ShuriZma
/
xqemu
mirror of https://github.com/xqemu/xqemu.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

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

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

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20181011205206.3552-21-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This commit is contained in:
Richard Henderson 2018-10-24 07:50:20 +01:00 committed by Peter Maydell
parent e23f12b3a2
commit 2d6ac92083
1 changed files with 50 additions and 42 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

92
target/arm/translate.c
View File

@ -1611,6 +1611,25 @@ static TCGv_i32 neon_load_reg(int reg, int pass)
return tmp; return tmp;
} }
static void neon_load_element(TCGv_i32 var, int reg, int ele, TCGMemOp mop)
{
long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
switch (mop) {
case MO_UB:
tcg_gen_ld8u_i32(var, cpu_env, offset);
break;
case MO_UW:
tcg_gen_ld16u_i32(var, cpu_env, offset);
break;
case MO_UL:
tcg_gen_ld_i32(var, cpu_env, offset);
break;
default:
g_assert_not_reached();
}
}
static void neon_load_element64(TCGv_i64 var, int reg, int ele, TCGMemOp mop) static void neon_load_element64(TCGv_i64 var, int reg, int ele, TCGMemOp mop)
{ {
long offset = neon_element_offset(reg, ele, mop & MO_SIZE); long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
@ -1639,6 +1658,25 @@ static void neon_store_reg(int reg, int pass, TCGv_i32 var)
tcg_temp_free_i32(var); tcg_temp_free_i32(var);
} }
static void neon_store_element(int reg, int ele, TCGMemOp size, TCGv_i32 var)
{
long offset = neon_element_offset(reg, ele, size);
switch (size) {
case MO_8:
tcg_gen_st8_i32(var, cpu_env, offset);
break;
case MO_16:
tcg_gen_st16_i32(var, cpu_env, offset);
break;
case MO_32:
tcg_gen_st_i32(var, cpu_env, offset);
break;
default:
g_assert_not_reached();
}
}
static void neon_store_element64(int reg, int ele, TCGMemOp size, TCGv_i64 var) static void neon_store_element64(int reg, int ele, TCGMemOp size, TCGv_i64 var)
{ {
long offset = neon_element_offset(reg, ele, size); long offset = neon_element_offset(reg, ele, size);
@ -4954,9 +4992,7 @@ static int disas_neon_ls_insn(DisasContext *s, uint32_t insn)
int stride; int stride;
int size; int size;
int reg; int reg;
int pass;
int load; int load;
int shift;
int n; int n;
int vec_size; int vec_size;
int mmu_idx; int mmu_idx;
@ -5104,18 +5140,18 @@ static int disas_neon_ls_insn(DisasContext *s, uint32_t insn)
} else { } else {
/* Single element. */ /* Single element. */
int idx = (insn >> 4) & 0xf; int idx = (insn >> 4) & 0xf;
pass = (insn >> 7) & 1; int reg_idx;
switch (size) { switch (size) {
case 0: case 0:
shift = ((insn >> 5) & 3) * 8; reg_idx = (insn >> 5) & 7;
stride = 1; stride = 1;
break; break;
case 1: case 1:
shift = ((insn >> 6) & 1) * 16; reg_idx = (insn >> 6) & 3;
stride = (insn & (1 << 5)) ? 2 : 1; stride = (insn & (1 << 5)) ? 2 : 1;
break; break;
case 2: case 2:
shift = 0; reg_idx = (insn >> 7) & 1;
stride = (insn & (1 << 6)) ? 2 : 1; stride = (insn & (1 << 6)) ? 2 : 1;
break; break;
default: default:
@ -5155,52 +5191,24 @@ static int disas_neon_ls_insn(DisasContext *s, uint32_t insn)
*/ */
return 1; return 1;
} }
tmp = tcg_temp_new_i32();
addr = tcg_temp_new_i32(); addr = tcg_temp_new_i32();
load_reg_var(s, addr, rn); load_reg_var(s, addr, rn);
for (reg = 0; reg < nregs; reg++) { for (reg = 0; reg < nregs; reg++) {
if (load) { if (load) {
tmp = tcg_temp_new_i32(); gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s),
switch (size) { s->be_data | size);
case 0: neon_store_element(rd, reg_idx, size, tmp);
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(rd, pass);
tcg_gen_deposit_i32(tmp, tmp2, tmp,
shift, size ? 16 : 8);
tcg_temp_free_i32(tmp2);
}
neon_store_reg(rd, pass, tmp);
} else { /* Store */ } else { /* Store */
tmp = neon_load_reg(rd, pass); neon_load_element(tmp, rd, reg_idx, size);
if (shift) gen_aa32_st_i32(s, tmp, addr, get_mem_index(s),
tcg_gen_shri_i32(tmp, tmp, shift); s->be_data | size);
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(tmp);
} }
rd += stride; rd += stride;
tcg_gen_addi_i32(addr, addr, 1 << size); tcg_gen_addi_i32(addr, addr, 1 << size);
} }
tcg_temp_free_i32(addr); tcg_temp_free_i32(addr);
tcg_temp_free_i32(tmp);
stride = nregs * (1 << size); stride = nregs * (1 << size);
} }
} }