* [PATCH v2 00/21] target/sparc: Cleanup condition codes etc
@ 2023-11-01 4:11 Richard Henderson
2023-11-01 4:11 ` [PATCH v2 01/21] target/sparc: Introduce cpu_put_psr_icc Richard Henderson
` (21 more replies)
0 siblings, 22 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
This was part of my guess for some of the performance problems.
I saw compute_all_sub quite high in the profile at some point, and I
believe that the test case has a partially rotated loop such that "cmp"
is in a delay slot, and so the gen_compare fast path for CC_OP_SUB is
not visible to the conditional branch that uses the output of the compare.
Which means that helper_compute_psr gets called more often that we'd like.
Since almost all Sparc instructions that set cc also have a version of
the instruction that does not set cc, we can trust that the compiler
has only used the cc-setting version when it is actually required.
Thus, unlike CISC processors, there is very little scope for optimization
of the flags -- we might as well compute them immediately.
Move away from CC_OP to explicit computation of conditions. This is
modeled on target/arm for the (mostly) separate representation of the bits.
We can pack icc.[NV] and xcc.[NV] into the same target_ulong, but Z and C
cannot share. (For "normal" setting of Z, we could share, but it is
possible to set xcc.Z and !icc.Z via explicit write to %ccr, and for
that we have to have two variables.)
After removing CC_OP, clean up the handling of conditions so that we can
minimize additional setcond required for env->cond.
Finally, inline some division, which can make use of the new out-of-line
exception path, which means we can expand UDIVX and SDIVX with very few
host insns. The 64/32 UDIV insn needs only a few more. Leave UDIVcc and
SDIV* out of line, as the overflow and saturation computation in these
cases is really too large to inline.
r~
Richard Henderson (21):
target/sparc: Introduce cpu_put_psr_icc
target/sparc: Split psr and xcc into components
target/sparc: Remove CC_OP_LOGIC
target/sparc: Remove CC_OP_DIV
target/sparc: Remove CC_OP_ADD, CC_OP_ADDX, CC_OP_TADD
target/sparc: Remove CC_OP_SUB, CC_OP_SUBX, CC_OP_TSUB
target/sparc: Remove CC_OP_TADDTV, CC_OP_TSUBTV
target/sparc: Remove CC_OP leftovers
target/sparc: Remove DisasCompare.is_bool
target/sparc: Change DisasCompare.c2 to int
target/sparc: Always copy conditions into a new temporary
target/sparc: Do flush_cond in advance_jump_cond
target/sparc: Merge gen_branch2 into advance_pc
target/sparc: Merge advance_jump_uncond_{never,always} into
advance_jump_cond
target/sparc: Pass displacement to advance_jump_cond
target/sparc: Merge gen_op_next_insn into only caller
target/sparc: Record entire jump condition in DisasContext
target/sparc: Discard cpu_cond at the end of each insn
target/sparc: Implement UDIVX and SDIVX inline
target/sparc: Implement UDIV inline
target/sparc: Check for invalid cond in gen_compare_reg
linux-user/sparc/target_cpu.h | 17 +-
target/sparc/cpu.h | 58 +-
target/sparc/helper.h | 12 +-
target/sparc/insns.decode | 7 +-
linux-user/sparc/cpu_loop.c | 11 +-
linux-user/sparc/signal.c | 2 +-
target/sparc/cc_helper.c | 471 ------------
target/sparc/cpu.c | 1 -
target/sparc/helper.c | 171 ++---
target/sparc/int32_helper.c | 5 -
target/sparc/int64_helper.c | 5 -
target/sparc/machine.c | 45 +-
target/sparc/translate.c | 1333 ++++++++++++++-------------------
target/sparc/win_helper.c | 56 +-
target/sparc/meson.build | 1 -
15 files changed, 789 insertions(+), 1406 deletions(-)
delete mode 100644 target/sparc/cc_helper.c
--
2.34.1
^ permalink raw reply [flat|nested] 23+ messages in thread
* [PATCH v2 01/21] target/sparc: Introduce cpu_put_psr_icc
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 02/21] target/sparc: Split psr and xcc into components Richard Henderson
` (20 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Isolate linux-user from changes to icc representation.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/cpu.h | 1 +
linux-user/sparc/signal.c | 2 +-
target/sparc/win_helper.c | 7 ++++++-
3 files changed, 8 insertions(+), 2 deletions(-)
diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 758a4e8aaa..955329f6c9 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -619,6 +619,7 @@ void sparc_restore_state_to_opc(CPUState *cs,
/* win_helper.c */
target_ulong cpu_get_psr(CPUSPARCState *env1);
void cpu_put_psr(CPUSPARCState *env1, target_ulong val);
+void cpu_put_psr_icc(CPUSPARCState *env1, target_ulong val);
void cpu_put_psr_raw(CPUSPARCState *env1, target_ulong val);
#ifdef TARGET_SPARC64
void cpu_change_pstate(CPUSPARCState *env1, uint32_t new_pstate);
diff --git a/linux-user/sparc/signal.c b/linux-user/sparc/signal.c
index 2be9000b9e..dfcae707e0 100644
--- a/linux-user/sparc/signal.c
+++ b/linux-user/sparc/signal.c
@@ -164,7 +164,7 @@ static void restore_pt_regs(struct target_pt_regs *regs, CPUSPARCState *env)
*/
uint32_t psr;
__get_user(psr, ®s->psr);
- env->psr = (psr & PSR_ICC) | (env->psr & ~PSR_ICC);
+ cpu_put_psr_icc(env, psr);
#endif
/* Note that pc and npc are handled in the caller. */
diff --git a/target/sparc/win_helper.c b/target/sparc/win_helper.c
index 3a7c0ff943..bf2c90c780 100644
--- a/target/sparc/win_helper.c
+++ b/target/sparc/win_helper.c
@@ -67,9 +67,14 @@ target_ulong cpu_get_psr(CPUSPARCState *env)
#endif
}
-void cpu_put_psr_raw(CPUSPARCState *env, target_ulong val)
+void cpu_put_psr_icc(CPUSPARCState *env, target_ulong val)
{
env->psr = val & PSR_ICC;
+}
+
+void cpu_put_psr_raw(CPUSPARCState *env, target_ulong val)
+{
+ cpu_put_psr_icc(env, val);
#if !defined(TARGET_SPARC64)
env->psref = (val & PSR_EF) ? 1 : 0;
env->psrpil = (val & PSR_PIL) >> 8;
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 02/21] target/sparc: Split psr and xcc into components
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
2023-11-01 4:11 ` [PATCH v2 01/21] target/sparc: Introduce cpu_put_psr_icc Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 03/21] target/sparc: Remove CC_OP_LOGIC Richard Henderson
` (19 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Step in removing CC_OP: change the representation of CC_OP_FLAGS.
The 8 bits are distributed between 6 variables, which should make
it easy to keep up to date.
The code within cc_helper.c is quite ugly but is only temporary.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
linux-user/sparc/target_cpu.h | 17 +-
target/sparc/cpu.h | 30 ++-
linux-user/sparc/cpu_loop.c | 6 +-
target/sparc/cc_helper.c | 51 +++--
target/sparc/machine.c | 45 ++++-
target/sparc/translate.c | 362 +++++++++++++---------------------
target/sparc/win_helper.c | 52 ++++-
7 files changed, 291 insertions(+), 272 deletions(-)
diff --git a/linux-user/sparc/target_cpu.h b/linux-user/sparc/target_cpu.h
index 1f4bed50f4..5f62c5eb75 100644
--- a/linux-user/sparc/target_cpu.h
+++ b/linux-user/sparc/target_cpu.h
@@ -26,6 +26,17 @@
# define TARGET_STACK_BIAS 0
#endif
+static void set_syscall_C(CPUSPARCState *env, bool val)
+{
+#ifndef TARGET_SPARC64
+ env->icc_C = val;
+#elif defined(TARGET_ABI32)
+ env->icc_C = (uint64_t)val << 32;
+#else
+ env->xcc_C = val;
+#endif
+}
+
static inline void cpu_clone_regs_child(CPUSPARCState *env, target_ulong newsp,
unsigned flags)
{
@@ -58,11 +69,7 @@ static inline void cpu_clone_regs_child(CPUSPARCState *env, target_ulong newsp,
* do the pc advance twice.
