Add fast forward requested char simd support for s390x.

This commit is contained in:
Zoltán Herczeg 2021-01-10 14:51:53 +00:00
parent b730793117
commit e5e1fab2db
1 changed files with 165 additions and 0 deletions

View File

@ -1447,6 +1447,171 @@ if (common->utf && offset > 0)
#endif
}
#define JIT_HAS_FAST_REQUESTED_CHAR_SIMD 1
static jump_list *fast_requested_char_simd(compiler_common *common, PCRE2_UCHAR char1, PCRE2_UCHAR char2)
{
DEFINE_COMPILER;
sljit_u16 instruction[3];
struct sljit_label *start;
struct sljit_jump *quit;
jump_list *not_found = NULL;
vector_compare_type compare_type = vector_compare_match1;
sljit_s32 tmp1_reg_ind = sljit_get_register_index(TMP1);
sljit_s32 tmp3_reg_ind = sljit_get_register_index(TMP3);
sljit_s32 data_ind = 0;
sljit_s32 tmp_ind = 1;
sljit_s32 cmp1_ind = 2;
sljit_s32 cmp2_ind = 3;
sljit_s32 zero_ind = 4;
sljit_u32 bit = 0;
int i;
if (char1 != char2)
{
bit = char1 ^ char2;
compare_type = vector_compare_match1i;
if (!is_powerof2(bit))
{
bit = 0;
compare_type = vector_compare_match2;
}
}
add_jump(compiler, &not_found, CMP(SLJIT_GREATER_EQUAL, TMP1, 0, STR_END, 0));
/* First part (unaligned start) */
OP2(SLJIT_ADD, TMP2, 0, TMP1, 0, SLJIT_IMM, 16);
#if PCRE2_CODE_UNIT_WIDTH != 32
/* VREPI */
instruction[0] = (sljit_u16)(0xe700 | (cmp1_ind << 4));
instruction[1] = (sljit_u16)(char1 | bit);
instruction[2] = (sljit_u16)((VECTOR_ELEMENT_SIZE << 12) | (0x8 << 8) | 0x45);
sljit_emit_op_custom(compiler, instruction, 6);
if (char1 != char2)
{
/* VREPI */
instruction[0] = (sljit_u16)(0xe700 | (cmp2_ind << 4));
instruction[1] = (sljit_u16)(bit != 0 ? bit : char2);
/* instruction[2] = (sljit_u16)((VECTOR_ELEMENT_SIZE << 12) | (0x8 << 8) | 0x45); */
sljit_emit_op_custom(compiler, instruction, 6);
}
#else /* PCRE2_CODE_UNIT_WIDTH == 32 */
for (int i = 0; i < 2; i++)
{
replicate_imm_vector(compiler, i, cmp1_ind, char1 | bit, TMP3);
if (char1 != char2)
replicate_imm_vector(compiler, i, cmp2_ind, bit != 0 ? bit : char2, TMP3);
}
#endif /* PCRE2_CODE_UNIT_WIDTH != 32 */
if (compare_type == vector_compare_match2)
{
/* VREPI */
instruction[0] = (sljit_u16)(0xe700 | (zero_ind << 4));
instruction[1] = 0;
instruction[2] = (sljit_u16)((0x8 << 8) | 0x45);
sljit_emit_op_custom(compiler, instruction, 6);
}
load_from_mem_vector(compiler, TRUE, data_ind, tmp1_reg_ind, 0);
OP2(SLJIT_AND, TMP2, 0, TMP2, 0, SLJIT_IMM, ~15);
if (compare_type != vector_compare_match2)
{
if (compare_type == vector_compare_match1i)
fast_forward_char_pair_sse2_compare(compiler, compare_type, 0, data_ind, cmp1_ind, cmp2_ind, tmp_ind);
/* VFEE */
instruction[0] = (sljit_u16)(0xe700 | (data_ind << 4) | data_ind);
