void CC1DNode::AssignParams_internal()
         // again, make batch the second dimension
         std::swap(row2colPlan->length.back(), row2colPlan->batch);
         row2colPlan->outputLength = {row2colPlan->length.back(), row2colPlan->length.front()};
-        row2colPlan->inStride     = {row2colPlan->length.back(), 1};
-        row2colPlan->iDist        = row2colPlan->length.front() * row2colPlan->length.back();
-        row2colPlan->outStride    = {1, row2colPlan->batch * row2colPlan->length.back()};
-        row2colPlan->oDist        = row2colPlan->length.back();
+        row2colPlan->inStride     = {inStride.front(), 1};
+        row2colPlan->iDist        = row2colPlan->length.front() * inStride.front();
+        row2colPlan->oDist        = outStride.front();
+        row2colPlan->outStride    = {1, row2colPlan->batch * outStride.front()};
     }
 }