@@ -366,19 +366,19 @@ function generic_matmatmul!(C::AbstractGPUMatrix{R}, A::AbstractGPUMatrix{T}, B:
366366 # number of tiles depends on inner dimension
367367 @uniform NUM_TILES = div (Q + TILE_DIM - 1 , TILE_DIM)
368368
369- I = (grow - 1 ) * TILE_DIM + tile_row
370- J = (gcol - 1 ) * TILE_DIM + tile_col
369+ glob_I = (grow - 1 ) * TILE_DIM + tile_row
370+ glob_J = (gcol - 1 ) * TILE_DIM + tile_col
371371
372372 # loop over all tiles needed for this calculation
373373 for t in 0 : (NUM_TILES - 1 )
374374 # load inputs into tiles, with bounds checking for non-square matrices
375- if I <= N && t * TILE_DIM + tile_col <= Q
376- @inbounds tile1[tile_row, tile_col] = input1[I , t * TILE_DIM + tile_col]
375+ if glob_I <= N && t * TILE_DIM + tile_col <= Q
376+ @inbounds tile1[tile_row, tile_col] = input1[glob_I , t * TILE_DIM + tile_col]
377377 else
378378 @inbounds tile1[tile_row, tile_col] = zero (R)
379379 end
380- if J <= M && t * TILE_DIM + tile_row <= Q
381- @inbounds tile2[tile_row, tile_col] = input2[t * TILE_DIM + tile_row, J ]
380+ if glob_J <= M && t * TILE_DIM + tile_row <= Q
381+ @inbounds tile2[tile_row, tile_col] = input2[t * TILE_DIM + tile_row, glob_J ]
382382 else
383383 @inbounds tile2[tile_row, tile_col] = zero (R)
384384 end
@@ -397,8 +397,8 @@ function generic_matmatmul!(C::AbstractGPUMatrix{R}, A::AbstractGPUMatrix{T}, B:
397397 end
398398
399399 # save if inbounds
400- if I <= N && J <= M
401- @inbounds output[I, J ] = add (outval[1 ], output[I, J ])
400+ if glob_I <= N && glob_J <= M
401+ @inbounds output[glob_I, glob_J ] = add (outval[1 ], output[glob_I, glob_J ])
402402 end
403403 end
404404
0 commit comments