yuzu/unicorn
1
0
Fork 0
mirror of https://github.com/yuzu-emu/unicorn.git synced 2025-06-20 17:08:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

target/arm: Use clear_vec_high more effectively

Browse source 
Do not explicitly store zero to the NEON high part
when we can pass !is_q to clear_vec_high.

Backports commit e1f778596ebfa8782276f4dd4651f2b285d734ff from qemu
This commit is contained in:
Richard Henderson 2020-06-14 21:06:38 -04:00 committed by Lioncash
parent 3ac9b9b206
commit acdd5c6065
1 changed files with 31 additions and 20 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

51
qemu/target/arm/translate-a64.c
View file

@ -1076,11 +1076,10 @@ static void do_fp_ld(DisasContext *s, int destidx, TCGv_i64 tcg_addr, int size)
TCGContext *tcg_ctx = s->uc->tcg_ctx; TCGContext *tcg_ctx = s->uc->tcg_ctx;
/* This always zero-extends and writes to a full 128 bit wide vector */ /* This always zero-extends and writes to a full 128 bit wide vector */
TCGv_i64 tmplo = tcg_temp_new_i64(tcg_ctx); TCGv_i64 tmplo = tcg_temp_new_i64(tcg_ctx);
TCGv_i64 tmphi; TCGv_i64 tmphi = NULL;
if (size < 4) { if (size < 4) {
MemOp memop = s->be_data + size; MemOp memop = s->be_data + size;
tmphi = tcg_const_i64(tcg_ctx, 0);
tcg_gen_qemu_ld_i64(s->uc, tmplo, tcg_addr, get_mem_index(s), memop); tcg_gen_qemu_ld_i64(s->uc, tmplo, tcg_addr, get_mem_index(s), memop);
} else { } else {
bool be = s->be_data == MO_BE; bool be = s->be_data == MO_BE;
@ -1098,12 +1097,14 @@ static void do_fp_ld(DisasContext *s, int destidx, TCGv_i64 tcg_addr, int size)
} }
tcg_gen_st_i64(tcg_ctx, tmplo, tcg_ctx->cpu_env, fp_reg_offset(s, destidx, MO_64)); tcg_gen_st_i64(tcg_ctx, tmplo, tcg_ctx->cpu_env, fp_reg_offset(s, destidx, MO_64));
tcg_gen_st_i64(tcg_ctx, tmphi, tcg_ctx->cpu_env, fp_reg_hi_offset(s, destidx));
tcg_temp_free_i64(tcg_ctx, tmplo); tcg_temp_free_i64(tcg_ctx, tmplo);
tcg_temp_free_i64(tcg_ctx, tmphi);
clear_vec_high(s, true, destidx); if (tmphi) {
tcg_gen_st_i64(tcg_ctx, tmphi, tcg_ctx->cpu_env, fp_reg_hi_offset(s, destidx));
tcg_temp_free_i64(tcg_ctx, tmphi);
}
clear_vec_high(s, tmphi != NULL, destidx);
} }
/* /*
@ -7182,7 +7183,6 @@ static void disas_simd_ext(DisasContext *s, uint32_t insn)
read_vec_element(s, tcg_resh, rm, 0, MO_64); read_vec_element(s, tcg_resh, rm, 0, MO_64);
do_ext64(s, tcg_resh, tcg_resl, pos); do_ext64(s, tcg_resh, tcg_resl, pos);
} }
tcg_gen_movi_i64(tcg_ctx, tcg_resh, 0);
} else { } else {
TCGv_i64 tcg_hh; TCGv_i64 tcg_hh;
typedef struct { typedef struct {
@ -7212,9 +7212,11 @@ static void disas_simd_ext(DisasContext *s, uint32_t insn)