*/
env->regwptr[WREG_O0] = 0;
-#if defined(TARGET_SPARC64) && !defined(TARGET_ABI32)
- env->xcc &= ~PSR_CARRY;
-#else
- env->psr &= ~PSR_CARRY;
-#endif
+ set_syscall_C(env, 0);
env->pc = env->npc;
env->npc = env->npc + 4;
}
diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 955329f6c9..ea8a04c6e3 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -149,7 +149,7 @@ enum {
*/
enum {
CC_OP_DYNAMIC, /* must use dynamic code to get cc_op */
- CC_OP_FLAGS, /* all cc are back in status register */
+ CC_OP_FLAGS, /* all cc are back in cc_*_[NZCV] registers */
CC_OP_DIV, /* modify N, Z and V, C = 0*/
CC_OP_ADD, /* modify all flags, CC_DST = res, CC_SRC = src1 */
CC_OP_ADDX, /* modify all flags, CC_DST = res, CC_SRC = src1 */
@@ -458,6 +458,32 @@ struct CPUArchState {
target_ulong npc; /* next program counter */
target_ulong y; /* multiply/divide register */
+ /*
+ * Bit 31 is for icc, bit 63 for xcc.
+ * Other bits are garbage.
+ */
+ target_long cc_N;
+ target_long cc_V;
+
+ /*
+ * Z is represented as == 0; any non-zero value is !Z.
+ * For sparc64, the high 32-bits of icc.Z are garbage.
+ */
+ target_ulong icc_Z;
+#ifdef TARGET_SPARC64
+ target_ulong xcc_Z;
+#endif
+
+ /*
+ * For sparc32, icc.C is boolean.
+ * For sparc64, xcc.C is boolean;
+ * icc.C is bit 32 with other bits garbage.
+ */
+ target_ulong icc_C;
+#ifdef TARGET_SPARC64
+ target_ulong xcc_C;
+#endif
+
/* emulator internal flags handling */
target_ulong cc_src, cc_src2;
target_ulong cc_dst;
@@ -466,7 +492,6 @@ struct CPUArchState {
target_ulong cond; /* conditional branch result (XXX: save it in a
temporary register when possible) */
- uint32_t psr; /* processor state register */
target_ulong fsr; /* FPU state register */
CPU_DoubleU fpr[TARGET_DPREGS]; /* floating point registers */
uint32_t cwp; /* index of current register window (extracted
@@ -522,7 +547,6 @@ struct CPUArchState {
#define MAXTL_MAX 8
#define MAXTL_MASK (MAXTL_MAX - 1)
trap_state ts[MAXTL_MAX];
- uint32_t xcc; /* Extended integer condition codes */
uint32_t asi;
uint32_t pstate;
uint32_t tl;
diff --git a/linux-user/sparc/cpu_loop.c b/linux-user/sparc/cpu_loop.c
index b36bb2574b..c1a2362041 100644
--- a/linux-user/sparc/cpu_loop.c
+++ b/linux-user/sparc/cpu_loop.c
@@ -197,10 +197,8 @@ static uint32_t do_getpsr(CPUSPARCState *env)
/* Avoid ifdefs below for the abi32 and abi64 paths. */
#ifdef TARGET_ABI32
#define TARGET_TT_SYSCALL (TT_TRAP + 0x10) /* t_linux */
-#define syscall_cc psr
#else
#define TARGET_TT_SYSCALL (TT_TRAP + 0x6d) /* tl0_linux64 */
-#define syscall_cc xcc
#endif
/* Avoid ifdefs below for the v9 and pre-v9 hw traps. */
@@ -240,10 +238,10 @@ void cpu_loop (CPUSPARCState *env)
break;
}
if ((abi_ulong)ret >= (abi_ulong)(-515)) {
- env->syscall_cc |= PSR_CARRY;
+ set_syscall_C(env, 1);
ret = -ret;
} else {
- env->syscall_cc &= ~PSR_CARRY;
+ set_syscall_C(env, 0);
}
env->regwptr[0] = ret;
/* next instruction */
diff --git a/target/sparc/cc_helper.c b/target/sparc/cc_helper.c
index 7ad5b9b29e..46bec69d96 100644
--- a/target/sparc/cc_helper.c
+++ b/target/sparc/cc_helper.c
@@ -21,16 +21,6 @@
#include "cpu.h"
#include "exec/helper-proto.h"
-static uint32_t compute_all_flags(CPUSPARCState *env)
-{
- return env->psr & PSR_ICC;
-}
-
-static uint32_t compute_C_flags(CPUSPARCState *env)
-{
- return env->psr & PSR_CARRY;
-}
-
static inline uint32_t get_NZ_icc(int32_t dst)
{
uint32_t ret = 0;
@@ -44,16 +34,6 @@ static inline uint32_t get_NZ_icc(int32_t dst)
}
#ifdef TARGET_SPARC64
-static uint32_t compute_all_flags_xcc(CPUSPARCState *env)
-{
- return env->xcc & PSR_ICC;
-}
-
-static uint32_t compute_C_flags_xcc(CPUSPARCState *env)
-{
- return env->xcc & PSR_CARRY;
-}
-
static inline uint32_t get_NZ_xcc(target_long dst)
{
uint32_t ret = 0;
@@ -422,7 +402,6 @@ typedef struct CCTable {
static const CCTable icc_table[CC_OP_NB] = {
/* CC_OP_DYNAMIC should never happen */
- [CC_OP_FLAGS] = { compute_all_flags, compute_C_flags },
[CC_OP_DIV] = { compute_all_div, compute_C_div },
[CC_OP_ADD] = { compute_all_add, compute_C_add },
[CC_OP_ADDX] = { compute_all_addx, compute_C_addx },
@@ -438,7 +417,6 @@ static const CCTable icc_table[CC_OP_NB] = {
#ifdef TARGET_SPARC64
static const CCTable xcc_table[CC_OP_NB] = {
/* CC_OP_DYNAMIC should never happen */
- [CC_OP_FLAGS] = { compute_all_flags_xcc, compute_C_flags_xcc },
[CC_OP_DIV] = { compute_all_logic_xcc, compute_C_logic },
[CC_OP_ADD] = { compute_all_add_xcc, compute_C_add_xcc },
[CC_OP_ADDX] = { compute_all_addx_xcc, compute_C_addx_xcc },
@@ -454,18 +432,37 @@ static const CCTable xcc_table[CC_OP_NB] = {
void helper_compute_psr(CPUSPARCState *env)
{
- uint32_t new_psr;
+ if (CC_OP == CC_OP_FLAGS) {
+ return;
+ }
- new_psr = icc_table[CC_OP].compute_all(env);
- env->psr = new_psr;
+ uint32_t icc = icc_table[CC_OP].compute_all(env);
#ifdef TARGET_SPARC64
- new_psr = xcc_table[CC_OP].compute_all(env);
- env->xcc = new_psr;
+ uint32_t xcc = xcc_table[CC_OP].compute_all(env);
+
+ env->cc_N = deposit64(-(icc & PSR_NEG), 32, 32, -(xcc & PSR_NEG));
+ env->cc_V = deposit64(-(icc & PSR_OVF), 32, 32, -(xcc & PSR_OVF));
+ env->icc_C = (uint64_t)icc << (32 - PSR_CARRY_SHIFT);
+ env->xcc_C = (xcc >> PSR_CARRY_SHIFT) & 1;
+ env->xcc_Z = ~xcc & PSR_ZERO;
+#else
+ env->cc_N = -(icc & PSR_NEG);
+ env->cc_V = -(icc & PSR_OVF);
+ env->icc_C = (icc >> PSR_CARRY_SHIFT) & 1;
#endif
+ env->icc_Z = ~icc & PSR_ZERO;
+
CC_OP = CC_OP_FLAGS;
}
uint32_t helper_compute_C_icc(CPUSPARCState *env)
