https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128444
Bug ID: 128444
Summary: missed optimization for fold-mem-offsets
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: rtl-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: law at gcc dot gnu.org
Blocks: 26163
Target Milestone: ---
rv64gcbv. Also taken from x264:
typedef unsigned char uint8_t;
typedef struct
{
int i_low;
int i_range;
int i_queue;
int i_bytes_outstanding;
uint8_t *p_start;
uint8_t *p;
uint8_t *p_end;
int f8_bits_encoded;
int i_pad[2];
uint8_t state[1024];
} x264_cabac_t;
extern const uint8_t x264_cabac_range_lps[64][4];
void
f (x264_cabac_t *cb, int i_ctx, int b)
{
int i_state = cb->state[i_ctx];
int i_range_lps = x264_cabac_range_lps[i_state >> 1][(cb->i_range >> 6) - 4];
cb->i_range -= i_range_lps;
if (b != (i_state & 1))
{
cb->i_low += cb->i_range;
cb->i_range = i_range_lps;
}
}
Emits this fragment:
sraiw t1,a4,6 # 13 [c=8 l=4] ashrsi3_extend
sh2add t2,a5,t0 # 23 [c=4 l=4] *shNadd
addiw a7,t1,-4 # 15 [c=8 l=4] addsi3_extended/1
add t3,t2,a7 # 24 [c=4 l=4] *adddi3/0
lbu t4,0(t3) # 27 [c=28 l=4]
*zero_extendqidi2_internal/1
Note the addiw a7, t1, -4. Conceptually we can sink the -4 offset into the
lbu instructions. That leaves insn 15 as a sext.w a7, t1. And that extension
is redundant with the one performed at insn 13. This is probably around 40m
instructions saved in x264, so small.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=26163
[Bug 26163] [meta-bug] missed optimization in SPEC (2026, 2k17, 2k and 2k6 and
95)
