leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 26 Packages Projects Releases Wiki Activity

Add grouped b2b GEMM (#970)

Browse Source
This commit is contained in:
Jack Kosaian
2023-06-05 17:16:57 -04:00
committed by GitHub
parent fde824af21
commit 87349d3496
15 changed files with 1644 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cutlass/gemm/threadblock/threadblock_swizzle.h
+6 -6
View File
@@ -114,7 +114,7 @@ struct GemmIdentityThreadblockSwizzle {
/// Calculates optimal swizzle width
CUTLASS_HOST_DEVICE
int get_log_tile(GemmCoord tiled_shape) const {
static int get_log_tile(GemmCoord tiled_shape) {
auto n = tiled_shape.n();
// Thresholds picked so that it doesn't cause too many no-op CTAs
if (N >= 8 && n >= 6)
@@ -187,7 +187,7 @@ struct GemmHorizontalThreadblockSwizzle {
/// Calculates optimal swizzle width
CUTLASS_HOST_DEVICE
int get_log_tile(GemmCoord tiled_shape) const {
static int get_log_tile(GemmCoord tiled_shape) {
return 0;
}
@@ -228,7 +228,7 @@ struct GemmBatchedIdentityThreadblockSwizzle {
/// Calculates optimal swizzle width
CUTLASS_HOST_DEVICE
int get_log_tile(GemmCoord tiled_shape) const {
static int get_log_tile(GemmCoord tiled_shape) {
return 0;
}
@@ -284,7 +284,7 @@ struct GemmSplitKIdentityThreadblockSwizzle {
/// Calculates optimal swizzle width
CUTLASS_HOST_DEVICE
int get_log_tile(GemmCoord tiled_shape) const {
static int get_log_tile(GemmCoord tiled_shape) {
auto n = tiled_shape.n();
// Thresholds picked so that it doesn't cause too many no-op CTAs
if (N >= 8 && n >= 6)
@@ -361,7 +361,7 @@ struct GemmSplitKHorizontalThreadblockSwizzle {
/// Calculates optimal swizzle width
CUTLASS_HOST_DEVICE
int get_log_tile(GemmCoord tiled_shape) const {
static int get_log_tile(GemmCoord tiled_shape) {
return 0;
}
@@ -412,7 +412,7 @@ struct GemvBatchedStridedThreadblockDefaultSwizzle {
/// Calculates optimal swizzle width
CUTLASS_HOST_DEVICE
int get_log_tile(GemmCoord tiled_shape) const {
static int get_log_tile(GemmCoord tiled_shape) {
return 0;
}