* [Qemu-devel] [PATCH v3 0/8] target-sh4: optimizations and cleanups
@ 2015-05-24 23:37 Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 1/8] target-sh4: use bit number for SR constants Aurelien Jarno
` (7 more replies)
0 siblings, 8 replies; 16+ messages in thread
From: Aurelien Jarno @ 2015-05-24 23:37 UTC (permalink / raw)
To: qemu-devel; +Cc: Aurelien Jarno
This patchset improves the SH4 emulation by using the recently added
TCG instructions, namely add2, sub2 and movcond. For that the T, Q and
M bits are split out from the SR register. This allow the optimizer and
the liveness analysis to do some more optimisations.
The last two patches are doing cleanup in the code.
Changes v2 -> v3:
- rebased
- patch 2: don't mask out T bit from env->sr when reading
- patch 6: don't mask out Q and M bits from env->sr when reading
correctly compute M bit in div0s
add some comments to explain the tricks used in div1
Changes v1 -> v2:
- rebased
- added last patch
Aurelien Jarno (8):
target-sh4: use bit number for SR constants
target-sh4: Split out T from SR
target-sh4: optimize addc using add2
target-sh4: optimize subc using sub2
target-sh4: optimize negc using add2 and sub2
target-sh4: split out Q and M from of SR and optimize div1
target-sh4: factorize fmov implementation
target-sh4: remove dead code
target-sh4/cpu.c | 3 +-
target-sh4/cpu.h | 50 +++++---
target-sh4/gdbstub.c | 8 +-
target-sh4/helper.c | 29 ++---
target-sh4/helper.h | 1 -
target-sh4/op_helper.c | 148 +-----------------------
target-sh4/translate.c | 307 +++++++++++++++++++++++++------------------------
7 files changed, 217 insertions(+), 329 deletions(-)
--
2.1.4
^ permalink raw reply [flat|nested] 16+ messages in thread
* [Qemu-devel] [PATCH v3 1/8] target-sh4: use bit number for SR constants
2015-05-24 23:37 [Qemu-devel] [PATCH v3 0/8] target-sh4: optimizations and cleanups Aurelien Jarno
@ 2015-05-24 23:37 ` Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 2/8] target-sh4: Split out T from SR Aurelien Jarno
` (6 subsequent siblings)
7 siblings, 0 replies; 16+ messages in thread
From: Aurelien Jarno @ 2015-05-24 23:37 UTC (permalink / raw)
To: qemu-devel; +Cc: Aurelien Jarno
Use the bit number for SR constants instead of using a bit mask. This
make possible to also use the constants for shifts.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
---
target-sh4/cpu.c | 3 +-
target-sh4/cpu.h | 30 ++++++++++----------
target-sh4/gdbstub.c | 4 +--
target-sh4/helper.c | 27 +++++++++---------
target-sh4/op_helper.c | 26 ++++++++---------
target-sh4/translate.c | 75 ++++++++++++++++++++++++++------------------------
6 files changed, 85 insertions(+), 80 deletions(-)
diff --git a/target-sh4/cpu.c b/target-sh4/cpu.c
index d187a2b..cccb14f 100644
--- a/target-sh4/cpu.c
+++ b/target-sh4/cpu.c
@@ -61,7 +61,8 @@ static void superh_cpu_reset(CPUState *s)
env->fpscr = FPSCR_PR; /* value for userspace according to the kernel */
set_float_rounding_mode(float_round_nearest_even, &env->fp_status); /* ?! */
#else
- env->sr = SR_MD | SR_RB | SR_BL | SR_I3 | SR_I2 | SR_I1 | SR_I0;
+ env->sr = (1u << SR_MD) | (1u << SR_RB) | (1u << SR_BL) |
+ (1u << SR_I3) | (1u << SR_I2) | (1u << SR_I1) | (1u << SR_I0);
env->fpscr = FPSCR_DN | FPSCR_RM_ZERO; /* CPU reset value according to SH4 manual */
set_float_rounding_mode(float_round_to_zero, &env->fp_status);
set_flush_to_zero(1, &env->fp_status);
diff --git a/target-sh4/cpu.h b/target-sh4/cpu.h
index c8dea6c..76fda35 100644
--- a/target-sh4/cpu.h
+++ b/target-sh4/cpu.h
@@ -47,18 +47,18 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 32
#define TARGET_VIRT_ADDR_SPACE_BITS 32
-#define SR_MD (1 << 30)
-#define SR_RB (1 << 29)
-#define SR_BL (1 << 28)
-#define SR_FD (1 << 15)
-#define SR_M (1 << 9)
-#define SR_Q (1 << 8)
-#define SR_I3 (1 << 7)
-#define SR_I2 (1 << 6)
-#define SR_I1 (1 << 5)
-#define SR_I0 (1 << 4)
-#define SR_S (1 << 1)
-#define SR_T (1 << 0)
+#define SR_MD 30
+#define SR_RB 29
+#define SR_BL 28
+#define SR_FD 15
+#define SR_M 9
+#define SR_Q 8
+#define SR_I3 7
+#define SR_I2 6
+#define SR_I1 5
+#define SR_I0 4
+#define SR_S 1
+#define SR_T 0
#define FPSCR_MASK (0x003fffff)
#define FPSCR_FR (1 << 21)
@@ -234,7 +234,7 @@ void cpu_load_tlb(CPUSH4State * env);
#define MMU_USER_IDX 1
static inline int cpu_mmu_index (CPUSH4State *env)
{
- return (env->sr & SR_MD) == 0 ? 1 : 0;
+ return (env->sr & (1u << SR_MD)) == 0 ? 1 : 0;
}
#include "exec/cpu-all.h"
@@ -339,8 +339,8 @@ static inline void cpu_get_tb_cpu_state(CPUSH4State *env, target_ulong *pc,
*flags = (env->flags & (DELAY_SLOT | DELAY_SLOT_CONDITIONAL
| DELAY_SLOT_TRUE | DELAY_SLOT_CLEARME)) /* Bits 0- 3 */
| (env->fpscr & (FPSCR_FR | FPSCR_SZ | FPSCR_PR)) /* Bits 19-21 */
- | (env->sr & (SR_MD | SR_RB)) /* Bits 29-30 */
- | (env->sr & SR_FD) /* Bit 15 */
+ | (env->sr & ((1u << SR_MD) | (1u << SR_RB))) /* Bits 29-30 */
+ | (env->sr & (1u << SR_FD)) /* Bit 15 */
| (env->movcal_backup ? TB_FLAG_PENDING_MOVCA : 0); /* Bit 4 */
}
diff --git a/target-sh4/gdbstub.c b/target-sh4/gdbstub.c
index df4fa2a..05ba728 100644
--- a/target-sh4/gdbstub.c
+++ b/target-sh4/gdbstub.c
@@ -31,7 +31,7 @@ int superh_cpu_gdb_read_register(CPUState *cs, uint8_t *mem_buf, int n)
switch (n) {
case 0 ... 7:
- if ((env->sr & (SR_MD | SR_RB)) == (SR_MD | SR_RB)) {
+ if ((env->sr & (1u << SR_MD)) && (env->sr & (1u << SR_RB))) {
return gdb_get_regl(mem_buf, env->gregs[n + 16]);
} else {
return gdb_get_regl(mem_buf, env->gregs[n]);
@@ -83,7 +83,7 @@ int superh_cpu_gdb_write_register(CPUState *cs, uint8_t *mem_buf, int n)
switch (n) {
case 0 ... 7:
- if ((env->sr & (SR_MD | SR_RB)) == (SR_MD | SR_RB)) {
+ if ((env->sr & (1u << SR_MD)) && (env->sr & (1u << SR_RB))) {
env->gregs[n + 16] = ldl_p(mem_buf);
} else {
env->gregs[n] = ldl_p(mem_buf);
diff --git a/target-sh4/helper.c b/target-sh4/helper.c
index 5811360..1cb0e8d 100644
--- a/target-sh4/helper.c
+++ b/target-sh4/helper.c
@@ -93,7 +93,7 @@ void superh_cpu_do_interrupt(CPUState *cs)
do_exp = cs->exception_index != -1;
do_irq = do_irq && (cs->exception_index == -1);
- if (env->sr & SR_BL) {
+ if (env->sr & (1u << SR_BL)) {
if (do_exp && cs->exception_index != 0x1e0) {
cs->exception_index = 0x000; /* masked exception -> reset */
}
@@ -165,7 +165,7 @@ void superh_cpu_do_interrupt(CPUState *cs)
env->ssr = env->sr;
env->spc = env->pc;
env->sgr = env->gregs[15];
- env->sr |= SR_BL | SR_MD | SR_RB;
+ env->sr |= (1u << SR_BL) | (1u << SR_MD) | (1u << SR_RB);
if (env->flags & (DELAY_SLOT | DELAY_SLOT_CONDITIONAL)) {
/* Branch instruction should be executed again before delay slot. */
@@ -182,7 +182,7 @@ void superh_cpu_do_interrupt(CPUState *cs)
case 0x000:
case 0x020:
case 0x140:
- env->sr &= ~SR_FD;
+ env->sr &= ~(1u << SR_FD);
env->sr |= 0xf << 4; /* IMASK */
env->pc = 0xa0000000;
break;
@@ -355,23 +355,24 @@ static int get_mmu_address(CPUSH4State * env, target_ulong * physical,
int use_asid, n;
tlb_t *matching = NULL;
- use_asid = (env->mmucr & MMUCR_SV) == 0 || (env->sr & SR_MD) == 0;
+ use_asid = !(env->mmucr & MMUCR_SV) || !(env->sr & (1u << SR_MD));
if (rw == 2) {
n = find_itlb_entry(env, address, use_asid);
if (n >= 0) {
matching = &env->itlb[n];
- if (!(env->sr & SR_MD) && !(matching->pr & 2))
+ if (!(env->sr & (1u << SR_MD)) && !(matching->pr & 2)) {
n = MMU_ITLB_VIOLATION;
- else
+ } else {
*prot = PAGE_EXEC;
+ }
} else {
n = find_utlb_entry(env, address, use_asid);
if (n >= 0) {
n = copy_utlb_entry_itlb(env, n);
matching = &env->itlb[n];
- if (!(env->sr & SR_MD) && !(matching->pr & 2)) {
- n = MMU_ITLB_VIOLATION;
+ if (!(env->sr & (1u << SR_MD)) && !(matching->pr & 2)) {
+ n = MMU_ITLB_VIOLATION;
} else {
*prot = PAGE_READ | PAGE_EXEC;
if ((matching->pr & 1) && matching->d) {
@@ -388,7 +389,7 @@ static int get_mmu_address(CPUSH4State * env, target_ulong * physical,
n = find_utlb_entry(env, address, use_asid);
if (n >= 0) {
matching = &env->utlb[n];
- if (!(env->sr & SR_MD) && !(matching->pr & 2)) {
+ if (!(env->sr & (1u << SR_MD)) && !(matching->pr & 2)) {
n = (rw == 1) ? MMU_DTLB_VIOLATION_WRITE :
MMU_DTLB_VIOLATION_READ;
} else if ((rw == 1) && !(matching->pr & 1)) {
@@ -421,7 +422,7 @@ static int get_physical_address(CPUSH4State * env, target_ulong * physical,
/* P1, P2 and P4 areas do not use translation */
if ((address >= 0x80000000 && address < 0xc0000000) ||
address >= 0xe0000000) {
- if (!(env->sr & SR_MD)
+ if (!(env->sr & (1u << SR_MD))
&& (address < 0xe0000000 || address >= 0xe4000000)) {
/* Unauthorized access in user mode (only store queues are available) */
