[PATCH] merge: optimization to skip evaluate_result for single strategy

anng.sw@gmail.com· May 18, 2020, 15:36 UTC · lore

From: Andrew Ng <andrew.ng@sony.com>

For a merge with a single strategy, the result of evaluate_result() is effectively not used and therefore is not needed, so avoid altogether.

On Windows, this optimization can half the time required to perform a recursive merge of a single commit with the LLVM repo.

Signed-off-by: Andrew Ng <andrew.ng@sony.com>
---
 builtin/merge.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Show changes to builtin/merge.c +1 −1

diff --git a/builtin/merge.c b/builtin/merge.c
index ca6a5dc4bf..7da707bf55 100644
--- a/builtin/merge.c
+++ b/builtin/merge.c
@@ -1656,7 +1656,7 @@ int cmd_merge(int argc, const char **argv, const char *prefix)
 				}
 				merge_was_ok = 1;
 			}
-			cnt = evaluate_result();
+			cnt = (use_strategies_nr > 1) ? evaluate_result() : 0;
 			if (best_cnt <= 0 || cnt <= best_cnt) {
 				best_strategy = use_strategies[i]->name;
 				best_cnt = cnt;

-- 
2.26.2.windows.1

Junio C Hamano· May 18, 2020, 17:23 UTC · re: anng.sw@gmail.com · lore

Re: [PATCH] merge: optimization to skip evaluate_result for single strategy

anng.sw@gmail.com writes:

> From: Andrew Ng <andrew.ng@sony.com>
>
> For a merge with a single strategy, the result of evaluate_result() is
> effectively not used and therefore is not needed, so avoid altogether.

Excellent observation.

> On Windows, this optimization can half the time required to perform a

s/half/halve/

Show 17 quoted lines

> recursive merge of a single commit with the LLVM repo.
>
> Signed-off-by: Andrew Ng <andrew.ng@sony.com>
> ---
>  builtin/merge.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/builtin/merge.c b/builtin/merge.c
> index ca6a5dc4bf..7da707bf55 100644
> --- a/builtin/merge.c
> +++ b/builtin/merge.c
> @@ -1656,7 +1656,7 @@ int cmd_merge(int argc, const char **argv, const char *prefix)
>  				}
>  				merge_was_ok = 1;
>  			}
> -			cnt = evaluate_result();
> +			cnt = (use_strategies_nr > 1) ? evaluate_result() : 0;

As best_cnt is initialized to (-1), any value would trigger the logic to nominate the current one as "best so far", and 0 is just as good as it gets ;-) And we'll finish the loop soon after doing so.

Looks good.

>  			if (best_cnt <= 0 || cnt <= best_cnt) {
>  				best_strategy = use_strategies[i]->name;
>  				best_cnt = cnt;

patchmerge: optimization to skip evaluate_result for single strategy