yuzu/unicorn
1
0
Fork 0
mirror of https://github.com/yuzu-emu/unicorn.git synced 2025-03-23 06:25:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

target/arm: Reorg NEON VLD/VST single element to one lane

Browse source 
Instead of shifts and masks, use direct loads and stores from
the neon register file.

Backports commit 2d6ac920837f558be214ad2ddd28cad7f3b15e5c from qemu
This commit is contained in:
Richard Henderson 2018-11-10 11:23:34 -05:00 committed by Lioncash
parent 37103f1bc4
commit 1bcba0737e
No known key found for this signature in database
GPG key ID: 4E3C3CC1031BA9C7
1 changed files with 52 additions and 43 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

95
qemu/target/arm/translate.c
View file

@ -1682,6 +1682,26 @@ static TCGv_i32 neon_load_reg(DisasContext *s, int reg, int pass)
return tmp;
}
static void neon_load_element(DisasContext *s, TCGv_i32 var, int reg, int ele, TCGMemOp mop)
{
TCGContext *tcg_ctx = s->uc->tcg_ctx;
long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
switch (mop) {
case MO_UB:
tcg_gen_ld8u_i32(tcg_ctx, var, tcg_ctx->cpu_env, offset);
break;
case MO_UW:
tcg_gen_ld16u_i32(tcg_ctx, var, tcg_ctx->cpu_env, offset);
break;
case MO_UL:
tcg_gen_ld_i32(tcg_ctx, var, tcg_ctx->cpu_env, offset);
break;
default:
g_assert_not_reached();
}
}
static void neon_load_element64(DisasContext *s, TCGv_i64 var, int reg, int ele, TCGMemOp mop)
{
TCGContext *tcg_ctx = s->uc->tcg_ctx;
@ -1712,6 +1732,26 @@ static void neon_store_reg(DisasContext *s, int reg, int pass, TCGv_i32 var)
tcg_temp_free_i32(tcg_ctx, var);
}
static void neon_store_element(DisasContext *s, int reg, int ele, TCGMemOp size, TCGv_i32 var)
{
TCGContext *tcg_ctx = s->uc->tcg_ctx;
long offset = neon_element_offset(reg, ele, size);
switch (size) {
case MO_8:
tcg_gen_st8_i32(tcg_ctx, var, tcg_ctx->cpu_env, offset);
break;
case MO_16:
tcg_gen_st16_i32(tcg_ctx, var, tcg_ctx->cpu_env, offset);
break;
case MO_32:
tcg_gen_st_i32(tcg_ctx, var, tcg_ctx->cpu_env, offset);
break;
default:
g_assert_not_reached();
}
}
static void neon_store_element64(DisasContext *s, int reg, int ele, TCGMemOp size, TCGv_i64 var)
{
TCGContext *tcg_ctx = s->uc->tcg_ctx;
@ -5102,9 +5142,7 @@ static int disas_neon_ls_insn(DisasContext *s, uint32_t insn)
int stride;
int size;
int reg;
int pass;
int load;
int shift;
int n;
int vec_size;
int mmu_idx;
@ -5252,19 +5290,18 @@ static int disas_neon_ls_insn(DisasContext *s, uint32_t insn)
} else {
/* Single element. */
int idx = (insn >> 4) & 0xf;
pass = (insn >> 7) & 1;
int reg_idx;
switch (size) {
case 0:
shift = ((insn >> 5) & 3) * 8;
reg_idx = (insn >> 5) & 7;
stride = 1;
break;
case 1:
shift = ((insn >> 6) & 1) * 16;
reg_idx = (insn >> 6) & 3;
stride = (insn & (1 << 5)) ? 2 : 1;
break;
case 2:
shift = 0;
stride = (insn & (1 << 6)) ? 2 : 1;
reg_idx = (insn >> 7) & 1;
break;
default:
abort();
@ -5303,52 +5340,24 @@ static int disas_neon_ls_insn(DisasContext *s, uint32_t insn)
*/
return 1;
}
tmp = tcg_temp_new_i32(tcg_ctx);
addr = tcg_temp_new_i32(tcg_ctx);
load_reg_var(s, addr, rn);
for (reg = 0; reg < nregs; reg++) {
if (load) {
tmp = tcg_temp_new_i32(tcg_ctx);
switch (size) {
case 0:
gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
break;
case 1:
gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
break;
case 2:
gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
break;
default: /* Avoid compiler warnings. */
abort();
}
if (size != 2) {
tmp2 = neon_load_reg(s, rd, pass);
tcg_gen_deposit_i32(tcg_ctx, tmp, tmp2, tmp,
shift, size ? 16 : 8);
tcg_temp_free_i32(tcg_ctx, tmp2);
}
neon_store_reg(s, rd, pass, tmp);
gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s),
s->be_data | size);
neon_store_element(s, rd, reg_idx, size, tmp);
} else { /* Store */
tmp = neon_load_reg(s, rd, pass);
if (shift)
tcg_gen_shri_i32(tcg_ctx, tmp, tmp, shift);
switch (size) {
case 0:
gen_aa32_st8(s, tmp, addr, get_mem_index(s));
break;
case 1:
gen_aa32_st16(s, tmp, addr, get_mem_index(s));
break;
case 2:
gen_aa32_st32(s, tmp, addr, get_mem_index(s));
break;
}
tcg_temp_free_i32(tcg_ctx, tmp);
neon_load_element(s, tmp, rd, reg_idx, size);
gen_aa32_st_i32(s, tmp, addr, get_mem_index(s),
s->be_data | size);
}
rd += stride;
tcg_gen_addi_i32(tcg_ctx, addr, addr, 1 << size);
}
tcg_temp_free_i32(tcg_ctx, addr);
tcg_temp_free_i32(tcg_ctx, tmp);
stride = nregs * (1 << size);
}
}