fprintf(stderr, "Unauthorized access\n");
@@ -690,7 +691,7 @@ void cpu_sh4_write_mmaped_utlb_addr(CPUSH4State *s, hwaddr addr,
uint8_t d = (uint8_t)((mem_value & 0x00000200) >> 9);
uint8_t v = (uint8_t)((mem_value & 0x00000100) >> 8);
uint8_t asid = (uint8_t)(mem_value & 0x000000ff);
- int use_asid = (s->mmucr & MMUCR_SV) == 0 || (s->sr & SR_MD) == 0;
+ int use_asid = !(s->mmucr & MMUCR_SV) || !(s->sr & (1u << SR_MD));
if (associate) {
int i;
@@ -821,10 +822,10 @@ void cpu_sh4_write_mmaped_utlb_data(CPUSH4State *s, hwaddr addr,
int cpu_sh4_is_cached(CPUSH4State * env, target_ulong addr)
{
int n;
- int use_asid = (env->mmucr & MMUCR_SV) == 0 || (env->sr & SR_MD) == 0;
+ int use_asid = !(env->mmucr & MMUCR_SV) || !(env->sr & (1u << SR_MD));
/* check area */
- if (env->sr & SR_MD) {
+ if (env->sr & (1u << SR_MD)) {
/* For previledged mode, P2 and P4 area is not cachable. */
if ((0xA0000000 <= addr && addr < 0xC0000000) || 0xE0000000 <= addr)
return 0;
diff --git a/target-sh4/op_helper.c b/target-sh4/op_helper.c
index 74a5c4e..6f34292 100644
--- a/target-sh4/op_helper.c
+++ b/target-sh4/op_helper.c
@@ -156,15 +156,15 @@ void helper_ocbi(CPUSH4State *env, uint32_t address)
}
}
-#define T (env->sr & SR_T)
-#define Q (env->sr & SR_Q ? 1 : 0)
-#define M (env->sr & SR_M ? 1 : 0)
-#define SETT env->sr |= SR_T
-#define CLRT env->sr &= ~SR_T
-#define SETQ env->sr |= SR_Q
-#define CLRQ env->sr &= ~SR_Q
-#define SETM env->sr |= SR_M
-#define CLRM env->sr &= ~SR_M
+#define T (env->sr & (1u << SR_T))
+#define Q (env->sr & (1u << SR_Q) ? 1 : 0)
+#define M (env->sr & (1u << SR_M) ? 1 : 0)
+#define SETT (env->sr |= (1u << SR_T))
+#define CLRT (env->sr &= ~(1u << SR_T))
+#define SETQ (env->sr |= (1u << SR_Q))
+#define CLRQ (env->sr &= ~(1u << SR_Q))
+#define SETM (env->sr |= (1u << SR_M))
+#define CLRM (env->sr &= ~(1u << SR_M))
uint32_t helper_div1(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
{
@@ -282,7 +282,7 @@ void helper_macl(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
res += (int64_t) (int32_t) arg0 *(int64_t) (int32_t) arg1;
env->mach = (res >> 32) & 0xffffffff;
env->macl = res & 0xffffffff;
- if (env->sr & SR_S) {
+ if (env->sr & (1u << SR_S)) {
if (res < 0)
env->mach |= 0xffff0000;
else
@@ -298,7 +298,7 @@ void helper_macw(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
res += (int64_t) (int16_t) arg0 *(int64_t) (int16_t) arg1;
env->mach = (res >> 32) & 0xffffffff;
env->macl = res & 0xffffffff;
- if (env->sr & SR_S) {
+ if (env->sr & (1u << SR_S)) {
if (res < -0x80000000) {
env->mach = 1;
env->macl = 0x80000000;
@@ -311,12 +311,12 @@ void helper_macw(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
static inline void set_t(CPUSH4State *env)
{
- env->sr |= SR_T;
+ env->sr |= (1u << SR_T);
}
static inline void clr_t(CPUSH4State *env)
{
- env->sr &= ~SR_T;
+ env->sr &= ~(1u << SR_T);
}
void helper_ld_fpscr(CPUSH4State *env, uint32_t val)
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index 41aa928..b25f79c 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -47,7 +47,7 @@ typedef struct DisasContext {
#if defined(CONFIG_USER_ONLY)
#define IS_USER(ctx) 1
#else
-#define IS_USER(ctx) (!(ctx->flags & SR_MD))
+#define IS_USER(ctx) (!(ctx->flags & (1u << SR_MD)))
#endif
enum {
@@ -214,7 +214,7 @@ static inline void gen_branch_slot(uint32_t delayed_pc, int t)
TCGLabel *label = gen_new_label();
tcg_gen_movi_i32(cpu_delayed_pc, delayed_pc);
sr = tcg_temp_new();
- tcg_gen_andi_i32(sr, cpu_sr, SR_T);
+ tcg_gen_andi_i32(sr, cpu_sr, (1u << SR_T));
tcg_gen_brcondi_i32(t ? TCG_COND_EQ:TCG_COND_NE, sr, 0, label);
tcg_gen_ori_i32(cpu_flags, cpu_flags, DELAY_SLOT_TRUE);
gen_set_label(label);
@@ -229,7 +229,7 @@ static void gen_conditional_jump(DisasContext * ctx,
l1 = gen_new_label();
sr = tcg_temp_new();
- tcg_gen_andi_i32(sr, cpu_sr, SR_T);
+ tcg_gen_andi_i32(sr, cpu_sr, (1u << SR_T));
tcg_gen_brcondi_i32(TCG_COND_NE, sr, 0, l1);
gen_goto_tb(ctx, 0, ifnott);
gen_set_label(l1);
@@ -258,7 +258,7 @@ static inline void gen_cmp(int cond, TCGv t0, TCGv t1)
t = tcg_temp_new();
tcg_gen_setcond_i32(cond, t, t1, t0);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_T);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
tcg_gen_or_i32(cpu_sr, cpu_sr, t);
tcg_temp_free(t);
@@ -270,7 +270,7 @@ static inline void gen_cmp_imm(int cond, TCGv t0, int32_t imm)
t = tcg_temp_new();
tcg_gen_setcondi_i32(cond, t, t0, imm);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_T);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
tcg_gen_or_i32(cpu_sr, cpu_sr, t);
tcg_temp_free(t);
@@ -326,10 +326,12 @@ static inline void gen_store_fpr64 (TCGv_i64 t, int reg)
#define B11_8 ((ctx->opcode >> 8) & 0xf)
#define B15_12 ((ctx->opcode >> 12) & 0xf)
-#define REG(x) ((x) < 8 && (ctx->flags & (SR_MD | SR_RB)) == (SR_MD | SR_RB) \
+#define REG(x) ((x) < 8 && (ctx->flags & (1u << SR_MD))\
+ && (ctx->flags & (1u << SR_RB))\
? (cpu_gregs[x + 16]) : (cpu_gregs[x]))
-#define ALTREG(x) ((x) < 8 && (ctx->flags & (SR_MD | SR_RB)) != (SR_MD | SR_RB)\
+#define ALTREG(x) ((x) < 8 && (!(ctx->flags & (1u << SR_MD))\
+ || !(ctx->flags & (1u << SR_RB)))\
? (cpu_gregs[x + 16]) : (cpu_gregs[x]))
#define FREG(x) (ctx->flags & FPSCR_FR ? (x) ^ 0x10 : (x))
@@ -359,7 +361,7 @@ static inline void gen_store_fpr64 (TCGv_i64 t, int reg)
}
#define CHECK_FPU_ENABLED \
- if (ctx->flags & SR_FD) { \
+ if (ctx->flags & (1u << SR_FD)) { \
tcg_gen_movi_i32(cpu_pc, ctx->pc); \
if (ctx->flags & (DELAY_SLOT | DELAY_SLOT_CONDITIONAL)) { \
gen_helper_raise_slot_fpu_disable(cpu_env); \
@@ -409,7 +411,8 @@ static void _decode_opc(DisasContext * ctx)
switch (ctx->opcode) {
case 0x0019: /* div0u */
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(SR_M | SR_Q | SR_T));
+ tcg_gen_andi_i32(cpu_sr, cpu_sr,
+ ~((1u << SR_M) | (1u << SR_Q) | (1u << SR_T)));
return;
case 0x000b: /* rts */
CHECK_NOT_DELAY_SLOT
@@ -422,10 +425,10 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_movi_i32(cpu_macl, 0);
return;
case 0x0048: /* clrs */
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_S);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_S));
return;
case 0x0008: /* clrt */
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_T);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
return;
case 0x0038: /* ldtlb */
CHECK_PRIVILEGED
@@ -440,10 +443,10 @@ static void _decode_opc(DisasContext * ctx)
ctx->delayed_pc = (uint32_t) - 1;
return;
case 0x0058: /* sets */
- tcg_gen_ori_i32(cpu_sr, cpu_sr, SR_S);
+ tcg_gen_ori_i32(cpu_sr, cpu_sr, (1u << SR_S));
return;
case 0x0018: /* sett */
- tcg_gen_ori_i32(cpu_sr, cpu_sr, SR_T);
+ tcg_gen_ori_i32(cpu_sr, cpu_sr, (1u << SR_T));
return;
case 0xfbfd: /* frchg */
tcg_gen_xori_i32(cpu_fpscr, cpu_fpscr, FPSCR_FR);
@@ -661,7 +664,7 @@ static void _decode_opc(DisasContext * ctx)
{
TCGv t0, t1, t2;
t0 = tcg_temp_new();
- tcg_gen_andi_i32(t0, cpu_sr, SR_T);
+ tcg_gen_andi_i32(t0, cpu_sr, (1u << SR_T));
t1 = tcg_temp_new();
tcg_gen_add_i32(t1, REG(B7_4), REG(B11_8));
tcg_gen_add_i32(t0, t0, t1);
@@ -670,7 +673,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_setcond_i32(TCG_COND_GTU, t1, t1, t0);
tcg_gen_or_i32(t1, t1, t2);
tcg_temp_free(t2);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_T);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
tcg_temp_free(t1);
tcg_gen_mov_i32(REG(B11_8), t0);
@@ -689,7 +692,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_andc_i32(t1, t1, t2);
tcg_temp_free(t2);
tcg_gen_shri_i32(t1, t1, 31);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_T);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
tcg_temp_free(t1);
tcg_gen_mov_i32(REG(B7_4), t0);
@@ -718,7 +721,7 @@ static void _decode_opc(DisasContext * ctx)
{
TCGv cmp1 = tcg_temp_new();
TCGv cmp2 = tcg_temp_new();
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_T);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
tcg_gen_xor_i32(cmp1, REG(B7_4), REG(B11_8));
tcg_gen_andi_i32(cmp2, cmp1, 0xff000000);
tcg_gen_setcondi_i32(TCG_COND_EQ, cmp2, cmp2, 0);
@@ -738,11 +741,11 @@ static void _decode_opc(DisasContext * ctx)
return;
case 0x2007: /* div0s Rm,Rn */
{
- gen_copy_bit_i32(cpu_sr, 8, REG(B11_8), 31); /* SR_Q */
- gen_copy_bit_i32(cpu_sr, 9, REG(B7_4), 31); /* SR_M */
+ gen_copy_bit_i32(cpu_sr, SR_Q, REG(B11_8), 31); /* SR_Q */
+ gen_copy_bit_i32(cpu_sr, SR_M, REG(B7_4), 31); /* SR_M */
TCGv val = tcg_temp_new();
tcg_gen_xor_i32(val, REG(B7_4), REG(B11_8));
- gen_copy_bit_i32(cpu_sr, 0, val, 31); /* SR_T */
+ gen_copy_bit_i32(cpu_sr, SR_T, val, 31); /* SR_T */
tcg_temp_free(val);
}
return;
