Merge pull request #143 from YichengDWu/docs

Update docs, fix bugs
YichengDWu · May 1, 2024 · 0e5a43b · 0e5a43b
2 parents 4caf3a2 + 390f789
commit 0e5a43b
Show file tree

Hide file tree

Showing 16 changed files with 12,728 additions and 7,161 deletions.
diff --git a/docs/src/assets/TF32F32.svg b/docs/src/assets/TF32F32.svg
diff --git a/docs/src/assets/smem_copy_A.svg b/docs/src/assets/smem_copy_A.svg
diff --git a/docs/src/assets/smem_copy_B.svg b/docs/src/assets/smem_copy_B.svg
diff --git a/docs/src/manual/async.md b/docs/src/manual/async.md
@@ -58,7 +58,7 @@ function matmul_kernel(A, sA_layout, copy_A,
 
  cp_async_wait()
 
- @gc_preserve gemm!(mma_C, tCsA, tCsB, tCrC)
+ @gc_preserve gemm!(mma_C, tCrC, tCsA, tCsB, tCrC)
  sync_threads()
  end
 

diff --git a/docs/src/manual/matmul.md b/docs/src/manual/matmul.md
@@ -76,7 +76,7 @@ end
 ```
 You can also call [`gemm!`] to perform the same operation:
 ```julia
-gemm!(tCsA, tCsB, tCrC)
+gemm!(tCrC, tCsA, tCsB, tCrC)
 ```
 
 The complete kernel code is as follows:
@@ -121,7 +121,7 @@ function matmul_kernel(A, sA_layout, tA,
  cp_async_wait()
  sync_threads()
 
- @gc_preserve gemm!(tCsA, tCsB, tCrC)
+ @gc_preserve gemm!(tCrC, tCsA, tCsB, tCrC)
  sync_threads()
  end
 

diff --git a/docs/src/manual/pipeline.md b/docs/src/manual/pipeline.md
@@ -67,7 +67,7 @@ function matmul_kernel(A, sA_layout, copy_A,
  copyto!(copy_B, tBsB, view(tBgB, :, :, :, k+1))
  end
 
- @gc_preserve gemm!(mma_C, tCrA, tCrB, tCrC)
+ @gc_preserve gemm!(mma_C, tCrC, tCrA, tCrB, tCrC)
  end
 
  copyto!(tCgC, tCrC)
@@ -161,7 +161,7 @@ for the next tile. We prefetch the next tile from global memory to shared memory
  smem_read, smem_write = smem_write, smem_read
  end
 
- @gc_preserve gemm!(mma_C, tCrA[:, :, k_block], tCrB[:, :, k_block], tCrC)
+ @gc_preserve gemm!(mma_C, tCrC, tCrA[:, :, k_block], tCrB[:, :, k_block], tCrC)
  end
  end