gh-mirror/mikepaul-LuaJIT
1
0
Fork 0
mirror of https://github.com/LuaJIT/LuaJIT.git synced 2025-02-07 15:14:08 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

ARM64: Improve K13 constant rematerialization.

Browse Source 
Algorithm by Dougall Johnson: https://dougallj.wordpress.com/2021/10/30/
Thanks to Peter Cawley. #1065
This commit is contained in:
Mike Pall 2023-09-09 14:11:25 +02:00
parent 9bd2404137
commit 9daf9f9003
1 changed files with 12 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
src/lj_emit_arm64.h
View File

@ -43,26 +43,18 @@ static uint32_t emit_isk12(int64_t n)
/* Encode constant in K13 format for logical data processing instructions. */ /* Encode constant in K13 format for logical data processing instructions. */
static uint32_t emit_isk13(uint64_t n, int is64) static uint32_t emit_isk13(uint64_t n, int is64)
{ {
int inv = 0, w = 128, lz, tz; /* Thanks to: https://dougallj.wordpress.com/2021/10/30/ */
if (n & 1) { n = ~n; w = 64; inv = 1; } /* Avoid wrap-around of ones. */ int rot, ones, size, immr, imms;
if (!n) return 0; /* Neither all-zero nor all-ones are allowed. */ if (!is64) n = ((uint64_t)n << 32) | (uint32_t)n;
do { /* Find the repeat width. */ if ((n+1u) <= 1u) return 0; /* Neither all-zero nor all-ones are allowed. */
if (is64 && (uint32_t)(n^(n>>32))) break; rot = (n & (n+1u)) ? emit_ctz64(n & (n+1u)) : 64;
n = (uint32_t)n; n = lj_ror(n, rot & 63);
if (!n) return 0; /* Ditto when passing n=0xffffffff and is64=0. */ ones = emit_ctz64(~n);
w = 32; if ((n^(n>>16)) & 0xffff) break; size = emit_clz64(n) + ones;
n = n & 0xffff; w = 16; if ((n^(n>>8)) & 0xff) break; if (lj_ror(n, size & 63) != n) return 0; /* Non-repeating? */
n = n & 0xff; w = 8; if ((n^(n>>4)) & 0xf) break; immr = -rot & (size - 1);
n = n & 0xf; w = 4; if ((n^(n>>2)) & 0x3) break; imms = (-(size << 1) | (ones - 1)) & 63;
n = n & 0x3; w = 2; return A64I_K13 | A64F_IMMR(immr | (size & 64)) | A64F_IMMS(imms);
} while (0);
lz = emit_clz64(n);
tz = emit_ctz64(n);
if ((int64_t)(n << lz) >> (lz+tz) != -1ll) return 0; /* Non-contiguous? */
if (inv)
return A64I_K13 | (((lz-w) & 127) << 16) | (((lz+tz-w-1) & 63) << 10);
else
return A64I_K13 | ((w-tz) << 16) | (((63-lz-tz-w-w) & 63) << 10);
} }
static uint32_t emit_isfpk64(uint64_t n) static uint32_t emit_isfpk64(uint64_t n)