{
+ if (CC_OP == CC_OP_FLAGS) {
+#ifdef TARGET_SPARC64
+ return extract64(env->icc_C, 32, 1);
+#else
+ return env->icc_C;
+#endif
+ }
return icc_table[CC_OP].compute_c(env) >> PSR_CARRY_SHIFT;
}
diff --git a/target/sparc/machine.c b/target/sparc/machine.c
index 274e1217df..44dfc07014 100644
--- a/target/sparc/machine.c
+++ b/target/sparc/machine.c
@@ -83,6 +83,42 @@ static const VMStateInfo vmstate_psr = {
.put = put_psr,
};
+#ifdef TARGET_SPARC64
+static int get_xcc(QEMUFile *f, void *opaque, size_t size,
+ const VMStateField *field)
+{
+ SPARCCPU *cpu = opaque;
+ CPUSPARCState *env = &cpu->env;
+ uint32_t val = qemu_get_be32(f);
+
+ /* Do not clobber icc.[NV] */
+ env->cc_N = deposit64(env->cc_N, 32, 32, -(val & PSR_NEG));
+ env->cc_V = deposit64(env->cc_V, 32, 32, -(val & PSR_OVF));
+ env->xcc_Z = ~val & PSR_ZERO;
+ env->xcc_C = (val >> PSR_CARRY_SHIFT) & 1;
+
+ return 0;
+}
+
+static int put_xcc(QEMUFile *f, void *opaque, size_t size,
+ const VMStateField *field, JSONWriter *vmdesc)
+{
+ SPARCCPU *cpu = opaque;
+ CPUSPARCState *env = &cpu->env;
+ uint32_t val = cpu_get_ccr(env);
+
+ /* Extract just xcc out of ccr and shift into legacy position. */
+ qemu_put_be32(f, (val & 0xf0) << (20 - 4));
+ return 0;
+}
+
+static const VMStateInfo vmstate_xcc = {
+ .name = "xcc",
+ .get = get_xcc,
+ .put = put_xcc,
+};
+#endif
+
static int cpu_pre_save(void *opaque)
{
SPARCCPU *cpu = opaque;
@@ -155,7 +191,14 @@ const VMStateDescription vmstate_sparc_cpu = {
VMSTATE_UINT32(env.mmu_version, SPARCCPU),
VMSTATE_STRUCT_ARRAY(env.ts, SPARCCPU, MAXTL_MAX, 0,
vmstate_trap_state, trap_state),
- VMSTATE_UINT32(env.xcc, SPARCCPU),
+ {
+ .name = "xcc",
+ .version_id = 0,
+ .size = sizeof(uint32_t),
+ .info = &vmstate_xcc,
+ .flags = VMS_SINGLE,
+ .offset = 0,
+ },
VMSTATE_UINT32(env.asi, SPARCCPU),
VMSTATE_UINT32(env.pstate, SPARCCPU),
VMSTATE_UINT32(env.tl, SPARCCPU),
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 986a88c4e1..261f142636 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -107,19 +107,35 @@
static TCGv_ptr cpu_regwptr;
static TCGv cpu_cc_src, cpu_cc_src2, cpu_cc_dst;
static TCGv_i32 cpu_cc_op;
-static TCGv_i32 cpu_psr;
static TCGv cpu_fsr, cpu_pc, cpu_npc;
static TCGv cpu_regs[32];
static TCGv cpu_y;
static TCGv cpu_tbr;
static TCGv cpu_cond;
+static TCGv cpu_cc_N;
+static TCGv cpu_cc_V;
+static TCGv cpu_icc_Z;
+static TCGv cpu_icc_C;
#ifdef TARGET_SPARC64
-static TCGv_i32 cpu_xcc, cpu_fprs;
+static TCGv cpu_xcc_Z;
+static TCGv cpu_xcc_C;
+static TCGv_i32 cpu_fprs;
static TCGv cpu_gsr;
#else
# define cpu_fprs ({ qemu_build_not_reached(); (TCGv)NULL; })
# define cpu_gsr ({ qemu_build_not_reached(); (TCGv)NULL; })
#endif
+
+#ifdef TARGET_SPARC64
+#define cpu_cc_Z cpu_xcc_Z
+#define cpu_cc_C cpu_xcc_C
+#else
+#define cpu_cc_Z cpu_icc_Z
+#define cpu_cc_C cpu_icc_C
+#define cpu_xcc_Z ({ qemu_build_not_reached(); NULL; })
+#define cpu_xcc_C ({ qemu_build_not_reached(); NULL; })
+#endif
+
/* Floating point registers */
static TCGv_i64 cpu_fpr[TARGET_DPREGS];
@@ -366,31 +382,6 @@ static void gen_goto_tb(DisasContext *s, int tb_num,
}
}
-// XXX suboptimal
-static void gen_mov_reg_N(TCGv reg, TCGv_i32 src)
-{
- tcg_gen_extu_i32_tl(reg, src);
- tcg_gen_extract_tl(reg, reg, PSR_NEG_SHIFT, 1);
-}
-
-static void gen_mov_reg_Z(TCGv reg, TCGv_i32 src)
-{
- tcg_gen_extu_i32_tl(reg, src);
- tcg_gen_extract_tl(reg, reg, PSR_ZERO_SHIFT, 1);
-}
-
-static void gen_mov_reg_V(TCGv reg, TCGv_i32 src)
-{
- tcg_gen_extu_i32_tl(reg, src);
- tcg_gen_extract_tl(reg, reg, PSR_OVF_SHIFT, 1);
-}
-
-static void gen_mov_reg_C(TCGv reg, TCGv_i32 src)
-{
- tcg_gen_extu_i32_tl(reg, src);
- tcg_gen_extract_tl(reg, reg, PSR_CARRY_SHIFT, 1);
-}
-
static void gen_op_add_cc(TCGv dst, TCGv src1, TCGv src2)
{
tcg_gen_mov_tl(cpu_cc_src, src1);
@@ -640,13 +631,11 @@ static void gen_op_mulscc(TCGv dst, TCGv src1, TCGv src2)
tcg_gen_deposit_tl(cpu_y, t0, cpu_cc_src, 31, 1);
// b1 = N ^ V;
- gen_mov_reg_N(t0, cpu_psr);
- gen_mov_reg_V(r_temp, cpu_psr);
- tcg_gen_xor_tl(t0, t0, r_temp);
+ tcg_gen_xor_tl(t0, cpu_cc_N, cpu_cc_V);
// T0 = (b1 << 31) | (T0 >> 1);
// src1 = T0;
- tcg_gen_shli_tl(t0, t0, 31);
+ tcg_gen_andi_tl(t0, t0, 1u << 31);
tcg_gen_shri_tl(cpu_cc_src, cpu_cc_src, 1);
tcg_gen_or_tl(cpu_cc_src, cpu_cc_src, t0);
@@ -825,114 +814,12 @@ static void gen_op_eval_ba(TCGv dst)
tcg_gen_movi_tl(dst, 1);
}
-// Z
-static void gen_op_eval_be(TCGv dst, TCGv_i32 src)
-{
- gen_mov_reg_Z(dst, src);
-}
-
-// Z | (N ^ V)
-static void gen_op_eval_ble(TCGv dst, TCGv_i32 src)
-{
- TCGv t0 = tcg_temp_new();
- gen_mov_reg_N(t0, src);
- gen_mov_reg_V(dst, src);
- tcg_gen_xor_tl(dst, dst, t0);
- gen_mov_reg_Z(t0, src);
- tcg_gen_or_tl(dst, dst, t0);
-}
-
-// N ^ V
-static void gen_op_eval_bl(TCGv dst, TCGv_i32 src)
-{
- TCGv t0 = tcg_temp_new();
- gen_mov_reg_V(t0, src);
- gen_mov_reg_N(dst, src);
- tcg_gen_xor_tl(dst, dst, t0);
-}
-
-// C | Z
-static void gen_op_eval_bleu(TCGv dst, TCGv_i32 src)
-{
- TCGv t0 = tcg_temp_new();
- gen_mov_reg_Z(t0, src);
- gen_mov_reg_C(dst, src);
- tcg_gen_or_tl(dst, dst, t0);
-}
-
-// C
-static void gen_op_eval_bcs(TCGv dst, TCGv_i32 src)
-{
- gen_mov_reg_C(dst, src);
-}
-
-// V
-static void gen_op_eval_bvs(TCGv dst, TCGv_i32 src)
-{
- gen_mov_reg_V(dst, src);
-}
-
// 0
static void gen_op_eval_bn(TCGv dst)
{
tcg_gen_movi_tl(dst, 0);
}
-// N
-static void gen_op_eval_bneg(TCGv dst, TCGv_i32 src)
-{
- gen_mov_reg_N(dst, src);
-}
-
-// !Z
