tcg-arm: Use ldrd/strd for appropriate qemu_ld/st64

Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
This commit is contained in:
Richard Henderson 2013-07-27 08:42:51 -10:00
parent a684f3cf9b
commit 23bbc25085

View File

@ -320,6 +320,9 @@ typedef enum {
INSN_STRB_REG = 0x06400000, INSN_STRB_REG = 0x06400000,
INSN_LDRD_IMM = 0x004000d0, INSN_LDRD_IMM = 0x004000d0,
INSN_LDRD_REG = 0x000000d0,
INSN_STRD_IMM = 0x004000f0,
INSN_STRD_REG = 0x000000f0,
} ARMInsn; } ARMInsn;
#define SHIFT_IMM_LSL(im) (((im) << 7) | 0x00) #define SHIFT_IMM_LSL(im) (((im) << 7) | 0x00)
@ -810,6 +813,30 @@ static inline void tcg_out_st32_r(TCGContext *s, int cond, TCGReg rt,
tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 0); tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 0);
} }
static inline void tcg_out_ldrd_8(TCGContext *s, int cond, TCGReg rt,
TCGReg rn, int imm8)
{
tcg_out_memop_8(s, cond, INSN_LDRD_IMM, rt, rn, imm8, 1, 0);
}
static inline void tcg_out_ldrd_r(TCGContext *s, int cond, TCGReg rt,
TCGReg rn, TCGReg rm)
{
tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 0);
}
static inline void tcg_out_strd_8(TCGContext *s, int cond, TCGReg rt,
TCGReg rn, int imm8)
{
tcg_out_memop_8(s, cond, INSN_STRD_IMM, rt, rn, imm8, 1, 0);
}
static inline void tcg_out_strd_r(TCGContext *s, int cond, TCGReg rt,
TCGReg rn, TCGReg rm)
{
tcg_out_memop_r(s, cond, INSN_STRD_REG, rt, rn, rm, 1, 1, 0);
}
/* Register pre-increment with base writeback. */ /* Register pre-increment with base writeback. */
static inline void tcg_out_ld32_rwb(TCGContext *s, int cond, TCGReg rt, static inline void tcg_out_ld32_rwb(TCGContext *s, int cond, TCGReg rt,
TCGReg rn, TCGReg rm) TCGReg rn, TCGReg rm)
@ -1397,6 +1424,9 @@ static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, int opc)
tcg_out_ld32_12(s, COND_AL, data_reg, TCG_REG_R1, 4); tcg_out_ld32_12(s, COND_AL, data_reg, TCG_REG_R1, 4);
tcg_out_bswap32(s, COND_AL, data_reg2, data_reg2); tcg_out_bswap32(s, COND_AL, data_reg2, data_reg2);
tcg_out_bswap32(s, COND_AL, data_reg, data_reg); tcg_out_bswap32(s, COND_AL, data_reg, data_reg);
} else if (use_armv6_instructions
&& (data_reg & 1) == 0 && data_reg2 == data_reg + 1) {
tcg_out_ldrd_r(s, COND_AL, data_reg, addr_reg, TCG_REG_R1);
} else { } else {
tcg_out_ld32_rwb(s, COND_AL, data_reg, TCG_REG_R1, addr_reg); tcg_out_ld32_rwb(s, COND_AL, data_reg, TCG_REG_R1, addr_reg);
tcg_out_ld32_12(s, COND_AL, data_reg2, TCG_REG_R1, 4); tcg_out_ld32_12(s, COND_AL, data_reg2, TCG_REG_R1, 4);
@ -1450,9 +1480,13 @@ static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, int opc)
} }
break; break;
case 3: case 3:
/* TODO: use block load - if (use_armv6_instructions && !bswap
* check that data_reg2 > data_reg or the other way */ && (data_reg & 1) == 0 && data_reg2 == data_reg + 1) {
if (data_reg == addr_reg) { tcg_out_ldrd_8(s, COND_AL, data_reg, addr_reg, 0);
} else if (use_armv6_instructions && bswap
&& (data_reg2 & 1) == 0 && data_reg == data_reg2 + 1) {
tcg_out_ldrd_8(s, COND_AL, data_reg2, addr_reg, 0);
} else if (data_reg == addr_reg) {
tcg_out_ld32_12(s, COND_AL, data_reg2, addr_reg, bswap ? 0 : 4); tcg_out_ld32_12(s, COND_AL, data_reg2, addr_reg, bswap ? 0 : 4);
tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, bswap ? 4 : 0); tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, bswap ? 4 : 0);
} else { } else {
@ -1529,6 +1563,9 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
tcg_out_st32_rwb(s, COND_AL, TCG_REG_R0, TCG_REG_R1, addr_reg); tcg_out_st32_rwb(s, COND_AL, TCG_REG_R0, TCG_REG_R1, addr_reg);
tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg); tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg);
tcg_out_st32_12(s, COND_AL, TCG_REG_R0, TCG_REG_R1, 4); tcg_out_st32_12(s, COND_AL, TCG_REG_R0, TCG_REG_R1, 4);
} else if (use_armv6_instructions
&& (data_reg & 1) == 0 && data_reg2 == data_reg + 1) {
tcg_out_strd_r(s, COND_AL, data_reg, addr_reg, TCG_REG_R1);
} else { } else {
tcg_out_st32_rwb(s, COND_AL, data_reg, TCG_REG_R1, addr_reg); tcg_out_st32_rwb(s, COND_AL, data_reg, TCG_REG_R1, addr_reg);
tcg_out_st32_12(s, COND_AL, data_reg2, TCG_REG_R1, 4); tcg_out_st32_12(s, COND_AL, data_reg2, TCG_REG_R1, 4);
@ -1576,13 +1613,14 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
} }
break; break;
case 3: case 3:
/* TODO: use block store -
* check that data_reg2 > data_reg or the other way */
if (bswap) { if (bswap) {
tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg2); tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg2);
tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addr_reg, 0); tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addr_reg, 0);
tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg); tcg_out_bswap32(s, COND_AL, TCG_REG_R0, data_reg);
tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addr_reg, 4); tcg_out_st32_12(s, COND_AL, TCG_REG_R0, addr_reg, 4);
} else if (use_armv6_instructions
&& (data_reg & 1) == 0 && data_reg2 == data_reg + 1) {
tcg_out_strd_8(s, COND_AL, data_reg, addr_reg, 0);
} else { } else {
tcg_out_st32_12(s, COND_AL, data_reg, addr_reg, 0); tcg_out_st32_12(s, COND_AL, data_reg, addr_reg, 0);
tcg_out_st32_12(s, COND_AL, data_reg2, addr_reg, 4); tcg_out_st32_12(s, COND_AL, data_reg2, addr_reg, 4);