@@ -831,9 +834,9 @@ static void _decode_opc(DisasContext * ctx)
t0 = tcg_temp_new();
tcg_gen_neg_i32(t0, REG(B7_4));
t1 = tcg_temp_new();
- tcg_gen_andi_i32(t1, cpu_sr, SR_T);
+ tcg_gen_andi_i32(t1, cpu_sr, (1u << SR_T));
tcg_gen_sub_i32(REG(B11_8), t0, t1);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_T);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
tcg_gen_setcondi_i32(TCG_COND_GTU, t1, t0, 0);
tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
tcg_gen_setcond_i32(TCG_COND_GTU, t1, REG(B11_8), t0);
@@ -920,7 +923,7 @@ static void _decode_opc(DisasContext * ctx)
{
TCGv t0, t1, t2;
t0 = tcg_temp_new();
- tcg_gen_andi_i32(t0, cpu_sr, SR_T);
+ tcg_gen_andi_i32(t0, cpu_sr, (1u << SR_T));
t1 = tcg_temp_new();
tcg_gen_sub_i32(t1, REG(B11_8), REG(B7_4));
tcg_gen_sub_i32(t0, t1, t0);
@@ -929,7 +932,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_setcond_i32(TCG_COND_LTU, t1, t1, t0);
tcg_gen_or_i32(t1, t1, t2);
tcg_temp_free(t2);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_T);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
tcg_temp_free(t1);
tcg_gen_mov_i32(REG(B11_8), t0);
@@ -948,7 +951,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_and_i32(t1, t1, t2);
tcg_temp_free(t2);
tcg_gen_shri_i32(t1, t1, 31);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_T);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
tcg_temp_free(t1);
tcg_gen_mov_i32(REG(B11_8), t0);
@@ -1545,7 +1548,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_addi_i32(REG(B11_8), REG(B11_8), 4);
return;
case 0x0029: /* movt Rn */
- tcg_gen_andi_i32(REG(B11_8), cpu_sr, SR_T);
+ tcg_gen_andi_i32(REG(B11_8), cpu_sr, (1u << SR_T));
return;
case 0x0073:
/* MOVCO.L
@@ -1555,7 +1558,7 @@ static void _decode_opc(DisasContext * ctx)
*/
if (ctx->features & SH_FEATURE_SH4A) {
TCGLabel *label = gen_new_label();
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~SR_T);
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
tcg_gen_or_i32(cpu_sr, cpu_sr, cpu_ldst);
tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_ldst, 0, label);
tcg_gen_qemu_st_i32(REG(0), REG(B11_8), ctx->memidx, MO_TEUL);
@@ -1610,9 +1613,9 @@ static void _decode_opc(DisasContext * ctx)
{
TCGv tmp = tcg_temp_new();
tcg_gen_mov_i32(tmp, cpu_sr);
- gen_copy_bit_i32(cpu_sr, 0, REG(B11_8), 31);
+ gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 31);
tcg_gen_shli_i32(REG(B11_8), REG(B11_8), 1);
- gen_copy_bit_i32(REG(B11_8), 0, tmp, 0);
+ gen_copy_bit_i32(REG(B11_8), SR_T, tmp, 0);
tcg_temp_free(tmp);
}
return;
@@ -1620,7 +1623,7 @@ static void _decode_opc(DisasContext * ctx)
{
TCGv tmp = tcg_temp_new();
tcg_gen_mov_i32(tmp, cpu_sr);
- gen_copy_bit_i32(cpu_sr, 0, REG(B11_8), 0);
+ gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 0);
tcg_gen_shri_i32(REG(B11_8), REG(B11_8), 1);
gen_copy_bit_i32(REG(B11_8), 31, tmp, 0);
tcg_temp_free(tmp);
@@ -1628,23 +1631,23 @@ static void _decode_opc(DisasContext * ctx)
return;
case 0x4004: /* rotl Rn */
tcg_gen_rotli_i32(REG(B11_8), REG(B11_8), 1);
- gen_copy_bit_i32(cpu_sr, 0, REG(B11_8), 0);
+ gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 0);
return;
case 0x4005: /* rotr Rn */
- gen_copy_bit_i32(cpu_sr, 0, REG(B11_8), 0);
+ gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 0);
tcg_gen_rotri_i32(REG(B11_8), REG(B11_8), 1);
return;
case 0x4000: /* shll Rn */
case 0x4020: /* shal Rn */
- gen_copy_bit_i32(cpu_sr, 0, REG(B11_8), 31);
+ gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 31);
tcg_gen_shli_i32(REG(B11_8), REG(B11_8), 1);
return;
case 0x4021: /* shar Rn */
- gen_copy_bit_i32(cpu_sr, 0, REG(B11_8), 0);
+ gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 0);
tcg_gen_sari_i32(REG(B11_8), REG(B11_8), 1);
return;
case 0x4001: /* shlr Rn */
- gen_copy_bit_i32(cpu_sr, 0, REG(B11_8), 0);
+ gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 0);
tcg_gen_shri_i32(REG(B11_8), REG(B11_8), 1);
return;
case 0x4008: /* shll2 Rn */
@@ -1874,7 +1877,7 @@ gen_intermediate_code_internal(SuperHCPU *cpu, TranslationBlock *tb,
ctx.pc = pc_start;
ctx.flags = (uint32_t)tb->flags;
ctx.bstate = BS_NONE;
- ctx.memidx = (ctx.flags & SR_MD) == 0 ? 1 : 0;
+ ctx.memidx = (ctx.flags & (1u << SR_MD)) == 0 ? 1 : 0;
/* We don't know if the delayed pc came from a dynamic or static branch,
so assume it is a dynamic branch. */
ctx.delayed_pc = -1; /* use delayed pc from env pointer */
--
2.1.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [Qemu-devel] [PATCH v3 2/8] target-sh4: Split out T from SR
2015-05-24 23:37 [Qemu-devel] [PATCH v3 0/8] target-sh4: optimizations and cleanups Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 1/8] target-sh4: use bit number for SR constants Aurelien Jarno
@ 2015-05-24 23:37 ` Aurelien Jarno
2015-06-04 5:01 ` Richard Henderson
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 3/8] target-sh4: optimize addc using add2 Aurelien Jarno
` (5 subsequent siblings)
7 siblings, 1 reply; 16+ messages in thread
From: Aurelien Jarno @ 2015-05-24 23:37 UTC (permalink / raw)
To: qemu-devel; +Cc: Aurelien Jarno
In preparation for more efficient setting of this field.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
---
target-sh4/cpu.h | 14 +++-
target-sh4/gdbstub.c | 4 +-
target-sh4/helper.c | 2 +-
target-sh4/op_helper.c | 32 ++------
target-sh4/translate.c | 212 ++++++++++++++++++++-----------------------------
5 files changed, 111 insertions(+), 153 deletions(-)
diff --git a/target-sh4/cpu.h b/target-sh4/cpu.h
index 76fda35..a308c53 100644
--- a/target-sh4/cpu.h
+++ b/target-sh4/cpu.h
@@ -138,7 +138,8 @@ typedef struct CPUSH4State {
uint32_t flags; /* general execution flags */
uint32_t gregs[24]; /* general registers */
float32 fregs[32]; /* floating point registers */
- uint32_t sr; /* status register */
+ uint32_t sr; /* status register (with T split out) */
+ uint32_t sr_t; /* T bit of status register */
uint32_t ssr; /* saved status register */
uint32_t spc; /* saved program counter */
uint32_t gbr; /* global base register */
@@ -331,6 +332,17 @@ static inline int cpu_ptel_pr (uint32_t ptel)
#define TB_FLAG_PENDING_MOVCA (1 << 4)
+static inline target_ulong cpu_read_sr(CPUSH4State *env)
+{
+ return env->sr | (env->sr_t << SR_T);
+}
+
+static inline void cpu_write_sr(CPUSH4State *env, target_ulong sr)
+{
+ env->sr_t = sr & (1u << SR_T);
+ env->sr = sr & ~(1u << SR_T);
+}
+
static inline void cpu_get_tb_cpu_state(CPUSH4State *env, target_ulong *pc,
target_ulong *cs_base, int *flags)
{
diff --git a/target-sh4/gdbstub.c b/target-sh4/gdbstub.c
index 05ba728..a365a27 100644
--- a/target-sh4/gdbstub.c
+++ b/target-sh4/gdbstub.c
@@ -51,7 +51,7 @@ int superh_cpu_gdb_read_register(CPUState *cs, uint8_t *mem_buf, int n)
case 21:
return gdb_get_regl(mem_buf, env->macl);
case 22:
- return gdb_get_regl(mem_buf, env->sr);
+ return gdb_get_regl(mem_buf, cpu_read_sr(env));
case 23:
return gdb_get_regl(mem_buf, env->fpul);
case 24:
@@ -111,7 +111,7 @@ int superh_cpu_gdb_write_register(CPUState *cs, uint8_t *mem_buf, int n)
env->macl = ldl_p(mem_buf);
break;
case 22:
- env->sr = ldl_p(mem_buf);
+ cpu_write_sr(env, ldl_p(mem_buf));
break;
case 23:
env->fpul = ldl_p(mem_buf);
diff --git a/target-sh4/helper.c b/target-sh4/helper.c
index 1cb0e8d..a533f08 100644
--- a/target-sh4/helper.c
+++ b/target-sh4/helper.c
@@ -162,7 +162,7 @@ void superh_cpu_do_interrupt(CPUState *cs)
log_cpu_state(cs, 0);
}
- env->ssr = env->sr;
+ env->ssr = cpu_read_sr(env);
env->spc = env->pc;
env->sgr = env->gregs[15];
env->sr |= (1u << SR_BL) | (1u << SR_MD) | (1u << SR_RB);
diff --git a/target-sh4/op_helper.c b/target-sh4/op_helper.c
index 6f34292..524d7f6 100644
--- a/target-sh4/op_helper.c
+++ b/target-sh4/op_helper.c
@@ -156,11 +156,11 @@ void helper_ocbi(CPUSH4State *env, uint32_t address)
}
}
-#define T (env->sr & (1u << SR_T))
+#define T (env->sr_t)
#define Q (env->sr & (1u << SR_Q) ? 1 : 0)
#define M (env->sr & (1u << SR_M) ? 1 : 0)
-#define SETT (env->sr |= (1u << SR_T))
-#define CLRT (env->sr &= ~(1u << SR_T))
+#define SETT (env->sr_t = 1)
+#define CLRT (env->sr_t = 0)
#define SETQ (env->sr |= (1u << SR_Q))
#define CLRQ (env->sr &= ~(1u << SR_Q))
#define SETM (env->sr |= (1u << SR_M))
@@ -309,16 +309,6 @@ void helper_macw(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
}
}
-static inline void set_t(CPUSH4State *env)
-{
- env->sr |= (1u << SR_T);
-}
-
-static inline void clr_t(CPUSH4State *env)
-{
- env->sr &= ~(1u << SR_T);
-}
-
void helper_ld_fpscr(CPUSH4State *env, uint32_t val)
{
env->fpscr = val & FPSCR_MASK;