-static void gen_op_eval_bne(TCGv dst, TCGv_i32 src)
-{
- gen_mov_reg_Z(dst, src);
- tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// !(Z | (N ^ V))
-static void gen_op_eval_bg(TCGv dst, TCGv_i32 src)
-{
- gen_op_eval_ble(dst, src);
- tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// !(N ^ V)
-static void gen_op_eval_bge(TCGv dst, TCGv_i32 src)
-{
- gen_op_eval_bl(dst, src);
- tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// !(C | Z)
-static void gen_op_eval_bgu(TCGv dst, TCGv_i32 src)
-{
- gen_op_eval_bleu(dst, src);
- tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// !C
-static void gen_op_eval_bcc(TCGv dst, TCGv_i32 src)
-{
- gen_mov_reg_C(dst, src);
- tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// !N
-static void gen_op_eval_bpos(TCGv dst, TCGv_i32 src)
-{
- gen_mov_reg_N(dst, src);
- tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// !V
-static void gen_op_eval_bvc(TCGv dst, TCGv_i32 src)
-{
- gen_mov_reg_V(dst, src);
- tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
/*
FPSR bit field FCC1 | FCC0:
0 =
@@ -1249,34 +1136,22 @@ static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
TCG_COND_ALWAYS, /* vc: !V -> 1 */
};
- TCGv_i32 r_src;
- TCGv r_dst;
+ TCGv t1, t2;
-#ifdef TARGET_SPARC64
- if (xcc) {
- r_src = cpu_xcc;
- } else {
- r_src = cpu_psr;
- }
-#else
- r_src = cpu_psr;
-#endif
+ cmp->is_bool = false;
switch (dc->cc_op) {
case CC_OP_LOGIC:
cmp->cond = logic_cond[cond];
do_compare_dst_0:
- cmp->is_bool = false;
cmp->c2 = tcg_constant_tl(0);
-#ifdef TARGET_SPARC64
- if (!xcc) {
- cmp->c1 = tcg_temp_new();
- tcg_gen_ext32s_tl(cmp->c1, cpu_cc_dst);
- break;
+ if (TARGET_LONG_BITS == 32 || xcc) {
+ cmp->c1 = cpu_cc_dst;
+ } else {
+ cmp->c1 = t1 = tcg_temp_new();
+ tcg_gen_ext32s_tl(t1, cpu_cc_dst);
}
-#endif
- cmp->c1 = cpu_cc_dst;
- break;
+ return;
case CC_OP_SUB:
switch (cond) {
@@ -1287,92 +1162,127 @@ static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
case 7: /* overflow */
case 15: /* !overflow */
- goto do_dynamic;
+ break;
default:
cmp->cond = subcc_cond[cond];
- cmp->is_bool = false;
-#ifdef TARGET_SPARC64
- if (!xcc) {
+ if (TARGET_LONG_BITS == 32 || xcc) {
+ cmp->c1 = cpu_cc_src;
+ cmp->c2 = cpu_cc_src2;
+ } else {
/* Note that sign-extension works for unsigned compares as
long as both operands are sign-extended. */
- cmp->c1 = tcg_temp_new();
- cmp->c2 = tcg_temp_new();
- tcg_gen_ext32s_tl(cmp->c1, cpu_cc_src);
- tcg_gen_ext32s_tl(cmp->c2, cpu_cc_src2);
- break;
+ cmp->c1 = t1 = tcg_temp_new();
+ tcg_gen_ext32s_tl(t1, cpu_cc_src);
+ cmp->c2 = t2 = tcg_temp_new();
+ tcg_gen_ext32s_tl(t2, cpu_cc_src2);
}
-#endif
- cmp->c1 = cpu_cc_src;
- cmp->c2 = cpu_cc_src2;
- break;
+ return;
}
break;
default:
- do_dynamic:
gen_helper_compute_psr(tcg_env);
dc->cc_op = CC_OP_FLAGS;
- /* FALLTHRU */
+ break;
case CC_OP_FLAGS:
- /* We're going to generate a boolean result. */
- cmp->cond = TCG_COND_NE;
- cmp->is_bool = true;
- cmp->c1 = r_dst = tcg_temp_new();
- cmp->c2 = tcg_constant_tl(0);
+ break;
+ }
- switch (cond) {
- case 0x0:
- gen_op_eval_bn(r_dst);
- break;
- case 0x1:
- gen_op_eval_be(r_dst, r_src);
- break;
- case 0x2:
- gen_op_eval_ble(r_dst, r_src);
- break;
- case 0x3:
- gen_op_eval_bl(r_dst, r_src);
- break;
- case 0x4:
- gen_op_eval_bleu(r_dst, r_src);
- break;
- case 0x5:
- gen_op_eval_bcs(r_dst, r_src);
- break;
- case 0x6:
- gen_op_eval_bneg(r_dst, r_src);
- break;
- case 0x7:
- gen_op_eval_bvs(r_dst, r_src);
- break;
- case 0x8:
- gen_op_eval_ba(r_dst);
- break;
- case 0x9:
- gen_op_eval_bne(r_dst, r_src);
- break;
- case 0xa:
- gen_op_eval_bg(r_dst, r_src);
- break;
- case 0xb:
- gen_op_eval_bge(r_dst, r_src);
- break;
- case 0xc:
- gen_op_eval_bgu(r_dst, r_src);
- break;
- case 0xd:
- gen_op_eval_bcc(r_dst, r_src);
- break;
- case 0xe:
- gen_op_eval_bpos(r_dst, r_src);
- break;
- case 0xf:
- gen_op_eval_bvc(r_dst, r_src);
- break;
+ cmp->c1 = t1 = tcg_temp_new();
+ cmp->c2 = tcg_constant_tl(0);
+
+ switch (cond & 7) {
+ case 0x0: /* never */
+ cmp->cond = TCG_COND_NEVER;
+ cmp->c1 = cmp->c2;
+ break;
+
+ case 0x1: /* eq: Z */
+ cmp->cond = TCG_COND_EQ;
+ if (TARGET_LONG_BITS == 32 || xcc) {
+ tcg_gen_mov_tl(t1, cpu_cc_Z);
+ } else {
+ tcg_gen_ext32u_tl(t1, cpu_icc_Z);
}
break;
+
+ case 0x2: /* le: Z | (N ^ V) */
+ /*
+ * Simplify:
+ * cc_Z || (N ^ V) < 0 NE
+ * cc_Z && !((N ^ V) < 0) EQ
+ * cc_Z & ~((N ^ V) >> TLB) EQ
+ */
+ cmp->cond = TCG_COND_EQ;
+ tcg_gen_xor_tl(t1, cpu_cc_N, cpu_cc_V);
+ tcg_gen_sextract_tl(t1, t1, xcc ? 63 : 31, 1);
+ tcg_gen_andc_tl(t1, xcc ? cpu_cc_Z : cpu_icc_Z, t1);
+ if (TARGET_LONG_BITS == 64 && !xcc) {
+ tcg_gen_ext32u_tl(t1, t1);
+ }
+ break;
+
+ case 0x3: /* lt: N ^ V */
+ cmp->cond = TCG_COND_LT;
+ tcg_gen_xor_tl(t1, cpu_cc_N, cpu_cc_V);
+ if (TARGET_LONG_BITS == 64 && !xcc) {
+ tcg_gen_ext32s_tl(t1, t1);
+ }
+ break;
+
+ case 0x4: /* leu: Z | C */
+ /*
+ * Simplify:
+ * cc_Z == 0 || cc_C != 0 NE
+ * cc_Z != 0 && cc_C == 0 EQ
+ * cc_Z & (cc_C ? 0 : -1) EQ
+ * cc_Z & (cc_C - 1) EQ
+ */
+ cmp->cond = TCG_COND_EQ;
+ if (TARGET_LONG_BITS == 32 || xcc) {
+ tcg_gen_subi_tl(t1, cpu_cc_C, 1);
+ tcg_gen_and_tl(t1, t1, cpu_cc_Z);
+ } else {
+ tcg_gen_extract_tl(t1, cpu_icc_C, 32, 1);
+ tcg_gen_subi_tl(t1, t1, 1);
+ tcg_gen_and_tl(t1, t1, cpu_icc_Z);
+ tcg_gen_ext32u_tl(t1, t1);
+ }
+ break;
+
+ case 0x5: /* ltu: C */
+ cmp->cond = TCG_COND_NE;
+ cmp->is_bool = true;
+ if (TARGET_LONG_BITS == 32 || xcc) {
+ tcg_gen_mov_tl(t1, cpu_cc_C);