instruction[1] = (sljit_u16)((cmp1_ind << 12) | (1 << 4));
instruction[2] = (sljit_u16)((VECTOR_ELEMENT_SIZE << 12) | (0xe << 8) | 0x80);
sljit_emit_op_custom(compiler, instruction, 6);
}
else
{
for (i = 0; i < 3; i++)
fast_forward_char_pair_sse2_compare(compiler, compare_type, i, data_ind, cmp1_ind, cmp2_ind, tmp_ind);
/* VFENE */
instruction[0] = (sljit_u16)(0xe700 | (data_ind << 4) | data_ind);
instruction[1] = (sljit_u16)((zero_ind << 12) | (1 << 4));
instruction[2] = (sljit_u16)((0xe << 8) | 0x81);
sljit_emit_op_custom(compiler, instruction, 6);
}
/* TODO: use sljit_set_current_flags */
/* VLGVB */
instruction[0] = (sljit_u16)(0xe700 | (tmp3_reg_ind << 4) | data_ind);
instruction[1] = 7;
instruction[2] = (sljit_u16)((0x4 << 8) | 0x21);
sljit_emit_op_custom(compiler, instruction, 6);
OP2(SLJIT_ADD, TMP1, 0, TMP1, 0, TMP3, 0);
quit = CMP(SLJIT_LESS, TMP1, 0, TMP2, 0);
OP2(SLJIT_SUB, TMP1, 0, TMP2, 0, SLJIT_IMM, 16);
/* Second part (aligned) */
start = LABEL();
OP2(SLJIT_ADD, TMP1, 0, TMP1, 0, SLJIT_IMM, 16);
add_jump(compiler, &not_found, CMP(SLJIT_GREATER_EQUAL, TMP1, 0, STR_END, 0));
load_from_mem_vector(compiler, TRUE, data_ind, tmp1_reg_ind, 0);
if (compare_type != vector_compare_match2)
{
if (compare_type == vector_compare_match1i)
fast_forward_char_pair_sse2_compare(compiler, compare_type, 0, data_ind, cmp1_ind, cmp2_ind, tmp_ind);
/* VFEE */
instruction[0] = (sljit_u16)(0xe700 | (data_ind << 4) | data_ind);
instruction[1] = (sljit_u16)((cmp1_ind << 12) | (1 << 4));
instruction[2] = (sljit_u16)((VECTOR_ELEMENT_SIZE << 12) | (0xe << 8) | 0x80);
sljit_emit_op_custom(compiler, instruction, 6);
}
else
{
for (i = 0; i < 3; i++)
fast_forward_char_pair_sse2_compare(compiler, compare_type, i, data_ind, cmp1_ind, cmp2_ind, tmp_ind);
/* VFENE */
instruction[0] = (sljit_u16)(0xe700 | (data_ind << 4) | data_ind);
instruction[1] = (sljit_u16)((zero_ind << 12) | (1 << 4));
instruction[2] = (sljit_u16)((0xe << 8) | 0x81);
sljit_emit_op_custom(compiler, instruction, 6);
}
/* TODO: use sljit_set_current_flags */
/* VLGVB */
instruction[0] = (sljit_u16)(0xe700 | (tmp3_reg_ind << 4) | data_ind);
instruction[1] = 7;
instruction[2] = (sljit_u16)((0x4 << 8) | 0x21);
sljit_emit_op_custom(compiler, instruction, 6);
CMPTO(SLJIT_GREATER_EQUAL, TMP3, 0, SLJIT_IMM, 16, start);
OP2(SLJIT_ADD, TMP1, 0, TMP1, 0, TMP3, 0);
JUMPHERE(quit);
add_jump(compiler, &not_found, CMP(SLJIT_GREATER_EQUAL, TMP1, 0, STR_END, 0));
return not_found;
}
#define JIT_HAS_FAST_FORWARD_CHAR_PAIR_SIMD 1
static void fast_forward_char_pair_simd(compiler_common *common, sljit_s32 offs1,