diff options
author | Richard Biener <rguenther@suse.de> | 2021-02-09 11:50:23 +0100 |
---|---|---|
committer | Richard Biener <rguenther@suse.de> | 2021-02-11 12:46:34 +0100 |
commit | 22a6d99d0a0d383856440ea479b4a9edabf23961 (patch) | |
tree | 4a5ec286822ed47f8e47285beb5d744e5f260f73 /gcc | |
parent | 5ee5415af8691640b0f7a5332b78d04ba309f4f0 (diff) | |
download | gcc-22a6d99d0a0d383856440ea479b4a9edabf23961.zip gcc-22a6d99d0a0d383856440ea479b4a9edabf23961.tar.gz gcc-22a6d99d0a0d383856440ea479b4a9edabf23961.tar.bz2 |
reduce sparseset memory requirement
Currently we use HOST_WIDEST_FAST_INT for the sparseset element
type which maps to a 64bit type on 64bit hosts. That's excessive
for the only current sparseset users which are LRA and IRA and
which store register numbers in it which are unsigned int. The
following changes the sparseset element type to unsigned int.
2021-02-09 Richard Biener <rguenther@suse.de>
* sparseset.h (SPARSESET_ELT_BITS): Remove.
(SPARSESET_ELT_TYPE): Use unsigned int.
* fwprop.c: Do not include sparseset.h.
Diffstat (limited to 'gcc')
-rw-r--r-- | gcc/fwprop.c | 1 | ||||
-rw-r--r-- | gcc/sparseset.h | 5 |
2 files changed, 2 insertions, 4 deletions
diff --git a/gcc/fwprop.c b/gcc/fwprop.c index 123cc22..4b8a554 100644 --- a/gcc/fwprop.c +++ b/gcc/fwprop.c @@ -28,7 +28,6 @@ along with GCC; see the file COPYING3. If not see #include "df.h" #include "rtl-ssa.h" -#include "sparseset.h" #include "predict.h" #include "cfgrtl.h" #include "cfgcleanup.h" diff --git a/gcc/sparseset.h b/gcc/sparseset.h index c72b4fe..017884e 100644 --- a/gcc/sparseset.h +++ b/gcc/sparseset.h @@ -76,15 +76,14 @@ along with GCC; see the file COPYING3. If not see Sparse sets storage requirements are relatively large: O(U) with a larger constant than sbitmaps (if the storage requirement for an sbitmap with universe U is S, then the storage required for a sparse - set for the same universe are 2*HOST_BITS_PER_WIDEST_FAST_INT * S). + set for the same universe are 2 * sizeof (SPARSESET_ELT_TYPE) * 8 * S). Accessing the sparse vector is not very cache-friendly, but iterating over the members in the set is cache-friendly because only the dense vector is used. */ /* Data Structure used for the SparseSet representation. */ -#define SPARSESET_ELT_BITS ((unsigned) HOST_BITS_PER_WIDEST_FAST_INT) -#define SPARSESET_ELT_TYPE unsigned HOST_WIDEST_FAST_INT +#define SPARSESET_ELT_TYPE unsigned int typedef struct sparseset_def { |