+ } else {
+ tcg_gen_extract_tl(t1, cpu_icc_C, 32, 1);
+ }
+ break;
+
+ case 0x6: /* neg: N */
+ cmp->cond = TCG_COND_LT;
+ if (TARGET_LONG_BITS == 32 || xcc) {
+ tcg_gen_mov_tl(t1, cpu_cc_N);
+ } else {
+ tcg_gen_ext32s_tl(t1, cpu_cc_N);
+ }
+ break;
+
+ case 0x7: /* vs: V */
+ cmp->cond = TCG_COND_LT;
+ if (TARGET_LONG_BITS == 32 || xcc) {
+ tcg_gen_mov_tl(t1, cpu_cc_V);
+ } else {
+ tcg_gen_ext32s_tl(t1, cpu_cc_V);
+ }
+ break;
+ }
+ if (cond & 8) {
+ cmp->cond = tcg_invert_cond(cmp->cond);
+ cmp->is_bool = false;
}
}
@@ -5513,17 +5423,21 @@ void sparc_tcg_init(void)
static const struct { TCGv_i32 *ptr; int off; const char *name; } r32[] = {
#ifdef TARGET_SPARC64
- { &cpu_xcc, offsetof(CPUSPARCState, xcc), "xcc" },
{ &cpu_fprs, offsetof(CPUSPARCState, fprs), "fprs" },
#endif
{ &cpu_cc_op, offsetof(CPUSPARCState, cc_op), "cc_op" },
- { &cpu_psr, offsetof(CPUSPARCState, psr), "psr" },
};
static const struct { TCGv *ptr; int off; const char *name; } rtl[] = {
#ifdef TARGET_SPARC64
{ &cpu_gsr, offsetof(CPUSPARCState, gsr), "gsr" },
+ { &cpu_xcc_Z, offsetof(CPUSPARCState, xcc_Z), "xcc_Z" },
+ { &cpu_xcc_C, offsetof(CPUSPARCState, xcc_C), "xcc_C" },
#endif
+ { &cpu_cc_N, offsetof(CPUSPARCState, cc_N), "cc_N" },
+ { &cpu_cc_V, offsetof(CPUSPARCState, cc_V), "cc_V" },
+ { &cpu_icc_Z, offsetof(CPUSPARCState, icc_Z), "icc_Z" },
+ { &cpu_icc_C, offsetof(CPUSPARCState, icc_C), "icc_C" },
{ &cpu_cond, offsetof(CPUSPARCState, cond), "cond" },
{ &cpu_cc_src, offsetof(CPUSPARCState, cc_src), "cc_src" },
{ &cpu_cc_src2, offsetof(CPUSPARCState, cc_src2), "cc_src2" },
diff --git a/target/sparc/win_helper.c b/target/sparc/win_helper.c
index bf2c90c780..f0ff6bf5db 100644
--- a/target/sparc/win_helper.c
+++ b/target/sparc/win_helper.c
@@ -53,23 +53,44 @@ void cpu_set_cwp(CPUSPARCState *env, int new_cwp)
target_ulong cpu_get_psr(CPUSPARCState *env)
{
+ target_ulong icc = 0;
+
helper_compute_psr(env);
+ icc |= ((int32_t)env->cc_N < 0) << PSR_NEG_SHIFT;
+ icc |= ((int32_t)env->cc_V < 0) << PSR_OVF_SHIFT;
+ icc |= ((int32_t)env->icc_Z == 0) << PSR_ZERO_SHIFT;
+ if (TARGET_LONG_BITS == 64) {
+ icc |= extract64(env->icc_C, 32, 1) << PSR_CARRY_SHIFT;
+ } else {
+ icc |= env->icc_C << PSR_CARRY_SHIFT;
+ }
+
#if !defined(TARGET_SPARC64)
- return env->version | (env->psr & PSR_ICC) |
+ return env->version | icc |
(env->psref ? PSR_EF : 0) |
(env->psrpil << 8) |
(env->psrs ? PSR_S : 0) |
(env->psrps ? PSR_PS : 0) |
(env->psret ? PSR_ET : 0) | env->cwp;
#else
- return env->psr & PSR_ICC;
+ return icc;
#endif
}
void cpu_put_psr_icc(CPUSPARCState *env, target_ulong val)
{
- env->psr = val & PSR_ICC;
+ if (TARGET_LONG_BITS == 64) {
+ /* Do not clobber xcc.[NV] */
+ env->cc_N = deposit64(env->cc_N, 0, 32, -(val & PSR_NEG));
+ env->cc_V = deposit64(env->cc_V, 0, 32, -(val & PSR_OVF));
+ env->icc_C = -(val & PSR_CARRY);
+ } else {
+ env->cc_N = -(val & PSR_NEG);
+ env->cc_V = -(val & PSR_OVF);
+ env->icc_C = (val >> PSR_CARRY_SHIFT) & 1;
+ }
+ env->icc_Z = ~val & PSR_ZERO;
}
void cpu_put_psr_raw(CPUSPARCState *env, target_ulong val)
@@ -249,17 +270,32 @@ void helper_restored(CPUSPARCState *env)
target_ulong cpu_get_ccr(CPUSPARCState *env)
{
- target_ulong psr;
+ target_ulong ccr = 0;
- psr = cpu_get_psr(env);
+ helper_compute_psr(env);
- return ((env->xcc >> 20) << 4) | ((psr & PSR_ICC) >> 20);
+ ccr |= (env->icc_C >> 32) & 1;
+ ccr |= ((int32_t)env->cc_V < 0) << 1;
+ ccr |= ((int32_t)env->icc_Z == 0) << 2;
+ ccr |= ((int32_t)env->cc_N < 0) << 3;
+
+ ccr |= env->xcc_C << 4;
+ ccr |= (env->cc_V < 0) << 5;
+ ccr |= (env->xcc_Z == 0) << 6;
+ ccr |= (env->cc_N < 0) << 7;
+
+ return ccr;
}
void cpu_put_ccr(CPUSPARCState *env, target_ulong val)
{
- env->xcc = (val >> 4) << 20;
- env->psr = (val & 0xf) << 20;
+ env->cc_N = deposit64(-(val & 0x08), 32, 32, -(val & 0x80));
+ env->cc_V = deposit64(-(val & 0x02), 32, 32, -(val & 0x20));
+ env->icc_C = (uint64_t)val << 32;
+ env->xcc_C = (val >> 4) & 1;
+ env->icc_Z = ~val & 0x04;
+ env->xcc_Z = ~val & 0x40;
+
CC_OP = CC_OP_FLAGS;
}
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 03/21] target/sparc: Remove CC_OP_LOGIC
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
2023-11-01 4:11 ` [PATCH v2 01/21] target/sparc: Introduce cpu_put_psr_icc Richard Henderson
2023-11-01 4:11 ` [PATCH v2 02/21] target/sparc: Split psr and xcc into components Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 04/21] target/sparc: Remove CC_OP_DIV Richard Henderson
` (18 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/cpu.h | 1 -
target/sparc/cc_helper.c | 14 +--------
target/sparc/translate.c | 66 ++++++++++++++++------------------------
3 files changed, 28 insertions(+), 53 deletions(-)
diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index ea8a04c6e3..202c34f7ca 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -159,7 +159,6 @@ enum {
CC_OP_SUBX, /* modify all flags, CC_DST = res, CC_SRC = src1 */
CC_OP_TSUB, /* modify all flags, CC_DST = res, CC_SRC = src1 */
CC_OP_TSUBTV, /* modify all flags except V, CC_DST = res, CC_SRC = src1 */
- CC_OP_LOGIC, /* modify N and Z, C = V = 0, CC_DST = res */
CC_OP_NB,
};
diff --git a/target/sparc/cc_helper.c b/target/sparc/cc_helper.c
index 46bec69d96..1622300a14 100644
--- a/target/sparc/cc_helper.c
+++ b/target/sparc/cc_helper.c
@@ -378,16 +378,6 @@ static uint32_t compute_all_tsubtv(CPUSPARCState *env)
return ret;
}
-static uint32_t compute_all_logic(CPUSPARCState *env)
-{
