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

tcg-ia64: Load GUEST_BASE into a register. 

Saves one bundle per memory operation.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
This commit is contained in:
Richard Henderson 2010-05-06 08:50:44 -07:00 committed by Aurelien Jarno
parent 7221f058a6
commit 6781d08d2e
1 changed files with 88 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

132
tcg/ia64/tcg-target.c
View File

@ -40,6 +40,12 @@ static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
};
#endif
#ifdef CONFIG_USE_GUEST_BASE
#define TCG_GUEST_BASE_REG TCG_REG_R55
#else
#define TCG_GUEST_BASE_REG TCG_REG_R0
#endif
/* Branch registers */
enum {
TCG_REG_B0 = 0,
@ -1640,9 +1646,13 @@ static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, int opc)
{
static uint64_t const opc_ld_m1[4] = {
OPC_LD1_M1, OPC_LD2_M1, OPC_LD4_M1, OPC_LD8_M1
};
static uint64_t const opc_sxt_i29[4] = {
OPC_SXT1_I29, OPC_SXT2_I29, OPC_SXT4_I29, 0
};
int addr_reg, data_reg, mem_index, s_bits, bswap;
uint64_t opc_ld_m1[4] = { OPC_LD1_M1, OPC_LD2_M1, OPC_LD4_M1, OPC_LD8_M1 };
uint64_t opc_sxt_i29[8] = { OPC_SXT1_I29, OPC_SXT2_I29, OPC_SXT4_I29, 0 };
data_reg = *args++;
addr_reg = *args++;
@ -1655,19 +1665,21 @@ static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, int opc)
bswap = 0;
#endif
tcg_out_bundle(s, mLX,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_l2 ((tcg_target_long) GUEST_BASE),
tcg_opc_x2 (TCG_REG_P0, OPC_MOVL_X2, TCG_REG_R2,
GUEST_BASE));
#if TARGET_LONG_BITS == 32
tcg_out_bundle(s, mII,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_i29(TCG_REG_P0, OPC_ZXT4_I29,
TCG_REG_R3, addr_reg),
tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_REG_R2, TCG_REG_R3));
if (GUEST_BASE != 0) {
tcg_out_bundle(s, mII,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_i29(TCG_REG_P0, OPC_ZXT4_I29,
TCG_REG_R3, addr_reg),
tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_GUEST_BASE_REG, TCG_REG_R3));
} else {
tcg_out_bundle(s, miI,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_i29(TCG_REG_P0, OPC_ZXT4_I29,
TCG_REG_R2, addr_reg),
tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0));
}
if (!bswap || s_bits == 0) {
if (s_bits == opc) {
@ -1723,12 +1735,20 @@ static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, int opc)
}
}
#else
tcg_out_bundle(s, MmI,
tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_REG_R2, addr_reg),
tcg_opc_m1 (TCG_REG_P0, opc_ld_m1[s_bits],
data_reg, TCG_REG_R2),
tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0));
if (GUEST_BASE != 0) {
tcg_out_bundle(s, MmI,
tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_GUEST_BASE_REG, addr_reg),
tcg_opc_m1 (TCG_REG_P0, opc_ld_m1[s_bits],
data_reg, TCG_REG_R2),
tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0));
} else {
tcg_out_bundle(s, mmI,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_m1 (TCG_REG_P0, opc_ld_m1[s_bits],
data_reg, addr_reg),
tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0));
}
if (bswap && s_bits == 1) {
tcg_out_bundle(s, mII,
@ -1763,8 +1783,13 @@ static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, int opc)
static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
{
static uint64_t const opc_st_m4[4] = {
OPC_ST1_M4, OPC_ST2_M4, OPC_ST4_M4, OPC_ST8_M4
};
int addr_reg, data_reg, bswap;
uint64_t opc_st_m4[4] = { OPC_ST1_M4, OPC_ST2_M4, OPC_ST4_M4, OPC_ST8_M4 };
#if TARGET_LONG_BITS == 64
uint64_t add_guest_base;
#endif
data_reg = *args++;
addr_reg = *args++;