@@ -403,10 +393,8 @@ void helper_fcmp_eq_FT(CPUSH4State *env, float32 t0, float32 t1)
relation = float32_compare(t0, t1, &env->fp_status);
if (unlikely(relation == float_relation_unordered)) {
update_fpscr(env, GETPC());
- } else if (relation == float_relation_equal) {
- set_t(env);
} else {
- clr_t(env);
+ env->sr_t = (relation == float_relation_equal);
}
}
@@ -418,10 +406,8 @@ void helper_fcmp_eq_DT(CPUSH4State *env, float64 t0, float64 t1)
relation = float64_compare(t0, t1, &env->fp_status);
if (unlikely(relation == float_relation_unordered)) {
update_fpscr(env, GETPC());
- } else if (relation == float_relation_equal) {
- set_t(env);
} else {
- clr_t(env);
+ env->sr_t = (relation == float_relation_equal);
}
}
@@ -433,10 +419,8 @@ void helper_fcmp_gt_FT(CPUSH4State *env, float32 t0, float32 t1)
relation = float32_compare(t0, t1, &env->fp_status);
if (unlikely(relation == float_relation_unordered)) {
update_fpscr(env, GETPC());
- } else if (relation == float_relation_greater) {
- set_t(env);
} else {
- clr_t(env);
+ env->sr_t = (relation == float_relation_greater);
}
}
@@ -448,10 +432,8 @@ void helper_fcmp_gt_DT(CPUSH4State *env, float64 t0, float64 t1)
relation = float64_compare(t0, t1, &env->fp_status);
if (unlikely(relation == float_relation_unordered)) {
update_fpscr(env, GETPC());
- } else if (relation == float_relation_greater) {
- set_t(env);
} else {
- clr_t(env);
+ env->sr_t = (relation == float_relation_greater);
}
}
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index b25f79c..f9bc24c 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -62,7 +62,7 @@ enum {
/* global register indexes */
static TCGv_ptr cpu_env;
static TCGv cpu_gregs[24];
-static TCGv cpu_pc, cpu_sr, cpu_ssr, cpu_spc, cpu_gbr;
+static TCGv cpu_pc, cpu_sr, cpu_sr_t, cpu_ssr, cpu_spc, cpu_gbr;
static TCGv cpu_vbr, cpu_sgr, cpu_dbr, cpu_mach, cpu_macl;
static TCGv cpu_pr, cpu_fpscr, cpu_fpul, cpu_ldst;
static TCGv cpu_fregs[32];
@@ -110,6 +110,8 @@ void sh4_translate_init(void)
offsetof(CPUSH4State, pc), "PC");
cpu_sr = tcg_global_mem_new_i32(TCG_AREG0,
offsetof(CPUSH4State, sr), "SR");
+ cpu_sr_t = tcg_global_mem_new_i32(TCG_AREG0,
+ offsetof(CPUSH4State, sr_t), "SR_T");
cpu_ssr = tcg_global_mem_new_i32(TCG_AREG0,
offsetof(CPUSH4State, ssr), "SSR");
cpu_spc = tcg_global_mem_new_i32(TCG_AREG0,
@@ -156,7 +158,7 @@ void superh_cpu_dump_state(CPUState *cs, FILE *f,
CPUSH4State *env = &cpu->env;
int i;
cpu_fprintf(f, "pc=0x%08x sr=0x%08x pr=0x%08x fpscr=0x%08x\n",
- env->pc, env->sr, env->pr, env->fpscr);
+ env->pc, cpu_read_sr(env), env->pr, env->fpscr);
cpu_fprintf(f, "spc=0x%08x ssr=0x%08x gbr=0x%08x vbr=0x%08x\n",
env->spc, env->ssr, env->gbr, env->vbr);
cpu_fprintf(f, "sgr=0x%08x dbr=0x%08x delayed_pc=0x%08x fpul=0x%08x\n",
@@ -174,6 +176,16 @@ void superh_cpu_dump_state(CPUState *cs, FILE *f,
env->delayed_pc);
}
}
+static void gen_read_sr(TCGv dst)
+{
+ tcg_gen_or_i32(dst, cpu_sr, cpu_sr_t);
+}
+
+static void gen_write_sr(TCGv src)
+{
+ tcg_gen_andi_i32(cpu_sr, src, ~(1u << SR_T));
+ tcg_gen_andi_i32(cpu_sr_t, src, (1u << SR_T));
+}
static void gen_goto_tb(DisasContext * ctx, int n, target_ulong dest)
{
@@ -210,12 +222,9 @@ static void gen_jump(DisasContext * ctx)
static inline void gen_branch_slot(uint32_t delayed_pc, int t)
{
- TCGv sr;
TCGLabel *label = gen_new_label();
tcg_gen_movi_i32(cpu_delayed_pc, delayed_pc);
- sr = tcg_temp_new();
- tcg_gen_andi_i32(sr, cpu_sr, (1u << SR_T));
- tcg_gen_brcondi_i32(t ? TCG_COND_EQ:TCG_COND_NE, sr, 0, label);
+ tcg_gen_brcondi_i32(t ? TCG_COND_EQ : TCG_COND_NE, cpu_sr_t, 0, label);
tcg_gen_ori_i32(cpu_flags, cpu_flags, DELAY_SLOT_TRUE);
gen_set_label(label);
}
@@ -224,13 +233,8 @@ static inline void gen_branch_slot(uint32_t delayed_pc, int t)
static void gen_conditional_jump(DisasContext * ctx,
target_ulong ift, target_ulong ifnott)
{
- TCGLabel *l1;
- TCGv sr;
-
- l1 = gen_new_label();
- sr = tcg_temp_new();
- tcg_gen_andi_i32(sr, cpu_sr, (1u << SR_T));
- tcg_gen_brcondi_i32(TCG_COND_NE, sr, 0, l1);
+ TCGLabel *l1 = gen_new_label();
+ tcg_gen_brcondi_i32(TCG_COND_NE, cpu_sr_t, 0, l1);
gen_goto_tb(ctx, 0, ifnott);
gen_set_label(l1);
gen_goto_tb(ctx, 1, ift);
@@ -252,30 +256,6 @@ static void gen_delayed_conditional_jump(DisasContext * ctx)
gen_jump(ctx);
}
-static inline void gen_cmp(int cond, TCGv t0, TCGv t1)
-{
- TCGv t;
-
- t = tcg_temp_new();
- tcg_gen_setcond_i32(cond, t, t1, t0);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
- tcg_gen_or_i32(cpu_sr, cpu_sr, t);
-
- tcg_temp_free(t);
-}
-
-static inline void gen_cmp_imm(int cond, TCGv t0, int32_t imm)
-{
- TCGv t;
-
- t = tcg_temp_new();
- tcg_gen_setcondi_i32(cond, t, t0, imm);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
- tcg_gen_or_i32(cpu_sr, cpu_sr, t);
-
- tcg_temp_free(t);
-}
-
static inline void gen_store_flags(uint32_t flags)
{
tcg_gen_andi_i32(cpu_flags, cpu_flags, DELAY_SLOT_TRUE);
@@ -411,8 +391,8 @@ static void _decode_opc(DisasContext * ctx)
switch (ctx->opcode) {
case 0x0019: /* div0u */
- tcg_gen_andi_i32(cpu_sr, cpu_sr,
- ~((1u << SR_M) | (1u << SR_Q) | (1u << SR_T)));
+ tcg_gen_andi_i32(cpu_sr, cpu_sr, ~((1u << SR_M) | (1u << SR_Q)));
+ tcg_gen_movi_i32(cpu_sr_t, 0);
return;
case 0x000b: /* rts */
CHECK_NOT_DELAY_SLOT
@@ -428,7 +408,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_S));
return;
case 0x0008: /* clrt */
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
+ tcg_gen_movi_i32(cpu_sr_t, 0);
return;
case 0x0038: /* ldtlb */
CHECK_PRIVILEGED
@@ -437,7 +417,7 @@ static void _decode_opc(DisasContext * ctx)
case 0x002b: /* rte */
CHECK_PRIVILEGED
CHECK_NOT_DELAY_SLOT
- tcg_gen_mov_i32(cpu_sr, cpu_ssr);
+ gen_write_sr(cpu_ssr);
tcg_gen_mov_i32(cpu_delayed_pc, cpu_spc);
ctx->flags |= DELAY_SLOT;
ctx->delayed_pc = (uint32_t) - 1;
@@ -446,7 +426,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_ori_i32(cpu_sr, cpu_sr, (1u << SR_S));
return;
case 0x0018: /* sett */
- tcg_gen_ori_i32(cpu_sr, cpu_sr, (1u << SR_T));
+ tcg_gen_movi_i32(cpu_sr_t, 1);
return;
case 0xfbfd: /* frchg */
tcg_gen_xori_i32(cpu_fpscr, cpu_fpscr, FPSCR_FR);
@@ -662,22 +642,17 @@ static void _decode_opc(DisasContext * ctx)
return;
case 0x300e: /* addc Rm,Rn */
{
- TCGv t0, t1, t2;
+ TCGv t0, t1;
t0 = tcg_temp_new();
- tcg_gen_andi_i32(t0, cpu_sr, (1u << SR_T));
t1 = tcg_temp_new();
- tcg_gen_add_i32(t1, REG(B7_4), REG(B11_8));
- tcg_gen_add_i32(t0, t0, t1);
- t2 = tcg_temp_new();
- tcg_gen_setcond_i32(TCG_COND_GTU, t2, REG(B11_8), t1);
- tcg_gen_setcond_i32(TCG_COND_GTU, t1, t1, t0);
- tcg_gen_or_i32(t1, t1, t2);
- tcg_temp_free(t2);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
- tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
- tcg_temp_free(t1);
- tcg_gen_mov_i32(REG(B11_8), t0);
+ tcg_gen_add_i32(t0, REG(B7_4), REG(B11_8));
+ tcg_gen_add_i32(t1, cpu_sr_t, t0);
+ tcg_gen_setcond_i32(TCG_COND_GTU, cpu_sr_t, REG(B11_8), t0);
+ tcg_gen_setcond_i32(TCG_COND_GTU, t0, t0, t1);
+ tcg_gen_or_i32(cpu_sr_t, cpu_sr_t, t0);
tcg_temp_free(t0);
+ tcg_gen_mov_i32(REG(B11_8), t1);
+ tcg_temp_free(t1);
}
return;
case 0x300f: /* addv Rm,Rn */
@@ -689,11 +664,9 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_xor_i32(t1, t0, REG(B11_8));
t2 = tcg_temp_new();
tcg_gen_xor_i32(t2, REG(B7_4), REG(B11_8));
- tcg_gen_andc_i32(t1, t1, t2);
+ tcg_gen_andc_i32(cpu_sr_t, t1, t2);
tcg_temp_free(t2);
- tcg_gen_shri_i32(t1, t1, 31);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
- tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
+ tcg_gen_shri_i32(cpu_sr_t, cpu_sr_t, 31);
tcg_temp_free(t1);
tcg_gen_mov_i32(REG(B7_4), t0);
tcg_temp_free(t0);
@@ -703,51 +676,45 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_and_i32(REG(B11_8), REG(B11_8), REG(B7_4));
return;
case 0x3000: /* cmp/eq Rm,Rn */
- gen_cmp(TCG_COND_EQ, REG(B7_4), REG(B11_8));
+ tcg_gen_setcond_i32(TCG_COND_EQ, cpu_sr_t, REG(B11_8), REG(B7_4));
return;
case 0x3003: /* cmp/ge Rm,Rn */
- gen_cmp(TCG_COND_GE, REG(B7_4), REG(B11_8));
+ tcg_gen_setcond_i32(TCG_COND_GE, cpu_sr_t, REG(B11_8), REG(B7_4));
return;
case 0x3007: /* cmp/gt Rm,Rn */
- gen_cmp(TCG_COND_GT, REG(B7_4), REG(B11_8));
+ tcg_gen_setcond_i32(TCG_COND_GT, cpu_sr_t, REG(B11_8), REG(B7_4));
return;
case 0x3006: /* cmp/hi Rm,Rn */
- gen_cmp(TCG_COND_GTU, REG(B7_4), REG(B11_8));
+ tcg_gen_setcond_i32(TCG_COND_GTU, cpu_sr_t, REG(B11_8), REG(B7_4));
return;
case 0x3002: /* cmp/hs Rm,Rn */
- gen_cmp(TCG_COND_GEU, REG(B7_4), REG(B11_8));
+ tcg_gen_setcond_i32(TCG_COND_GEU, cpu_sr_t, REG(B11_8), REG(B7_4));
return;
case 0x200c: /* cmp/str Rm,Rn */
{
TCGv cmp1 = tcg_temp_new();