- return get_NZ_icc(CC_DST);
-}
-
-static uint32_t compute_C_logic(CPUSPARCState *env)
-{
- return 0;
-}
-
#ifdef TARGET_SPARC64
static uint32_t compute_all_logic_xcc(CPUSPARCState *env)
{
@@ -411,13 +401,12 @@ static const CCTable icc_table[CC_OP_NB] = {
[CC_OP_SUBX] = { compute_all_subx, compute_C_subx },
[CC_OP_TSUB] = { compute_all_tsub, compute_C_sub },
[CC_OP_TSUBTV] = { compute_all_tsubtv, compute_C_sub },
- [CC_OP_LOGIC] = { compute_all_logic, compute_C_logic },
};
#ifdef TARGET_SPARC64
static const CCTable xcc_table[CC_OP_NB] = {
/* CC_OP_DYNAMIC should never happen */
- [CC_OP_DIV] = { compute_all_logic_xcc, compute_C_logic },
+ [CC_OP_DIV] = { compute_all_logic_xcc, compute_C_div },
[CC_OP_ADD] = { compute_all_add_xcc, compute_C_add_xcc },
[CC_OP_ADDX] = { compute_all_addx_xcc, compute_C_addx_xcc },
[CC_OP_TADD] = { compute_all_add_xcc, compute_C_add_xcc },
@@ -426,7 +415,6 @@ static const CCTable xcc_table[CC_OP_NB] = {
[CC_OP_SUBX] = { compute_all_subx_xcc, compute_C_subx_xcc },
[CC_OP_TSUB] = { compute_all_sub_xcc, compute_C_sub_xcc },
[CC_OP_TSUBTV] = { compute_all_sub_xcc, compute_C_sub_xcc },
- [CC_OP_LOGIC] = { compute_all_logic_xcc, compute_C_logic },
};
#endif
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 261f142636..b11d89343b 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -1117,48 +1117,24 @@ static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
-1, /* no overflow */
};
- static int logic_cond[16] = {
- TCG_COND_NEVER,
- TCG_COND_EQ, /* eq: Z */
- TCG_COND_LE, /* le: Z | (N ^ V) -> Z | N */
- TCG_COND_LT, /* lt: N ^ V -> N */
- TCG_COND_EQ, /* leu: C | Z -> Z */
- TCG_COND_NEVER, /* ltu: C -> 0 */
- TCG_COND_LT, /* neg: N */
- TCG_COND_NEVER, /* vs: V -> 0 */
- TCG_COND_ALWAYS,
- TCG_COND_NE, /* ne: !Z */
- TCG_COND_GT, /* gt: !(Z | (N ^ V)) -> !(Z | N) */
- TCG_COND_GE, /* ge: !(N ^ V) -> !N */
- TCG_COND_NE, /* gtu: !(C | Z) -> !Z */
- TCG_COND_ALWAYS, /* geu: !C -> 1 */
- TCG_COND_GE, /* pos: !N */
- TCG_COND_ALWAYS, /* vc: !V -> 1 */
- };
-
TCGv t1, t2;
cmp->is_bool = false;
switch (dc->cc_op) {
- case CC_OP_LOGIC:
- cmp->cond = logic_cond[cond];
- do_compare_dst_0:
- cmp->c2 = tcg_constant_tl(0);
- if (TARGET_LONG_BITS == 32 || xcc) {
- cmp->c1 = cpu_cc_dst;
- } else {
- cmp->c1 = t1 = tcg_temp_new();
- tcg_gen_ext32s_tl(t1, cpu_cc_dst);
- }
- return;
-
case CC_OP_SUB:
switch (cond) {
case 6: /* neg */
case 14: /* pos */
cmp->cond = (cond == 6 ? TCG_COND_LT : TCG_COND_GE);
- goto do_compare_dst_0;
+ cmp->c2 = tcg_constant_tl(0);
+ if (TARGET_LONG_BITS == 32 || xcc) {
+ cmp->c1 = cpu_cc_dst;
+ } else {
+ cmp->c1 = t1 = tcg_temp_new();
+ tcg_gen_ext32s_tl(t1, cpu_cc_dst);
+ }
+ return;
case 7: /* overflow */
case 15: /* !overflow */
@@ -3652,7 +3628,8 @@ TRANS(NOP_v9, 64, trans_NOP, a)
static bool do_arith_int(DisasContext *dc, arg_r_r_ri_cc *a, int cc_op,
void (*func)(TCGv, TCGv, TCGv),
- void (*funci)(TCGv, TCGv, target_long))
+ void (*funci)(TCGv, TCGv, target_long),
+ bool logic_cc)
{
TCGv dst, src1;
@@ -3661,7 +3638,9 @@ static bool do_arith_int(DisasContext *dc, arg_r_r_ri_cc *a, int cc_op,
return false;
}
- if (a->cc) {
+ if (logic_cc) {
+ dst = cpu_cc_N;
+ } else if (a->cc && cc_op > CC_OP_FLAGS) {
dst = cpu_cc_dst;
} else {
dst = gen_dest_gpr(dc, a->rd);
@@ -3677,6 +3656,17 @@ static bool do_arith_int(DisasContext *dc, arg_r_r_ri_cc *a, int cc_op,
} else {
func(dst, src1, cpu_regs[a->rs2_or_imm]);
}
+
+ if (logic_cc) {
+ if (TARGET_LONG_BITS == 64) {
+ tcg_gen_mov_tl(cpu_icc_Z, cpu_cc_N);
+ tcg_gen_movi_tl(cpu_icc_C, 0);
+ }
+ tcg_gen_mov_tl(cpu_cc_Z, cpu_cc_N);
+ tcg_gen_movi_tl(cpu_cc_C, 0);
+ tcg_gen_movi_tl(cpu_cc_V, 0);
+ }
+
gen_store_gpr(dc, a->rd, dst);
if (a->cc) {
@@ -3693,16 +3683,16 @@ static bool do_arith(DisasContext *dc, arg_r_r_ri_cc *a, int cc_op,
{
if (a->cc) {
assert(cc_op >= 0);
- return do_arith_int(dc, a, cc_op, func_cc, NULL);
+ return do_arith_int(dc, a, cc_op, func_cc, NULL, false);
}
- return do_arith_int(dc, a, cc_op, func, funci);
+ return do_arith_int(dc, a, cc_op, func, funci, false);
}
static bool do_logic(DisasContext *dc, arg_r_r_ri_cc *a,
void (*func)(TCGv, TCGv, TCGv),
void (*funci)(TCGv, TCGv, target_long))
{
- return do_arith_int(dc, a, CC_OP_LOGIC, func, funci);
+ return do_arith_int(dc, a, CC_OP_FLAGS, func, funci, a->cc);
}
TRANS(ADD, ALL, do_arith, a, CC_OP_ADD,
@@ -3754,7 +3744,6 @@ static bool trans_ADDC(DisasContext *dc, arg_r_r_ri_cc *a)
{
switch (dc->cc_op) {
case CC_OP_DIV:
- case CC_OP_LOGIC:
/* Carry is known to be zero. Fall back to plain ADD. */
return do_arith(dc, a, CC_OP_ADD,
tcg_gen_add_tl, tcg_gen_addi_tl, gen_op_add_cc);
@@ -3778,7 +3767,6 @@ static bool trans_SUBC(DisasContext *dc, arg_r_r_ri_cc *a)
{
switch (dc->cc_op) {
case CC_OP_DIV:
- case CC_OP_LOGIC:
/* Carry is known to be zero. Fall back to plain SUB. */
return do_arith(dc, a, CC_OP_SUB,
tcg_gen_sub_tl, tcg_gen_subi_tl, gen_op_sub_cc);
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 04/21] target/sparc: Remove CC_OP_DIV
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (2 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 03/21] target/sparc: Remove CC_OP_LOGIC Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 05/21] target/sparc: Remove CC_OP_ADD, CC_OP_ADDX, CC_OP_TADD Richard Henderson
` (17 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Return both result and overflow from helper_[us]div.
Compute all flags explicitly in gen_op_[us]divcc.