@ -1775,19 +1800,22 @@ static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
bswap = 0;
#endif
tcg_out_bundle(s, mLX,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_l2 ((tcg_target_long) GUEST_BASE),
tcg_opc_x2 (TCG_REG_P0, OPC_MOVL_X2, TCG_REG_R2,
GUEST_BASE));
#if TARGET_LONG_BITS == 32
tcg_out_bundle(s, mII,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_i29(TCG_REG_P0, OPC_ZXT4_I29,
TCG_REG_R3, addr_reg),
tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_REG_R2, TCG_REG_R3));
if (GUEST_BASE != 0) {
tcg_out_bundle(s, mII,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_i29(TCG_REG_P0, OPC_ZXT4_I29,
TCG_REG_R3, addr_reg),
tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_GUEST_BASE_REG, TCG_REG_R3));
} else {
tcg_out_bundle(s, miI,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_i29(TCG_REG_P0, OPC_ZXT4_I29,
TCG_REG_R3, addr_reg),
tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0));
}
if (bswap) {
if (opc == 1) {
tcg_out_bundle(s, mII,
@ -1820,18 +1848,24 @@ static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0));
#else
if (GUEST_BASE != 0) {
add_guest_base = tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_GUEST_BASE_REG, addr_reg);
addr_reg = TCG_REG_R2;
} else {
add_guest_base = tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0);
}
if (!bswap || opc == 0) {
tcg_out_bundle(s, MmI,
tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_REG_R2, addr_reg),
tcg_out_bundle(s, (GUEST_BASE ? MmI : mmI),
add_guest_base,
tcg_opc_m4 (TCG_REG_P0, opc_st_m4[opc],
data_reg, TCG_REG_R2),
data_reg, addr_reg),
tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0));
} else {
if (opc == 1) {
tcg_out_bundle(s, mII,
tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_REG_R2, addr_reg),
add_guest_base,
tcg_opc_i12(TCG_REG_P0, OPC_DEP_Z_I12,
TCG_REG_R3, data_reg, 15, 15),
tcg_opc_i3 (TCG_REG_P0, OPC_MUX1_I3,
@ -1839,8 +1873,7 @@ static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
data_reg = TCG_REG_R3;
} else if (opc == 2) {
tcg_out_bundle(s, mII,
tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_REG_R2, addr_reg),
add_guest_base,
tcg_opc_i12(TCG_REG_P0, OPC_DEP_Z_I12,
TCG_REG_R3, data_reg, 31, 31),
tcg_opc_i3 (TCG_REG_P0, OPC_MUX1_I3,
@ -1848,8 +1881,7 @@ static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
data_reg = TCG_REG_R3;
} else if (opc == 3) {
tcg_out_bundle(s, miI,
tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
TCG_REG_R2, addr_reg),
add_guest_base,
tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0),
tcg_opc_i3 (TCG_REG_P0, OPC_MUX1_I3,
TCG_REG_R3, data_reg, 0xb));
@ -1857,7 +1889,7 @@ static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)
}
tcg_out_bundle(s, miI,
tcg_opc_m4 (TCG_REG_P0, opc_st_m4[opc],
data_reg, TCG_REG_R2),
data_reg, addr_reg),
tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0),
tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0));
}
@ -2254,6 +2286,18 @@ void tcg_target_qemu_prologue(TCGContext *s)
TCG_REG_B6, TCG_REG_R32, 0),
tcg_opc_i22(TCG_REG_P0, OPC_MOV_I22,
TCG_REG_R32, TCG_REG_B0));
/* ??? If GUEST_BASE < 0x200000, we could load the register via
an ADDL in the M slot of the next bundle. */
if (GUEST_BASE != 0) {
tcg_out_bundle(s, mlx,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_l2 (GUEST_BASE),
tcg_opc_x2 (TCG_REG_P0, OPC_MOVL_X2,
TCG_GUEST_BASE_REG, GUEST_BASE));
tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
}
tcg_out_bundle(s, miB,
tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
tcg_opc_a4 (TCG_REG_P0, OPC_ADDS_A4,