TCGv cmp2 = tcg_temp_new();
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
tcg_gen_xor_i32(cmp1, REG(B7_4), REG(B11_8));
tcg_gen_andi_i32(cmp2, cmp1, 0xff000000);
- tcg_gen_setcondi_i32(TCG_COND_EQ, cmp2, cmp2, 0);
- tcg_gen_or_i32(cpu_sr, cpu_sr, cmp2);
+ tcg_gen_setcondi_i32(TCG_COND_EQ, cpu_sr_t, cmp2, 0);
tcg_gen_andi_i32(cmp2, cmp1, 0x00ff0000);
tcg_gen_setcondi_i32(TCG_COND_EQ, cmp2, cmp2, 0);
- tcg_gen_or_i32(cpu_sr, cpu_sr, cmp2);
+ tcg_gen_or_i32(cpu_sr_t, cpu_sr_t, cmp2);
tcg_gen_andi_i32(cmp2, cmp1, 0x0000ff00);
tcg_gen_setcondi_i32(TCG_COND_EQ, cmp2, cmp2, 0);
- tcg_gen_or_i32(cpu_sr, cpu_sr, cmp2);
+ tcg_gen_or_i32(cpu_sr_t, cpu_sr_t, cmp2);
tcg_gen_andi_i32(cmp2, cmp1, 0x000000ff);
tcg_gen_setcondi_i32(TCG_COND_EQ, cmp2, cmp2, 0);
- tcg_gen_or_i32(cpu_sr, cpu_sr, cmp2);
+ tcg_gen_or_i32(cpu_sr_t, cpu_sr_t, cmp2);
tcg_temp_free(cmp2);
tcg_temp_free(cmp1);
}
return;
case 0x2007: /* div0s Rm,Rn */
- {
- gen_copy_bit_i32(cpu_sr, SR_Q, REG(B11_8), 31); /* SR_Q */
- gen_copy_bit_i32(cpu_sr, SR_M, REG(B7_4), 31); /* SR_M */
- TCGv val = tcg_temp_new();
- tcg_gen_xor_i32(val, REG(B7_4), REG(B11_8));
- gen_copy_bit_i32(cpu_sr, SR_T, val, 31); /* SR_T */
- tcg_temp_free(val);
- }
+ gen_copy_bit_i32(cpu_sr, SR_Q, REG(B11_8), 31); /* SR_Q */
+ gen_copy_bit_i32(cpu_sr, SR_M, REG(B7_4), 31); /* SR_M */
+ tcg_gen_xor_i32(cpu_sr_t, REG(B7_4), REG(B11_8));
+ tcg_gen_shri_i32(cpu_sr_t, cpu_sr_t, 31); /* SR_T */
return;
case 0x3004: /* div1 Rm,Rn */
gen_helper_div1(REG(B11_8), cpu_env, REG(B7_4), REG(B11_8));
@@ -830,19 +797,13 @@ static void _decode_opc(DisasContext * ctx)
return;
case 0x600a: /* negc Rm,Rn */
{
- TCGv t0, t1;
- t0 = tcg_temp_new();
+ TCGv t0 = tcg_temp_new();
tcg_gen_neg_i32(t0, REG(B7_4));
- t1 = tcg_temp_new();
- tcg_gen_andi_i32(t1, cpu_sr, (1u << SR_T));
- tcg_gen_sub_i32(REG(B11_8), t0, t1);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
- tcg_gen_setcondi_i32(TCG_COND_GTU, t1, t0, 0);
- tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
- tcg_gen_setcond_i32(TCG_COND_GTU, t1, REG(B11_8), t0);
- tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
+ tcg_gen_sub_i32(REG(B11_8), t0, cpu_sr_t);
+ tcg_gen_setcondi_i32(TCG_COND_GTU, cpu_sr_t, t0, 0);
+ tcg_gen_setcond_i32(TCG_COND_GTU, t0, REG(B11_8), t0);
+ tcg_gen_or_i32(cpu_sr_t, cpu_sr_t, t0);
tcg_temp_free(t0);
- tcg_temp_free(t1);
}
return;
case 0x6007: /* not Rm,Rn */
@@ -923,17 +884,14 @@ static void _decode_opc(DisasContext * ctx)
{
TCGv t0, t1, t2;
t0 = tcg_temp_new();
- tcg_gen_andi_i32(t0, cpu_sr, (1u << SR_T));
t1 = tcg_temp_new();
tcg_gen_sub_i32(t1, REG(B11_8), REG(B7_4));
- tcg_gen_sub_i32(t0, t1, t0);
+ tcg_gen_sub_i32(t0, t1, cpu_sr_t);
t2 = tcg_temp_new();
tcg_gen_setcond_i32(TCG_COND_LTU, t2, REG(B11_8), t1);
tcg_gen_setcond_i32(TCG_COND_LTU, t1, t1, t0);
- tcg_gen_or_i32(t1, t1, t2);
+ tcg_gen_or_i32(cpu_sr_t, t1, t2);
tcg_temp_free(t2);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
- tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
tcg_temp_free(t1);
tcg_gen_mov_i32(REG(B11_8), t0);
tcg_temp_free(t0);
@@ -950,9 +908,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_xor_i32(t2, REG(B11_8), REG(B7_4));
tcg_gen_and_i32(t1, t1, t2);
tcg_temp_free(t2);
- tcg_gen_shri_i32(t1, t1, 31);
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
- tcg_gen_or_i32(cpu_sr, cpu_sr, t1);
+ tcg_gen_shri_i32(cpu_sr_t, t1, 31);
tcg_temp_free(t1);
tcg_gen_mov_i32(REG(B11_8), t0);
tcg_temp_free(t0);
@@ -962,7 +918,7 @@ static void _decode_opc(DisasContext * ctx)
{
TCGv val = tcg_temp_new();
tcg_gen_and_i32(val, REG(B7_4), REG(B11_8));
- gen_cmp_imm(TCG_COND_EQ, val, 0);
+ tcg_gen_setcondi_i32(TCG_COND_EQ, cpu_sr_t, val, 0);
tcg_temp_free(val);
}
return;
@@ -1213,7 +1169,7 @@ static void _decode_opc(DisasContext * ctx)
ctx->flags |= DELAY_SLOT_CONDITIONAL;
return;
case 0x8800: /* cmp/eq #imm,R0 */
- gen_cmp_imm(TCG_COND_EQ, REG(0), B7_0s);
+ tcg_gen_setcondi_i32(TCG_COND_EQ, cpu_sr_t, REG(0), B7_0s);
return;
case 0xc400: /* mov.b @(disp,GBR),R0 */
{
@@ -1329,7 +1285,7 @@ static void _decode_opc(DisasContext * ctx)
{
TCGv val = tcg_temp_new();
tcg_gen_andi_i32(val, REG(0), B7_0);
- gen_cmp_imm(TCG_COND_EQ, val, 0);
+ tcg_gen_setcondi_i32(TCG_COND_EQ, cpu_sr_t, val, 0);
tcg_temp_free(val);
}
return;
@@ -1339,7 +1295,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_add_i32(val, REG(0), cpu_gbr);
tcg_gen_qemu_ld_i32(val, val, ctx->memidx, MO_UB);
tcg_gen_andi_i32(val, val, B7_0);
- gen_cmp_imm(TCG_COND_EQ, val, 0);
+ tcg_gen_setcondi_i32(TCG_COND_EQ, cpu_sr_t, val, 0);
tcg_temp_free(val);
}
return;
@@ -1402,14 +1358,14 @@ static void _decode_opc(DisasContext * ctx)
ctx->delayed_pc = (uint32_t) - 1;
return;
case 0x4015: /* cmp/pl Rn */
- gen_cmp_imm(TCG_COND_GT, REG(B11_8), 0);
+ tcg_gen_setcondi_i32(TCG_COND_GT, cpu_sr_t, REG(B11_8), 0);
return;
case 0x4011: /* cmp/pz Rn */
- gen_cmp_imm(TCG_COND_GE, REG(B11_8), 0);
+ tcg_gen_setcondi_i32(TCG_COND_GE, cpu_sr_t, REG(B11_8), 0);
return;
case 0x4010: /* dt Rn */
tcg_gen_subi_i32(REG(B11_8), REG(B11_8), 1);
- gen_cmp_imm(TCG_COND_EQ, REG(B11_8), 0);
+ tcg_gen_setcondi_i32(TCG_COND_EQ, cpu_sr_t, REG(B11_8), 0);
return;
case 0x402b: /* jmp @Rn */
CHECK_NOT_DELAY_SLOT
@@ -1426,15 +1382,21 @@ static void _decode_opc(DisasContext * ctx)
return;
case 0x400e: /* ldc Rm,SR */
CHECK_PRIVILEGED
- tcg_gen_andi_i32(cpu_sr, REG(B11_8), 0x700083f3);
- ctx->bstate = BS_STOP;
+ {
+ TCGv val = tcg_temp_new();
+ tcg_gen_andi_i32(val, REG(B11_8), 0x700083f3);
+ gen_write_sr(val);
+ tcg_temp_free(val);
+ ctx->bstate = BS_STOP;
+ }
return;
case 0x4007: /* ldc.l @Rm+,SR */
CHECK_PRIVILEGED
{
TCGv val = tcg_temp_new();
tcg_gen_qemu_ld_i32(val, REG(B11_8), ctx->memidx, MO_TESL);
- tcg_gen_andi_i32(cpu_sr, val, 0x700083f3);
+ tcg_gen_andi_i32(val, val, 0x700083f3);
+ gen_write_sr(val);
tcg_temp_free(val);
tcg_gen_addi_i32(REG(B11_8), REG(B11_8), 4);
ctx->bstate = BS_STOP;
@@ -1442,15 +1404,18 @@ static void _decode_opc(DisasContext * ctx)
return;
case 0x0002: /* stc SR,Rn */
CHECK_PRIVILEGED
- tcg_gen_mov_i32(REG(B11_8), cpu_sr);
+ gen_read_sr(REG(B11_8));
return;
case 0x4003: /* stc SR,@-Rn */
CHECK_PRIVILEGED
{
TCGv addr = tcg_temp_new();
+ TCGv val = tcg_temp_new();
tcg_gen_subi_i32(addr, REG(B11_8), 4);
- tcg_gen_qemu_st_i32(cpu_sr, addr, ctx->memidx, MO_TEUL);
+ gen_read_sr(val);
+ tcg_gen_qemu_st_i32(val, addr, ctx->memidx, MO_TEUL);
tcg_gen_mov_i32(REG(B11_8), addr);
+ tcg_temp_free(val);
tcg_temp_free(addr);
}
return;
@@ -1548,7 +1513,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_addi_i32(REG(B11_8), REG(B11_8), 4);
return;
case 0x0029: /* movt Rn */
- tcg_gen_andi_i32(REG(B11_8), cpu_sr, (1u << SR_T));
+ tcg_gen_mov_i32(REG(B11_8), cpu_sr_t);
return;
case 0x0073:
/* MOVCO.L
@@ -1558,8 +1523,7 @@ static void _decode_opc(DisasContext * ctx)
*/
if (ctx->features & SH_FEATURE_SH4A) {
TCGLabel *label = gen_new_label();
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
- tcg_gen_or_i32(cpu_sr, cpu_sr, cpu_ldst);
+ tcg_gen_mov_i32(cpu_sr, cpu_ldst);
tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_ldst, 0, label);
tcg_gen_qemu_st_i32(REG(0), REG(B11_8), ctx->memidx, MO_TEUL);
gen_set_label(label);
@@ -1612,42 +1576,42 @@ static void _decode_opc(DisasContext * ctx)
case 0x4024: /* rotcl Rn */
{
TCGv tmp = tcg_temp_new();
- tcg_gen_mov_i32(tmp, cpu_sr);
- gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 31);
+ tcg_gen_mov_i32(tmp, cpu_sr_t);
+ tcg_gen_shri_i32(cpu_sr_t, REG(B11_8), 31);
tcg_gen_shli_i32(REG(B11_8), REG(B11_8), 1);
- gen_copy_bit_i32(REG(B11_8), SR_T, tmp, 0);
+ tcg_gen_or_i32(REG(B11_8), REG(B11_8), tmp);
tcg_temp_free(tmp);
}
return;
case 0x4025: /* rotcr Rn */
{
TCGv tmp = tcg_temp_new();
- tcg_gen_mov_i32(tmp, cpu_sr);
- gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 0);
+ tcg_gen_shli_i32(tmp, cpu_sr_t, 31);
+ tcg_gen_andi_i32(cpu_sr_t, REG(B11_8), 1);
tcg_gen_shri_i32(REG(B11_8), REG(B11_8), 1);
- gen_copy_bit_i32(REG(B11_8), 31, tmp, 0);
+ tcg_gen_or_i32(REG(B11_8), REG(B11_8), tmp);
tcg_temp_free(tmp);
}
return;
case 0x4004: /* rotl Rn */
tcg_gen_rotli_i32(REG(B11_8), REG(B11_8), 1);
- gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 0);
+ tcg_gen_andi_i32(cpu_sr_t, REG(B11_8), 0);
return;
case 0x4005: /* rotr Rn */
- gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 0);
+ tcg_gen_andi_i32(cpu_sr_t, REG(B11_8), 0);
tcg_gen_rotri_i32(REG(B11_8), REG(B11_8), 1);
return;
case 0x4000: /* shll Rn */
case 0x4020: /* shal Rn */
- gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 31);