Marginally improve the INT64_MIN special case in do_sdiv.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/cpu.h | 1 -
target/sparc/helper.h | 6 +--
target/sparc/cc_helper.c | 33 -------------
target/sparc/helper.c | 101 ++++++++++++++-------------------------
target/sparc/translate.c | 70 ++++++++++++++++++++++-----
5 files changed, 97 insertions(+), 114 deletions(-)
diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 202c34f7ca..b16d53b91f 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -150,7 +150,6 @@ enum {
enum {
CC_OP_DYNAMIC, /* must use dynamic code to get cc_op */
CC_OP_FLAGS, /* all cc are back in cc_*_[NZCV] registers */
- CC_OP_DIV, /* modify N, Z and V, C = 0*/
CC_OP_ADD, /* modify all flags, CC_DST = res, CC_SRC = src1 */
CC_OP_ADDX, /* modify all flags, CC_DST = res, CC_SRC = src1 */
CC_OP_TADD, /* modify all flags, CC_DST = res, CC_SRC = src1 */
diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index dd1721a340..a7b0079c3b 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -27,10 +27,8 @@ DEF_HELPER_FLAGS_2(tick_set_limit, TCG_CALL_NO_RWG, void, ptr, i64)
DEF_HELPER_1(debug, void, env)
DEF_HELPER_1(save, void, env)
DEF_HELPER_1(restore, void, env)
-DEF_HELPER_3(udiv, tl, env, tl, tl)
-DEF_HELPER_3(udiv_cc, tl, env, tl, tl)
-DEF_HELPER_3(sdiv, tl, env, tl, tl)
-DEF_HELPER_3(sdiv_cc, tl, env, tl, tl)
+DEF_HELPER_FLAGS_3(udiv, TCG_CALL_NO_WG, i64, env, tl, tl)
+DEF_HELPER_FLAGS_3(sdiv, TCG_CALL_NO_WG, i64, env, tl, tl)
DEF_HELPER_3(taddcctv, tl, env, tl, tl)
DEF_HELPER_3(tsubcctv, tl, env, tl, tl)
#ifdef TARGET_SPARC64
diff --git a/target/sparc/cc_helper.c b/target/sparc/cc_helper.c
index 1622300a14..5400dfec15 100644
--- a/target/sparc/cc_helper.c
+++ b/target/sparc/cc_helper.c
@@ -47,30 +47,6 @@ static inline uint32_t get_NZ_xcc(target_long dst)
}
#endif
-static inline uint32_t get_V_div_icc(target_ulong src2)
-{
- uint32_t ret = 0;
-
- if (src2 != 0) {
- ret = PSR_OVF;
- }
- return ret;
-}
-
-static uint32_t compute_all_div(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_icc(CC_DST);
- ret |= get_V_div_icc(CC_SRC2);
- return ret;
-}
-
-static uint32_t compute_C_div(CPUSPARCState *env)
-{
- return 0;
-}
-
static inline uint32_t get_C_add_icc(uint32_t dst, uint32_t src1)
{
uint32_t ret = 0;
@@ -378,13 +354,6 @@ static uint32_t compute_all_tsubtv(CPUSPARCState *env)
return ret;
}
-#ifdef TARGET_SPARC64
-static uint32_t compute_all_logic_xcc(CPUSPARCState *env)
-{
- return get_NZ_xcc(CC_DST);
-}
-#endif
-
typedef struct CCTable {
uint32_t (*compute_all)(CPUSPARCState *env); /* return all the flags */
uint32_t (*compute_c)(CPUSPARCState *env); /* return the C flag */
@@ -392,7 +361,6 @@ typedef struct CCTable {
static const CCTable icc_table[CC_OP_NB] = {
/* CC_OP_DYNAMIC should never happen */
- [CC_OP_DIV] = { compute_all_div, compute_C_div },
[CC_OP_ADD] = { compute_all_add, compute_C_add },
[CC_OP_ADDX] = { compute_all_addx, compute_C_addx },
[CC_OP_TADD] = { compute_all_tadd, compute_C_add },
@@ -406,7 +374,6 @@ static const CCTable icc_table[CC_OP_NB] = {
#ifdef TARGET_SPARC64
static const CCTable xcc_table[CC_OP_NB] = {
/* CC_OP_DYNAMIC should never happen */
- [CC_OP_DIV] = { compute_all_logic_xcc, compute_C_div },
[CC_OP_ADD] = { compute_all_add_xcc, compute_C_add_xcc },
[CC_OP_ADDX] = { compute_all_addx_xcc, compute_C_addx_xcc },
[CC_OP_TADD] = { compute_all_add_xcc, compute_C_add_xcc },
diff --git a/target/sparc/helper.c b/target/sparc/helper.c
index 2bcdc81d54..53eec693dd 100644
--- a/target/sparc/helper.c
+++ b/target/sparc/helper.c
@@ -81,79 +81,52 @@ void helper_tick_set_limit(void *opaque, uint64_t limit)
}
#endif
-static target_ulong do_udiv(CPUSPARCState *env, target_ulong a,
- target_ulong b, int cc, uintptr_t ra)
+uint64_t helper_udiv(CPUSPARCState *env, target_ulong a, target_ulong b)
{
- int overflow = 0;
- uint64_t x0;
- uint32_t x1;
+ uint64_t a64 = (uint32_t)a | ((uint64_t)env->y << 32);
+ uint32_t b32 = b;
+ uint32_t r;
- x0 = (a & 0xffffffff) | ((int64_t) (env->y) << 32);
- x1 = (b & 0xffffffff);
-
- if (x1 == 0) {
- cpu_raise_exception_ra(env, TT_DIV_ZERO, ra);
+ if (b32 == 0) {
+ cpu_raise_exception_ra(env, TT_DIV_ZERO, GETPC());
}
- x0 = x0 / x1;
- if (x0 > UINT32_MAX) {
- x0 = UINT32_MAX;
- overflow = 1;
+ a64 /= b32;
+ r = a64;
+ if (unlikely(a64 > UINT32_MAX)) {
+ return -1; /* r = UINT32_MAX, v = 1 */
+ }
+ return r;
+}
+
+uint64_t helper_sdiv(CPUSPARCState *env, target_ulong a, target_ulong b)
+{
+ int64_t a64 = (uint32_t)a | ((uint64_t)env->y << 32);
+ int32_t b32 = b;
+ int32_t r;
+
+ if (b32 == 0) {
+ cpu_raise_exception_ra(env, TT_DIV_ZERO, GETPC());
}
- if (cc) {
- env->cc_src2 = overflow;
- }
- return x0;
-}
-
-target_ulong helper_udiv(CPUSPARCState *env, target_ulong a, target_ulong b)
-{
- return do_udiv(env, a, b, 0, GETPC());
-}
-
-target_ulong helper_udiv_cc(CPUSPARCState *env, target_ulong a, target_ulong b)
-{
- return do_udiv(env, a, b, 1, GETPC());
-}
-
-static target_ulong do_sdiv(CPUSPARCState *env, target_ulong a,
- target_ulong b, int cc, uintptr_t ra)
-{
- int overflow = 0;
- int64_t x0;
- int32_t x1;
-
- x0 = (a & 0xffffffff) | ((int64_t) (env->y) << 32);
- x1 = (b & 0xffffffff);
-
- if (x1 == 0) {
- cpu_raise_exception_ra(env, TT_DIV_ZERO, ra);
- } else if (x1 == -1 && x0 == INT64_MIN) {
- x0 = INT32_MAX;
- overflow = 1;
- } else {
- x0 = x0 / x1;
- if ((int32_t) x0 != x0) {
- x0 = x0 < 0 ? INT32_MIN : INT32_MAX;
- overflow = 1;
- }
+ if (unlikely(a64 == INT64_MIN)) {
+ /*
+ * Special case INT64_MIN / -1 is required to avoid trap on x86 host.
+ * However, with a dividend of INT64_MIN, there is no 32-bit divisor
+ * which can yield a 32-bit result:
+ * INT64_MIN / INT32_MIN = 0x1_0000_0000
+ * INT64_MIN / INT32_MAX = -0x1_0000_0002
+ * Therefore we know we must overflow and saturate.