write_vec_element(s, tcg_resl, rd, 0, MO_64); write_vec_element(s, tcg_resl, rd, 0, MO_64);
tcg_temp_free_i64(tcg_ctx, tcg_resl); tcg_temp_free_i64(tcg_ctx, tcg_resl);
write_vec_element(s, tcg_resh, rd, 1, MO_64); if (is_q) {
write_vec_element(s, tcg_resh, rd, 1, MO_64);
}
tcg_temp_free_i64(tcg_ctx, tcg_resh); tcg_temp_free_i64(tcg_ctx, tcg_resh);
clear_vec_high(s, true, rd); clear_vec_high(s, is_q, rd);
} }
/* TBL/TBX /* TBL/TBX
@ -7252,17 +7254,21 @@ static void disas_simd_tb(DisasContext *s, uint32_t insn)
* the input. * the input.
*/ */
tcg_resl = tcg_temp_new_i64(tcg_ctx); tcg_resl = tcg_temp_new_i64(tcg_ctx);
tcg_resh = tcg_temp_new_i64(tcg_ctx); tcg_resh = NULL;
if (is_tblx) { if (is_tblx) {
read_vec_element(s, tcg_resl, rd, 0, MO_64); read_vec_element(s, tcg_resl, rd, 0, MO_64);
} else { } else {
tcg_gen_movi_i64(tcg_ctx, tcg_resl, 0); tcg_gen_movi_i64(tcg_ctx, tcg_resl, 0);
} }
if (is_tblx && is_q) {
read_vec_element(s, tcg_resh, rd, 1, MO_64); if (is_q) {
} else { tcg_resh = tcg_temp_new_i64(tcg_ctx);
tcg_gen_movi_i64(tcg_ctx, tcg_resh, 0); if (is_tblx) {
read_vec_element(s, tcg_resh, rd, 1, MO_64);
} else {
tcg_gen_movi_i64(tcg_ctx, tcg_resh, 0);
}
} }
tcg_idx = tcg_temp_new_i64(tcg_ctx); tcg_idx = tcg_temp_new_i64(tcg_ctx);
@ -7282,9 +7288,12 @@ static void disas_simd_tb(DisasContext *s, uint32_t insn)
write_vec_element(s, tcg_resl, rd, 0, MO_64); write_vec_element(s, tcg_resl, rd, 0, MO_64);
tcg_temp_free_i64(tcg_ctx, tcg_resl); tcg_temp_free_i64(tcg_ctx, tcg_resl);
write_vec_element(s, tcg_resh, rd, 1, MO_64);
tcg_temp_free_i64(tcg_ctx, tcg_resh); if (is_q) {
clear_vec_high(s, true, rd); write_vec_element(s, tcg_resh, rd, 1, MO_64);
tcg_temp_free_i64(tcg_ctx, tcg_resh);
}
clear_vec_high(s, is_q, rd);
} }
/* ZIP/UZP/TRN /* ZIP/UZP/TRN
@ -7322,7 +7331,7 @@ static void disas_simd_zip_trn(DisasContext *s, uint32_t insn)
} }
tcg_resl = tcg_const_i64(tcg_ctx, 0); tcg_resl = tcg_const_i64(tcg_ctx, 0);
tcg_resh = tcg_const_i64(tcg_ctx, 0); tcg_resh = is_q ? tcg_const_i64(tcg_ctx, 0) : NULL;
tcg_res = tcg_temp_new_i64(tcg_ctx); tcg_res = tcg_temp_new_i64(tcg_ctx);
for (i = 0; i < elements; i++) { for (i = 0; i < elements; i++) {
@ -7373,9 +7382,11 @@ static void disas_simd_zip_trn(DisasContext *s, uint32_t insn)
write_vec_element(s, tcg_resl, rd, 0, MO_64); write_vec_element(s, tcg_resl, rd, 0, MO_64);
tcg_temp_free_i64(tcg_ctx, tcg_resl); tcg_temp_free_i64(tcg_ctx, tcg_resl);
write_vec_element(s, tcg_resh, rd, 1, MO_64); if (is_q) {
tcg_temp_free_i64(tcg_ctx, tcg_resh); write_vec_element(s, tcg_resh, rd, 1, MO_64);
clear_vec_high(s, true, rd); tcg_temp_free_i64(tcg_ctx, tcg_resh);
}
clear_vec_high(s, is_q, rd);
} }
/* /*