+ tcg_gen_shri_i32(cpu_sr_t, REG(B11_8), 31);
tcg_gen_shli_i32(REG(B11_8), REG(B11_8), 1);
return;
case 0x4021: /* shar Rn */
- gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 0);
+ tcg_gen_andi_i32(cpu_sr_t, REG(B11_8), 1);
tcg_gen_sari_i32(REG(B11_8), REG(B11_8), 1);
return;
case 0x4001: /* shlr Rn */
- gen_copy_bit_i32(cpu_sr, SR_T, REG(B11_8), 0);
+ tcg_gen_andi_i32(cpu_sr_t, REG(B11_8), 1);
tcg_gen_shri_i32(REG(B11_8), REG(B11_8), 1);
return;
case 0x4008: /* shll2 Rn */
@@ -1675,7 +1639,7 @@ static void _decode_opc(DisasContext * ctx)
tcg_gen_mov_i32(addr, REG(B11_8));
val = tcg_temp_local_new();
tcg_gen_qemu_ld_i32(val, addr, ctx->memidx, MO_UB);
- gen_cmp_imm(TCG_COND_EQ, val, 0);
+ tcg_gen_setcondi_i32(TCG_COND_EQ, cpu_sr_t, val, 0);
tcg_gen_ori_i32(val, val, 0x80);
tcg_gen_qemu_st_i32(val, addr, ctx->memidx, MO_UB);
tcg_temp_free(val);
--
2.1.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [Qemu-devel] [PATCH v3 3/8] target-sh4: optimize addc using add2
2015-05-24 23:37 [Qemu-devel] [PATCH v3 0/8] target-sh4: optimizations and cleanups Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 1/8] target-sh4: use bit number for SR constants Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 2/8] target-sh4: Split out T from SR Aurelien Jarno
@ 2015-05-24 23:37 ` Aurelien Jarno
2015-06-04 5:03 ` Richard Henderson
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 4/8] target-sh4: optimize subc using sub2 Aurelien Jarno
` (4 subsequent siblings)
7 siblings, 1 reply; 16+ messages in thread
From: Aurelien Jarno @ 2015-05-24 23:37 UTC (permalink / raw)
To: qemu-devel; +Cc: Aurelien Jarno
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
---
target-sh4/translate.c | 14 ++++++--------
1 file changed, 6 insertions(+), 8 deletions(-)
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index f9bc24c..a7a8f39 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -642,17 +642,15 @@ static void _decode_opc(DisasContext * ctx)
return;
case 0x300e: /* addc Rm,Rn */
{
- TCGv t0, t1;
- t0 = tcg_temp_new();
+ TCGv t0, t1, t2;
+ t0 = tcg_const_tl(0);
t1 = tcg_temp_new();
- tcg_gen_add_i32(t0, REG(B7_4), REG(B11_8));
- tcg_gen_add_i32(t1, cpu_sr_t, t0);
- tcg_gen_setcond_i32(TCG_COND_GTU, cpu_sr_t, REG(B11_8), t0);
- tcg_gen_setcond_i32(TCG_COND_GTU, t0, t0, t1);
- tcg_gen_or_i32(cpu_sr_t, cpu_sr_t, t0);
+ t2 = tcg_temp_new();
+ tcg_gen_add2_i32(t1, t2, REG(B11_8), t0, REG(B7_4), t0);
+ tcg_gen_add2_i32(REG(B11_8), cpu_sr_t, t1, t2, cpu_sr_t, t0);
tcg_temp_free(t0);
- tcg_gen_mov_i32(REG(B11_8), t1);
tcg_temp_free(t1);
+ tcg_temp_free(t2);
}
return;
case 0x300f: /* addv Rm,Rn */
--
2.1.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [Qemu-devel] [PATCH v3 4/8] target-sh4: optimize subc using sub2
2015-05-24 23:37 [Qemu-devel] [PATCH v3 0/8] target-sh4: optimizations and cleanups Aurelien Jarno
` (2 preceding siblings ...)
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 3/8] target-sh4: optimize addc using add2 Aurelien Jarno
@ 2015-05-24 23:37 ` Aurelien Jarno
2015-06-04 5:05 ` Richard Henderson
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 5/8] target-sh4: optimize negc using add2 and sub2 Aurelien Jarno
` (3 subsequent siblings)
7 siblings, 1 reply; 16+ messages in thread
From: Aurelien Jarno @ 2015-05-24 23:37 UTC (permalink / raw)
To: qemu-devel; +Cc: Aurelien Jarno
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
---
target-sh4/translate.c | 15 ++++++---------
1 file changed, 6 insertions(+), 9 deletions(-)
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index a7a8f39..d5b448e 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -881,18 +881,15 @@ static void _decode_opc(DisasContext * ctx)
case 0x300a: /* subc Rm,Rn */
{
TCGv t0, t1, t2;
- t0 = tcg_temp_new();
+ t0 = tcg_const_tl(0);
t1 = tcg_temp_new();
- tcg_gen_sub_i32(t1, REG(B11_8), REG(B7_4));
- tcg_gen_sub_i32(t0, t1, cpu_sr_t);
t2 = tcg_temp_new();
- tcg_gen_setcond_i32(TCG_COND_LTU, t2, REG(B11_8), t1);
- tcg_gen_setcond_i32(TCG_COND_LTU, t1, t1, t0);
- tcg_gen_or_i32(cpu_sr_t, t1, t2);
- tcg_temp_free(t2);
- tcg_temp_free(t1);
- tcg_gen_mov_i32(REG(B11_8), t0);
+ tcg_gen_sub2_i32(t1, t2, REG(B11_8), t0, REG(B7_4), t0);
+ tcg_gen_sub2_i32(REG(B11_8), cpu_sr_t, t1, t2, cpu_sr_t, t0);
+ tcg_gen_andi_i32(cpu_sr_t, cpu_sr_t, 1);
tcg_temp_free(t0);
+ tcg_temp_free(t1);
+ tcg_temp_free(t2);
}
return;
case 0x300b: /* subv Rm,Rn */
--
2.1.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [Qemu-devel] [PATCH v3 5/8] target-sh4: optimize negc using add2 and sub2
2015-05-24 23:37 [Qemu-devel] [PATCH v3 0/8] target-sh4: optimizations and cleanups Aurelien Jarno
` (3 preceding siblings ...)
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 4/8] target-sh4: optimize subc using sub2 Aurelien Jarno
@ 2015-05-24 23:37 ` Aurelien Jarno
2015-06-04 5:07 ` Richard Henderson
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 6/8] target-sh4: split out Q and M from of SR and optimize div1 Aurelien Jarno
` (2 subsequent siblings)
7 siblings, 1 reply; 16+ messages in thread
From: Aurelien Jarno @ 2015-05-24 23:37 UTC (permalink / raw)
To: qemu-devel; +Cc: Aurelien Jarno
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
---
target-sh4/translate.c | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index d5b448e..250632a 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -795,12 +795,12 @@ static void _decode_opc(DisasContext * ctx)
return;
case 0x600a: /* negc Rm,Rn */
{
- TCGv t0 = tcg_temp_new();
- tcg_gen_neg_i32(t0, REG(B7_4));
- tcg_gen_sub_i32(REG(B11_8), t0, cpu_sr_t);
- tcg_gen_setcondi_i32(TCG_COND_GTU, cpu_sr_t, t0, 0);
- tcg_gen_setcond_i32(TCG_COND_GTU, t0, REG(B11_8), t0);
- tcg_gen_or_i32(cpu_sr_t, cpu_sr_t, t0);
+ TCGv t0 = tcg_const_i32(0);
+ tcg_gen_add2_i32(REG(B11_8), cpu_sr_t,
+ REG(B7_4), t0, cpu_sr_t, t0);
+ tcg_gen_sub2_i32(REG(B11_8), cpu_sr_t,
+ t0, t0, REG(B11_8), cpu_sr_t);
+ tcg_gen_andi_i32(cpu_sr_t, cpu_sr_t, 1);
tcg_temp_free(t0);
}
return;
--
2.1.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [Qemu-devel] [PATCH v3 6/8] target-sh4: split out Q and M from of SR and optimize div1
2015-05-24 23:37 [Qemu-devel] [PATCH v3 0/8] target-sh4: optimizations and cleanups Aurelien Jarno
` (4 preceding siblings ...)
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 5/8] target-sh4: optimize negc using add2 and sub2 Aurelien Jarno
@ 2015-05-24 23:37 ` Aurelien Jarno
2015-06-04 5:31 ` Richard Henderson
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 7/8] target-sh4: factorize fmov implementation Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 8/8] target-sh4: remove dead code Aurelien Jarno
7 siblings, 1 reply; 16+ messages in thread
From: Aurelien Jarno @ 2015-05-24 23:37 UTC (permalink / raw)
To: qemu-devel; +Cc: Aurelien Jarno
Splitting Q and M out of SR, it's possible to optimize div1 by using
TCG code instead of an helper.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
---
target-sh4/cpu.h | 12 +++--
target-sh4/helper.h | 1 -
target-sh4/op_helper.c | 118 -------------------------------------------------
target-sh4/translate.c | 70 ++++++++++++++++++++++++-----
4 files changed, 69 insertions(+), 132 deletions(-)
diff --git a/target-sh4/cpu.h b/target-sh4/cpu.h
index a308c53..4a027a6 100644
--- a/target-sh4/cpu.h
+++ b/target-sh4/cpu.h
@@ -139,6 +139,8 @@ typedef struct CPUSH4State {
uint32_t gregs[24]; /* general registers */
float32 fregs[32]; /* floating point registers */
uint32_t sr; /* status register (with T split out) */
+ uint32_t sr_m; /* M bit of status register */
+ uint32_t sr_q; /* Q bit of status register */
uint32_t sr_t; /* T bit of status register */
uint32_t ssr; /* saved status register */
uint32_t spc; /* saved program counter */
@@ -334,13 +336,17 @@ static inline int cpu_ptel_pr (uint32_t ptel)
static inline target_ulong cpu_read_sr(CPUSH4State *env)
{
- return env->sr | (env->sr_t << SR_T);
+ return env->sr | (env->sr_m << SR_M) |
+ (env->sr_q << SR_Q) |
+ (env->sr_t << SR_T);
}
static inline void cpu_write_sr(CPUSH4State *env, target_ulong sr)
{
- env->sr_t = sr & (1u << SR_T);
- env->sr = sr & ~(1u << SR_T);
+ env->sr_m = (sr >> SR_M) & 1;
+ env->sr_q = (sr >> SR_Q) & 1;
+ env->sr_t = (sr >> SR_T) & 1;
+ env->sr = sr & ~((1u << SR_M) | (1u << SR_Q) | (1u << SR_T));
}
static inline void cpu_get_tb_cpu_state(CPUSH4State *env, target_ulong *pc,
diff --git a/target-sh4/helper.h b/target-sh4/helper.h
index 3b5c436..c9bc407 100644
--- a/target-sh4/helper.h
+++ b/target-sh4/helper.h
@@ -11,7 +11,6 @@ DEF_HELPER_3(movcal, void, env, i32, i32)
