Author: Rik Huijzer Date: 2023-10-25T19:01:33-07:00 New Revision: d8a5215728322d1c72bafa6c84cd54936102b174
URL: https://github.com/llvm/llvm-project/commit/d8a5215728322d1c72bafa6c84cd54936102b174 DIFF: https://github.com/llvm/llvm-project/commit/d8a5215728322d1c72bafa6c84cd54936102b174.diff LOG: [mlir][doc] Slightly clarify bufferization documentation (#70212) I was reading through the very well written Destination-Passing Style docs. Even though I know not much about compilers, I managed to understand it pretty well. A few things tripped me up though, which this PR suggests to rewrite: 1. Write `buffer(%0)` instead of buffer(`%0`). While reading, I first interpreted the text as "the buffer (%0)", whereas it should be interpreted as pseudocode for "a function that determines the buffer applied to %0". Quickly introducing it and moving the backticks around as this PR does should make this more clear. Also, I verified that MLIR does not contain any other occurences of `"buffer(<BACKTICK>"`. It does contain many occurences of `"buffer("` (without the backtick after the opening bracket), so this PR makes notation a bit more consistent. 2. Quotation marks slowed me down during reading, so I removed them. I think it's also clear without. 3. The `outs` from `linalg` was suddenly introduced. I've tried to clarify in as few words as possible that `outs` stands for `outputs` but suggestions are welcome. Added: Modified: mlir/docs/Bufferization.md Removed: ################################################################################ diff --git a/mlir/docs/Bufferization.md b/mlir/docs/Bufferization.md index d9d0751cae8c9dd..4e2d4a69e74587c 100644 --- a/mlir/docs/Bufferization.md +++ b/mlir/docs/Bufferization.md @@ -110,13 +110,13 @@ As an example, consider the following op: `%0 = tensor.insert %cst into %t[%idx] : tensor<?xf32>` `%t` is the destination in this example. When choosing a buffer for the result -`%0`, One-Shot Bufferize considers only two options: +`%0`, denoted as `buffer(%0)`, One-Shot Bufferize considers only two options: -1. buffer(`%0`) = buffer(`%t`). -2. buffer(`%0`) is a newly allocated buffer. +1. `buffer(%0) = buffer(%t)`, or +2. `buffer(%0)` is a newly allocated buffer. There may be other buffers in the same function that could potentially be used -for buffer(`%0`), but those are not considered by One-Shot Bufferize to keep the +for `buffer(%0)`, but those are not considered by One-Shot Bufferize to keep the bufferization simple. One-Shot Bufferize could be extended to consider such buffers in the future to achieve a better quality of bufferization. @@ -131,10 +131,10 @@ memory allocation. E.g.: } : tensor<?xf32> ``` -The result of `tensor.generate` does not have a "destination", so bufferization -allocates a new buffer. This could be avoided by choosing an op such as -`linalg.generic`, which can express the same computation with a destination -("out") tensor: +The result of `tensor.generate` does not have a destination operand, so +bufferization allocates a new buffer. This could be avoided by choosing an +op such as `linalg.generic`, which can express the same computation with a +destination operand, as specified behind outputs (`outs`): ```mlir #map = affine_map<(i) -> (i)> @@ -165,7 +165,7 @@ such as a subsequent read of `%s`). RaW conflicts are detected with an analysis of SSA use-def chains (details later). One-Shot Bufferize works best if there is a single SSA use-def chain, -where the result of a tensor op is the "destination" operand of the next tensor +where the result of a tensor op is the destination operand of the next tensor ops, e.g.: ```mlir @@ -263,7 +263,7 @@ must be inserted due to a RaW conflict. E.g.: } ``` -In the above example, a buffer copy of buffer(`%another_tensor`) (with `%cst` +In the above example, a buffer copy of `buffer(%another_tensor)` (with `%cst` inserted) is yielded from the "then" branch. Note: Buffer allocations that are returned from a function are not deallocated. _______________________________________________ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
