leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

add unit test for non int4 load

Browse Source
This commit is contained in:
mengchi.hmc
2021-04-23 14:33:46 +08:00
parent bb35a3ba6f
commit f4b0a33633
6 changed files with 311 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
include/cutlass/conv/kernel/default_conv2d_fprop.h
View File

@@ -615,6 +615,11 @@ struct DefaultConv2dFprop <
using WarpMmaTensorOp = typename MmaCore::MmaTensorOp;
using MmaPolicy = typename MmaCore::MmaPolicy;
static cutlass::arch::CacheOperation::Kind const CacheOpB =
((sizeof_bits<ElementA>::value * AlignmentB) == 128)
? cutlass::arch::CacheOperation::Global
: cutlass::arch::CacheOperation::Always;
// Define the Mma
using Mma = threadblock::ImplicitGemmMultistage<
ThreadblockShape,
@@ -623,7 +628,7 @@ struct DefaultConv2dFprop <
arch::CacheOperation::Always,
IteratorB,
SmemIteratorB,
arch::CacheOperation::Always,
CacheOpB,
MmaPolicy,
Stages
>;

2
include/cutlass/conv/threadblock/conv2d_dgrad_output_gradient_tile_access_iterator_analytic.h
View File

@@ -341,6 +341,8 @@ public:
// Parameters structure
//
static int const kAccessesPerVector = ThreadMap::kElementsPerAccess / AccessType::kElements;
struct Params {
Layout layout;

3
include/cutlass/conv/threadblock/conv2d_fprop_filter_tile_access_iterator_optimized.h
View File

@@ -176,6 +176,7 @@ public:
}
}
CUTLASS_PRAGMA_UNROLL
for (int v_idx = 0; v_idx < kAccessesPerVector; ++v_idx) {
clear_mask_(filter_c_ + v_idx * AccessSize >= problem_size_.C, v_idx);
}
@@ -212,7 +213,6 @@ public:
#else
if (clear) {
predicates_[index] = 0;
predicates_[index] = 0;
}
#endif
}
@@ -247,6 +247,7 @@ public:
filter_c_ += params_.filter_c_delta;
}
CUTLASS_PRAGMA_UNROLL
for (int v_idx = 0; v_idx < kAccessesPerVector; ++v_idx) {
clear_mask_(filter_c_ + v_idx * AccessSize >= problem_size_.C, v_idx);
}