DEF_HELPER_1(discard_movcal_backup, void, env)
DEF_HELPER_2(ocbi, void, env, i32)
-DEF_HELPER_3(div1, i32, env, i32, i32)
DEF_HELPER_3(macl, void, env, i32, i32)
DEF_HELPER_3(macw, void, env, i32, i32)
diff --git a/target-sh4/op_helper.c b/target-sh4/op_helper.c
index 524d7f6..cbc11ae 100644
--- a/target-sh4/op_helper.c
+++ b/target-sh4/op_helper.c
@@ -156,124 +156,6 @@ void helper_ocbi(CPUSH4State *env, uint32_t address)
}
}
-#define T (env->sr_t)
-#define Q (env->sr & (1u << SR_Q) ? 1 : 0)
-#define M (env->sr & (1u << SR_M) ? 1 : 0)
-#define SETT (env->sr_t = 1)
-#define CLRT (env->sr_t = 0)
-#define SETQ (env->sr |= (1u << SR_Q))
-#define CLRQ (env->sr &= ~(1u << SR_Q))
-#define SETM (env->sr |= (1u << SR_M))
-#define CLRM (env->sr &= ~(1u << SR_M))
-
-uint32_t helper_div1(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
-{
- uint32_t tmp0, tmp2;
- uint8_t old_q, tmp1 = 0xff;
-
- //printf("div1 arg0=0x%08x arg1=0x%08x M=%d Q=%d T=%d\n", arg0, arg1, M, Q, T);
- old_q = Q;
- if ((0x80000000 & arg1) != 0)
- SETQ;
- else
- CLRQ;
- tmp2 = arg0;
- arg1 <<= 1;
- arg1 |= T;
- switch (old_q) {
- case 0:
- switch (M) {
- case 0:
- tmp0 = arg1;
- arg1 -= tmp2;
- tmp1 = arg1 > tmp0;
- switch (Q) {
- case 0:
- if (tmp1)
- SETQ;
- else
- CLRQ;
- break;
- case 1:
- if (tmp1 == 0)
- SETQ;
- else
- CLRQ;
- break;
- }
- break;
- case 1:
- tmp0 = arg1;
- arg1 += tmp2;
- tmp1 = arg1 < tmp0;
- switch (Q) {
- case 0:
- if (tmp1 == 0)
- SETQ;
- else
- CLRQ;
- break;
- case 1:
- if (tmp1)
- SETQ;
- else
- CLRQ;
- break;
- }
- break;
- }
- break;
- case 1:
- switch (M) {
- case 0:
- tmp0 = arg1;
- arg1 += tmp2;
- tmp1 = arg1 < tmp0;
- switch (Q) {
- case 0:
- if (tmp1)
- SETQ;
- else
- CLRQ;
- break;
- case 1:
- if (tmp1 == 0)
- SETQ;
- else
- CLRQ;
- break;
- }
- break;
- case 1:
- tmp0 = arg1;
- arg1 -= tmp2;
- tmp1 = arg1 > tmp0;
- switch (Q) {
- case 0:
- if (tmp1 == 0)
- SETQ;
- else
- CLRQ;
- break;
- case 1:
- if (tmp1)
- SETQ;
- else
- CLRQ;
- break;
- }
- break;
- }
- break;
- }
- if (Q == M)
- SETT;
- else
- CLRT;
- //printf("Output: arg1=0x%08x M=%d Q=%d T=%d\n", arg1, M, Q, T);
- return arg1;
-}
-
void helper_macl(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
{
int64_t res;
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index 250632a..daea268 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -62,7 +62,8 @@ enum {
/* global register indexes */
static TCGv_ptr cpu_env;
static TCGv cpu_gregs[24];
-static TCGv cpu_pc, cpu_sr, cpu_sr_t, cpu_ssr, cpu_spc, cpu_gbr;
+static TCGv cpu_sr, cpu_sr_m, cpu_sr_q, cpu_sr_t;
+static TCGv cpu_pc, cpu_ssr, cpu_spc, cpu_gbr;
static TCGv cpu_vbr, cpu_sgr, cpu_dbr, cpu_mach, cpu_macl;
static TCGv cpu_pr, cpu_fpscr, cpu_fpul, cpu_ldst;
static TCGv cpu_fregs[32];
@@ -110,6 +111,10 @@ void sh4_translate_init(void)
offsetof(CPUSH4State, pc), "PC");
cpu_sr = tcg_global_mem_new_i32(TCG_AREG0,
offsetof(CPUSH4State, sr), "SR");
+ cpu_sr_m = tcg_global_mem_new_i32(TCG_AREG0,
+ offsetof(CPUSH4State, sr_m), "SR_M");
+ cpu_sr_q = tcg_global_mem_new_i32(TCG_AREG0,
+ offsetof(CPUSH4State, sr_q), "SR_Q");
cpu_sr_t = tcg_global_mem_new_i32(TCG_AREG0,
offsetof(CPUSH4State, sr_t), "SR_T");
cpu_ssr = tcg_global_mem_new_i32(TCG_AREG0,
@@ -178,13 +183,26 @@ void superh_cpu_dump_state(CPUState *cs, FILE *f,
}
static void gen_read_sr(TCGv dst)
{
- tcg_gen_or_i32(dst, cpu_sr, cpu_sr_t);
+ TCGv t0 = tcg_temp_new();
+ tcg_gen_shli_i32(t0, cpu_sr_q, SR_Q);
+ tcg_gen_or_i32(dst, dst, t0);
+ tcg_gen_shli_i32(t0, cpu_sr_m, SR_M);
+ tcg_gen_or_i32(dst, dst, t0);
+ tcg_gen_shli_i32(t0, cpu_sr_t, SR_T);
+ tcg_gen_or_i32(dst, cpu_sr, t0);
+ tcg_temp_free_i32(t0);
}
static void gen_write_sr(TCGv src)
{
- tcg_gen_andi_i32(cpu_sr, src, ~(1u << SR_T));
- tcg_gen_andi_i32(cpu_sr_t, src, (1u << SR_T));
+ tcg_gen_andi_i32(cpu_sr, src,
+ ~((1u << SR_Q) | (1u << SR_M) | (1u << SR_T)));
+ tcg_gen_shri_i32(cpu_sr_q, src, SR_Q);
+ tcg_gen_andi_i32(cpu_sr_q, cpu_sr_q, 1);
+ tcg_gen_shri_i32(cpu_sr_m, src, SR_M);
+ tcg_gen_andi_i32(cpu_sr_m, cpu_sr_m, 1);
+ tcg_gen_shri_i32(cpu_sr_t, src, SR_T);
+ tcg_gen_andi_i32(cpu_sr_t, cpu_sr_t, 1);
}
static void gen_goto_tb(DisasContext * ctx, int n, target_ulong dest)
@@ -391,7 +409,8 @@ static void _decode_opc(DisasContext * ctx)
switch (ctx->opcode) {
case 0x0019: /* div0u */
- tcg_gen_andi_i32(cpu_sr, cpu_sr, ~((1u << SR_M) | (1u << SR_Q)));
+ tcg_gen_movi_i32(cpu_sr_m, 0);
+ tcg_gen_movi_i32(cpu_sr_q, 0);
tcg_gen_movi_i32(cpu_sr_t, 0);
return;
case 0x000b: /* rts */
@@ -709,13 +728,44 @@ static void _decode_opc(DisasContext * ctx)
}
return;
case 0x2007: /* div0s Rm,Rn */
- gen_copy_bit_i32(cpu_sr, SR_Q, REG(B11_8), 31); /* SR_Q */
- gen_copy_bit_i32(cpu_sr, SR_M, REG(B7_4), 31); /* SR_M */
- tcg_gen_xor_i32(cpu_sr_t, REG(B7_4), REG(B11_8));
- tcg_gen_shri_i32(cpu_sr_t, cpu_sr_t, 31); /* SR_T */
+ tcg_gen_shri_i32(cpu_sr_q, REG(B11_8), 31); /* SR_Q */
+ tcg_gen_shri_i32(cpu_sr_m, REG(B7_4), 31); /* SR_M */
+ tcg_gen_xor_i32(cpu_sr_t, cpu_sr_q, cpu_sr_m); /* SR_T */
return;
case 0x3004: /* div1 Rm,Rn */
- gen_helper_div1(REG(B11_8), cpu_env, REG(B7_4), REG(B11_8));
+ {
+ TCGv t0 = tcg_temp_new();
+ TCGv t1 = tcg_temp_new();
+ TCGv t2 = tcg_temp_new();
+ TCGv zero = tcg_const_i32(0);
+
+ /* shift left arg1, saving the bit being pushed out and inserting
+ T on the right */
+ tcg_gen_shri_i32(t0, REG(B11_8), 31);
+ tcg_gen_shli_i32(REG(B11_8), REG(B11_8), 1);
+ tcg_gen_or_i32(REG(B11_8), REG(B11_8), cpu_sr_t);
+
+ /* Add or subtract arg0 from arg1 depending if Q == M. To avoid
+ using 64-bit temps, we compute arg0's high part from q ^ m, so
+ that it is 0x00000000 when adding the value or 0xffffffff when
+ subtracting it. */
+ tcg_gen_xor_i32(t1, cpu_sr_q, cpu_sr_m);
+ tcg_gen_subi_i32(t1, t1, 1);
+ tcg_gen_neg_i32(t2, REG(B7_4));
+ tcg_gen_movcond_i32(TCG_COND_EQ, t2, t1, zero, REG(B7_4), t2);
+ tcg_gen_add2_i32(REG(B11_8), t1, REG(B11_8), zero, t2, t1);
+
+ /* compute T and Q depending on carry */
+ tcg_gen_andi_i32(t1, t1, 1);
+ tcg_gen_xor_i32(t1, t1, t0);
+ tcg_gen_xori_i32(cpu_sr_t, t1, 1);
+ tcg_gen_xor_i32(cpu_sr_q, cpu_sr_m, t1);
+
+ tcg_temp_free(zero);
+ tcg_temp_free(t2);
+ tcg_temp_free(t1);
+ tcg_temp_free(t0);
+ }
return;
case 0x300d: /* dmuls.l Rm,Rn */
tcg_gen_muls2_i32(cpu_macl, cpu_mach, REG(B7_4), REG(B11_8));
--
2.1.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [Qemu-devel] [PATCH v3 7/8] target-sh4: factorize fmov implementation
2015-05-24 23:37 [Qemu-devel] [PATCH v3 0/8] target-sh4: optimizations and cleanups Aurelien Jarno
` (5 preceding siblings ...)
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 6/8] target-sh4: split out Q and M from of SR and optimize div1 Aurelien Jarno
@ 2015-05-24 23:37 ` Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 8/8] target-sh4: remove dead code Aurelien Jarno
7 siblings, 0 replies; 16+ messages in thread
From: Aurelien Jarno @ 2015-05-24 23:37 UTC (permalink / raw)
To: qemu-devel; +Cc: Aurelien Jarno
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
---
target-sh4/translate.c | 18 ++++++------------
1 file changed, 6 insertions(+), 12 deletions(-)
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index daea268..882c8d8 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -1029,24 +1029,18 @@ static void _decode_opc(DisasContext * ctx)
return;
case 0xf00b: /* fmov {F,D,X}Rm,@-Rn - FPSCR: Nothing */
CHECK_FPU_ENABLED
+ const int fr = XREG(B7_4);
+ TCGv addr = tcg_temp_new_i32();
+ tcg_gen_subi_i32(addr, REG(B11_8), 4);
if (ctx->flags & FPSCR_SZ) {
- TCGv addr = tcg_temp_new_i32();
- int fr = XREG(B7_4);
- tcg_gen_subi_i32(addr, REG(B11_8), 4);
tcg_gen_qemu_st_i32(cpu_fregs[fr+1], addr, ctx->memidx, MO_TEUL);
tcg_gen_subi_i32(addr, addr, 4);
tcg_gen_qemu_st_i32(cpu_fregs[fr], addr, ctx->memidx, MO_TEUL);
- tcg_gen_mov_i32(REG(B11_8), addr);
- tcg_temp_free(addr);
} else {
- TCGv addr;
- addr = tcg_temp_new_i32();
- tcg_gen_subi_i32(addr, REG(B11_8), 4);
- tcg_gen_qemu_st_i32(cpu_fregs[FREG(B7_4)], addr,
- ctx->memidx, MO_TEUL);
- tcg_gen_mov_i32(REG(B11_8), addr);
- tcg_temp_free(addr);
+ tcg_gen_qemu_st_i32(cpu_fregs[fr], addr, ctx->memidx, MO_TEUL);
}
+ tcg_gen_mov_i32(REG(B11_8), addr);
+ tcg_temp_free(addr);
return;
case 0xf006: /* fmov @(R0,Rm),{F,D,X}Rm - FPSCR: Nothing */
CHECK_FPU_ENABLED
--
2.1.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [Qemu-devel] [PATCH v3 8/8] target-sh4: remove dead code
2015-05-24 23:37 [Qemu-devel] [PATCH v3 0/8] target-sh4: optimizations and cleanups Aurelien Jarno
` (6 preceding siblings ...)
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 7/8] target-sh4: factorize fmov implementation Aurelien Jarno
@ 2015-05-24 23:37 ` Aurelien Jarno
7 siblings, 0 replies; 16+ messages in thread
From: Aurelien Jarno @ 2015-05-24 23:37 UTC (permalink / raw)
To: qemu-devel; +Cc: Aurelien Jarno
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
---
target-sh4/translate.c | 1 -
1 file changed, 1 deletion(-)
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index 882c8d8..38d4a6f 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -18,7 +18,6 @@
*/
#define DEBUG_DISAS
-//#define SH4_SINGLE_STEP
#include "cpu.h"
#include "disas/disas.h"
--
2.1.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* Re: [Qemu-devel] [PATCH v3 2/8] target-sh4: Split out T from SR
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 2/8] target-sh4: Split out T from SR Aurelien Jarno
@ 2015-06-04 5:01 ` Richard Henderson
0 siblings, 0 replies; 16+ messages in thread
From: Richard Henderson @ 2015-06-04 5:01 UTC (permalink / raw)
To: Aurelien Jarno, qemu-devel
On 05/24/2015 04:37 PM, Aurelien Jarno wrote:
> @@ -174,6 +176,16 @@ void superh_cpu_dump_state(CPUState *cs, FILE *f,
> env->delayed_pc);
> }
> }
> +static void gen_read_sr(TCGv dst)
> +{
> + tcg_gen_or_i32(dst, cpu_sr, cpu_sr_t);
> +}
Watch the spacing.
> /* MOVCO.L
> @@ -1558,8 +1523,7 @@ static void _decode_opc(DisasContext * ctx)
> */
> if (ctx->features & SH_FEATURE_SH4A) {
> TCGLabel *label = gen_new_label();
> - tcg_gen_andi_i32(cpu_sr, cpu_sr, ~(1u << SR_T));
> - tcg_gen_or_i32(cpu_sr, cpu_sr, cpu_ldst);
> + tcg_gen_mov_i32(cpu_sr, cpu_ldst);
Move to cpu_sr_t.
r~
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [Qemu-devel] [PATCH v3 3/8] target-sh4: optimize addc using add2
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 3/8] target-sh4: optimize addc using add2 Aurelien Jarno
@ 2015-06-04 5:03 ` Richard Henderson
2015-06-04 10:54 ` Paolo Bonzini
0 siblings, 1 reply; 16+ messages in thread
From: Richard Henderson @ 2015-06-04 5:03 UTC (permalink / raw)
To: Aurelien Jarno, qemu-devel
On 05/24/2015 04:37 PM, Aurelien Jarno wrote:
> - TCGv t0, t1;
> - t0 = tcg_temp_new();
> + TCGv t0, t1, t2;
> + t0 = tcg_const_tl(0);
> t1 = tcg_temp_new();
> - tcg_gen_add_i32(t0, REG(B7_4), REG(B11_8));
> - tcg_gen_add_i32(t1, cpu_sr_t, t0);
> - tcg_gen_setcond_i32(TCG_COND_GTU, cpu_sr_t, REG(B11_8), t0);
> - tcg_gen_setcond_i32(TCG_COND_GTU, t0, t0, t1);
> - tcg_gen_or_i32(cpu_sr_t, cpu_sr_t, t0);
> + t2 = tcg_temp_new();
> + tcg_gen_add2_i32(t1, t2, REG(B11_8), t0, REG(B7_4), t0);
> + tcg_gen_add2_i32(REG(B11_8), cpu_sr_t, t1, t2, cpu_sr_t, t0);
Swap these two adds and you don't need t2. You can consume sr_t immediately
and start producing it in the same go.
r~
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [Qemu-devel] [PATCH v3 4/8] target-sh4: optimize subc using sub2
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 4/8] target-sh4: optimize subc using sub2 Aurelien Jarno
@ 2015-06-04 5:05 ` Richard Henderson
0 siblings, 0 replies; 16+ messages in thread
From: Richard Henderson @ 2015-06-04 5:05 UTC (permalink / raw)
To: Aurelien Jarno, qemu-devel
On 05/24/2015 04:37 PM, Aurelien Jarno wrote:
> + tcg_gen_sub2_i32(t1, t2, REG(B11_8), t0, REG(B7_4), t0);
> + tcg_gen_sub2_i32(REG(B11_8), cpu_sr_t, t1, t2, cpu_sr_t, t0);
Similarly.
r~
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [Qemu-devel] [PATCH v3 5/8] target-sh4: optimize negc using add2 and sub2
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 5/8] target-sh4: optimize negc using add2 and sub2 Aurelien Jarno
@ 2015-06-04 5:07 ` Richard Henderson
0 siblings, 0 replies; 16+ messages in thread
From: Richard Henderson @ 2015-06-04 5:07 UTC (permalink / raw)
To: Aurelien Jarno, qemu-devel
On 05/24/2015 04:37 PM, Aurelien Jarno wrote:
> Signed-off-by: Aurelien Jarno<aurelien@aurel32.net>
> ---
> target-sh4/translate.c | 12 ++++++------
> 1 file changed, 6 insertions(+), 6 deletions(-)
Reviewed-by: Richard Henderson <rth@twiddle.net>
r~
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [Qemu-devel] [PATCH v3 6/8] target-sh4: split out Q and M from of SR and optimize div1
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 6/8] target-sh4: split out Q and M from of SR and optimize div1 Aurelien Jarno
@ 2015-06-04 5:31 ` Richard Henderson
0 siblings, 0 replies; 16+ messages in thread
From: Richard Henderson @ 2015-06-04 5:31 UTC (permalink / raw)
To: Aurelien Jarno, qemu-devel
On 05/24/2015 04:37 PM, Aurelien Jarno wrote:
> Splitting Q and M out of SR, it's possible to optimize div1 by using
> TCG code instead of an helper.
>
> Signed-off-by: Aurelien Jarno<aurelien@aurel32.net>
> ---
> target-sh4/cpu.h | 12 +++--
> target-sh4/helper.h | 1 -
> target-sh4/op_helper.c | 118 -------------------------------------------------
> target-sh4/translate.c | 70 ++++++++++++++++++++++++-----
> 4 files changed, 69 insertions(+), 132 deletions(-)
Reviewed-by: Richard Henderson <rth@twiddle.net>
r~
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [Qemu-devel] [PATCH v3 3/8] target-sh4: optimize addc using add2
2015-06-04 5:03 ` Richard Henderson
@ 2015-06-04 10:54 ` Paolo Bonzini
2015-06-04 16:08 ` Aurelien Jarno
0 siblings, 1 reply; 16+ messages in thread
From: Paolo Bonzini @ 2015-06-04 10:54 UTC (permalink / raw)
To: Richard Henderson, Aurelien Jarno, qemu-devel
On 04/06/2015 07:03, Richard Henderson wrote:
>> + tcg_gen_add2_i32(t1, t2, REG(B11_8), t0, REG(B7_4), t0);
>> + tcg_gen_add2_i32(REG(B11_8), cpu_sr_t, t1, t2, cpu_sr_t,
>> t0);
>
> Swap these two adds and you don't need t2. You can consume sr_t
> immediately and start producing it in the same go.
Could TCG do some kind of intra-basic-block live range splitting? In
this case, the new sr_t could be allocated to a different register than
the old one, saving one instruction on 2-address targets.
The pseudocode below uses "dest, src" operand order:
// add2(t1, cpu_sr_t, cpu_sr_t, t0, REG(B7_4), t0)
add sr_t_in, B7_4 // instead of mov t1, sr_t; add t1, B7_4
mov sr_t_out, 0
adc sr_t_out, 0 // cout(B7_r + sr_t_in)
// add2(REG(B11_8), cpu_sr_t, t1, cpu_sr_t, REG(B11_8), t0)
add B11_8, sr_t_in // B11_8 + B7_4 + sr_t_in
adc sr_t_out, 0 // cout(B11_8 + B7_4 + sr_t_in)
Paolo
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [Qemu-devel] [PATCH v3 3/8] target-sh4: optimize addc using add2
2015-06-04 10:54 ` Paolo Bonzini
@ 2015-06-04 16:08 ` Aurelien Jarno
0 siblings, 0 replies; 16+ messages in thread
From: Aurelien Jarno @ 2015-06-04 16:08 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: qemu-devel, Richard Henderson
On 2015-06-04 12:54, Paolo Bonzini wrote:
>
>
> On 04/06/2015 07:03, Richard Henderson wrote:
> >> + tcg_gen_add2_i32(t1, t2, REG(B11_8), t0, REG(B7_4), t0);
> >> + tcg_gen_add2_i32(REG(B11_8), cpu_sr_t, t1, t2, cpu_sr_t,
> >> t0);
> >
> > Swap these two adds and you don't need t2. You can consume sr_t
> > immediately and start producing it in the same go.
>
> Could TCG do some kind of intra-basic-block live range splitting? In
> this case, the new sr_t could be allocated to a different register than
> the old one, saving one instruction on 2-address targets.
TCG doesn't use a fixed register to a temp, so it's kind of difficult to
know, but let's say it more or less do that (see below). On the other
hand it is really bad at handling the constant in that case.
> The pseudocode below uses "dest, src" operand order:
>
> // add2(t1, cpu_sr_t, cpu_sr_t, t0, REG(B7_4), t0)
> add sr_t_in, B7_4 // instead of mov t1, sr_t; add t1, B7_4
> mov sr_t_out, 0
> adc sr_t_out, 0 // cout(B7_r + sr_t_in)
The registers are allocated from left to right, started by the inputs
first.
- cpu_sr_t is already in register or in memory and loaded to a register
- t0 is a constant, and the add2 op on x86_64 do not accept a constant
three so it is loaded to a register. However it is aliased to the
output and not dead as used again in the second add2 instruction. It
is therefore copied into another register.
- REG(B7_4) is already in register or in memory and loaded to a register
- t0 appears again and has been loaded to a register and therefore not
anymore a constant.
We therefore end up with (Intel notation)
xor %ebx, %ebx // this is t0
mov %r12d, %ebx // a copy of t0
add %r13d, %ebp // %r13d contains B7_4 and %ebp contains sr_t
adc %r12d, %ebx // %r12d is the new sr_t
> // add2(REG(B11_8), cpu_sr_t, t1, cpu_sr_t, REG(B11_8), t0)
> add B11_8, sr_t_in // B11_8 + B7_4 + sr_t_in
> adc sr_t_out, 0 // cout(B11_8 + B7_4 + sr_t_in)
add %ebp, %r13d // %ebp is now B11_8
adc %ebx, %r12d // %ebx is now cpu_sr_t
--
Aurelien Jarno GPG: 4096R/1DDD8C9B
aurelien@aurel32.net http://www.aurel32.net
^ permalink raw reply [flat|nested] 16+ messages in thread
end of thread, other threads:[~2015-06-04 16:09 UTC | newest]
Thread overview: 16+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2015-05-24 23:37 [Qemu-devel] [PATCH v3 0/8] target-sh4: optimizations and cleanups Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 1/8] target-sh4: use bit number for SR constants Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 2/8] target-sh4: Split out T from SR Aurelien Jarno
2015-06-04 5:01 ` Richard Henderson
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 3/8] target-sh4: optimize addc using add2 Aurelien Jarno
2015-06-04 5:03 ` Richard Henderson
2015-06-04 10:54 ` Paolo Bonzini
2015-06-04 16:08 ` Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 4/8] target-sh4: optimize subc using sub2 Aurelien Jarno
2015-06-04 5:05 ` Richard Henderson
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 5/8] target-sh4: optimize negc using add2 and sub2 Aurelien Jarno
2015-06-04 5:07 ` Richard Henderson
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 6/8] target-sh4: split out Q and M from of SR and optimize div1 Aurelien Jarno
2015-06-04 5:31 ` Richard Henderson
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 7/8] target-sh4: factorize fmov implementation Aurelien Jarno
2015-05-24 23:37 ` [Qemu-devel] [PATCH v3 8/8] target-sh4: remove dead code Aurelien Jarno
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).