+ */
+ return (uint32_t)(b32 < 0 ? INT32_MAX : INT32_MIN) | (-1ull << 32);
}
- if (cc) {
- env->cc_src2 = overflow;
+ a64 /= b;
+ r = a64;
+ if (unlikely(r != a64)) {
+ return (uint32_t)(a64 < 0 ? INT32_MIN : INT32_MAX) | (-1ull << 32);
}
- return x0;
-}
-
-target_ulong helper_sdiv(CPUSPARCState *env, target_ulong a, target_ulong b)
-{
- return do_sdiv(env, a, b, 0, GETPC());
-}
-
-target_ulong helper_sdiv_cc(CPUSPARCState *env, target_ulong a, target_ulong b)
-{
- return do_sdiv(env, a, b, 1, GETPC());
+ return (uint32_t)r;
}
#ifdef TARGET_SPARC64
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index b11d89343b..fa4bad6d1f 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -693,22 +693,76 @@ static void gen_op_sdivx(TCGv dst, TCGv src1, TCGv src2)
static void gen_op_udiv(TCGv dst, TCGv src1, TCGv src2)
{
+#ifdef TARGET_SPARC64
gen_helper_udiv(dst, tcg_env, src1, src2);
+ tcg_gen_ext32u_tl(dst, dst);
+#else
+ TCGv_i64 t64 = tcg_temp_new_i64();
+ gen_helper_udiv(t64, tcg_env, src1, src2);
+ tcg_gen_trunc_i64_tl(dst, t64);
+#endif
}
static void gen_op_sdiv(TCGv dst, TCGv src1, TCGv src2)
{
+#ifdef TARGET_SPARC64
gen_helper_sdiv(dst, tcg_env, src1, src2);
+ tcg_gen_ext32s_tl(dst, dst);
+#else
+ TCGv_i64 t64 = tcg_temp_new_i64();
+ gen_helper_sdiv(t64, tcg_env, src1, src2);
+ tcg_gen_trunc_i64_tl(dst, t64);
+#endif
}
static void gen_op_udivcc(TCGv dst, TCGv src1, TCGv src2)
{
- gen_helper_udiv_cc(dst, tcg_env, src1, src2);
+ TCGv_i64 t64;
+
+#ifdef TARGET_SPARC64
+ t64 = cpu_cc_V;
+#else
+ t64 = tcg_temp_new_i64();
+#endif
+
+ gen_helper_udiv(t64, tcg_env, src1, src2);
+
+#ifdef TARGET_SPARC64
+ tcg_gen_ext32u_tl(cpu_cc_N, t64);
+ tcg_gen_shri_tl(cpu_cc_V, t64, 32);
+ tcg_gen_mov_tl(cpu_icc_Z, cpu_cc_N);
+ tcg_gen_movi_tl(cpu_icc_C, 0);
+#else
+ tcg_gen_extr_i64_tl(cpu_cc_N, cpu_cc_V, t64);
+#endif
+ tcg_gen_mov_tl(cpu_cc_Z, cpu_cc_N);
+ tcg_gen_movi_tl(cpu_cc_C, 0);
+ tcg_gen_mov_tl(dst, cpu_cc_N);
}
static void gen_op_sdivcc(TCGv dst, TCGv src1, TCGv src2)
{
- gen_helper_sdiv_cc(dst, tcg_env, src1, src2);
+ TCGv_i64 t64;
+
+#ifdef TARGET_SPARC64
+ t64 = cpu_cc_V;
+#else
+ t64 = tcg_temp_new_i64();
+#endif
+
+ gen_helper_sdiv(t64, tcg_env, src1, src2);
+
+#ifdef TARGET_SPARC64
+ tcg_gen_ext32s_tl(cpu_cc_N, t64);
+ tcg_gen_shri_tl(cpu_cc_V, t64, 32);
+ tcg_gen_mov_tl(cpu_icc_Z, cpu_cc_N);
+ tcg_gen_movi_tl(cpu_icc_C, 0);
+#else
+ tcg_gen_extr_i64_tl(cpu_cc_N, cpu_cc_V, t64);
+#endif
+ tcg_gen_mov_tl(cpu_cc_Z, cpu_cc_N);
+ tcg_gen_movi_tl(cpu_cc_C, 0);
+ tcg_gen_mov_tl(dst, cpu_cc_N);
}
static void gen_op_taddcctv(TCGv dst, TCGv src1, TCGv src2)
@@ -3717,8 +3771,8 @@ TRANS(SMUL, MUL, do_logic, a, gen_op_smul, NULL)
TRANS(UDIVX, 64, do_arith, a, -1, gen_op_udivx, NULL, NULL)
TRANS(SDIVX, 64, do_arith, a, -1, gen_op_sdivx, NULL, NULL)
-TRANS(UDIV, DIV, do_arith, a, CC_OP_DIV, gen_op_udiv, NULL, gen_op_udivcc)
-TRANS(SDIV, DIV, do_arith, a, CC_OP_DIV, gen_op_sdiv, NULL, gen_op_sdivcc)
+TRANS(UDIV, DIV, do_arith, a, CC_OP_FLAGS, gen_op_udiv, NULL, gen_op_udivcc)
+TRANS(SDIV, DIV, do_arith, a, CC_OP_FLAGS, gen_op_sdiv, NULL, gen_op_sdivcc)
/* TODO: Should have feature bit -- comes in with UltraSparc T2. */
TRANS(POPC, 64, do_arith, a, -1, gen_op_popc, NULL, NULL)
@@ -3743,10 +3797,6 @@ static bool trans_OR(DisasContext *dc, arg_r_r_ri_cc *a)
static bool trans_ADDC(DisasContext *dc, arg_r_r_ri_cc *a)
{
switch (dc->cc_op) {
- case CC_OP_DIV:
- /* Carry is known to be zero. Fall back to plain ADD. */
- return do_arith(dc, a, CC_OP_ADD,
- tcg_gen_add_tl, tcg_gen_addi_tl, gen_op_add_cc);
case CC_OP_ADD:
case CC_OP_TADD:
case CC_OP_TADDTV:
@@ -3766,10 +3816,6 @@ static bool trans_ADDC(DisasContext *dc, arg_r_r_ri_cc *a)
static bool trans_SUBC(DisasContext *dc, arg_r_r_ri_cc *a)
{
switch (dc->cc_op) {
- case CC_OP_DIV:
- /* Carry is known to be zero. Fall back to plain SUB. */
- return do_arith(dc, a, CC_OP_SUB,
- tcg_gen_sub_tl, tcg_gen_subi_tl, gen_op_sub_cc);
case CC_OP_ADD:
case CC_OP_TADD:
case CC_OP_TADDTV:
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 05/21] target/sparc: Remove CC_OP_ADD, CC_OP_ADDX, CC_OP_TADD
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (3 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 04/21] target/sparc: Remove CC_OP_DIV Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 06/21] target/sparc: Remove CC_OP_SUB, CC_OP_SUBX, CC_OP_TSUB Richard Henderson
` (16 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
These are all related and implementable with common code.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/cpu.h | 3 -
target/sparc/cc_helper.c | 92 ---------------
target/sparc/translate.c | 247 ++++++++++++++-------------------------
3 files changed, 87 insertions(+), 255 deletions(-)
diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index b16d53b91f..4ee8e2dc92 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -150,9 +150,6 @@ enum {
enum {
CC_OP_DYNAMIC, /* must use dynamic code to get cc_op */
CC_OP_FLAGS, /* all cc are back in cc_*_[NZCV] registers */
- CC_OP_ADD, /* modify all flags, CC_DST = res, CC_SRC = src1 */
- CC_OP_ADDX, /* modify all flags, CC_DST = res, CC_SRC = src1 */
- CC_OP_TADD, /* modify all flags, CC_DST = res, CC_SRC = src1 */
CC_OP_TADDTV, /* modify all flags except V, CC_DST = res, CC_SRC = src1 */
CC_OP_SUB, /* modify all flags, CC_DST = res, CC_SRC = src1 */
CC_OP_SUBX, /* modify all flags, CC_DST = res, CC_SRC = src1 */
diff --git a/target/sparc/cc_helper.c b/target/sparc/cc_helper.c
index 5400dfec15..55bac722d2 100644
--- a/target/sparc/cc_helper.c
+++ b/target/sparc/cc_helper.c
@@ -57,28 +57,6 @@ static inline uint32_t get_C_add_icc(uint32_t dst, uint32_t src1)
return ret;
}
-static inline uint32_t get_C_addx_icc(uint32_t dst, uint32_t src1,
- uint32_t src2)
-{
- uint32_t ret = 0;
-
- if (((src1 & src2) | (~dst & (src1 | src2))) & (1U << 31)) {
- ret = PSR_CARRY;
- }
- return ret;
-}
-
-static inline uint32_t get_V_add_icc(uint32_t dst, uint32_t src1,
- uint32_t src2)
-{
- uint32_t ret = 0;
-
- if (((src1 ^ src2 ^ -1) & (src1 ^ dst)) & (1U << 31)) {
- ret = PSR_OVF;
- }
- return ret;
-}
-
#ifdef TARGET_SPARC64
static inline uint32_t get_C_add_xcc(target_ulong dst, target_ulong src1)
{
@@ -90,17 +68,6 @@ static inline uint32_t get_C_add_xcc(target_ulong dst, target_ulong src1)
return ret;
}
-static inline uint32_t get_C_addx_xcc(target_ulong dst, target_ulong src1,
- target_ulong src2)
-{
- uint32_t ret = 0;
-
- if (((src1 & src2) | (~dst & (src1 | src2))) & (1ULL << 63)) {
- ret = PSR_CARRY;
- }
- return ret;
-}
-
static inline uint32_t get_V_add_xcc(target_ulong dst, target_ulong src1,
target_ulong src2)
{
@@ -128,53 +95,11 @@ static uint32_t compute_C_add_xcc(CPUSPARCState *env)
}
#endif
-static uint32_t compute_all_add(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_icc(CC_DST);
- ret |= get_C_add_icc(CC_DST, CC_SRC);
- ret |= get_V_add_icc(CC_DST, CC_SRC, CC_SRC2);
- return ret;
-}
-
static uint32_t compute_C_add(CPUSPARCState *env)
{
return get_C_add_icc(CC_DST, CC_SRC);
}
-#ifdef TARGET_SPARC64
-static uint32_t compute_all_addx_xcc(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_xcc(CC_DST);
- ret |= get_C_addx_xcc(CC_DST, CC_SRC, CC_SRC2);
- ret |= get_V_add_xcc(CC_DST, CC_SRC, CC_SRC2);
- return ret;
-}
-
-static uint32_t compute_C_addx_xcc(CPUSPARCState *env)
-{
- return get_C_addx_xcc(CC_DST, CC_SRC, CC_SRC2);
-}
-#endif
-
-static uint32_t compute_all_addx(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_icc(CC_DST);
- ret |= get_C_addx_icc(CC_DST, CC_SRC, CC_SRC2);
- ret |= get_V_add_icc(CC_DST, CC_SRC, CC_SRC2);
- return ret;
-}
-
-static uint32_t compute_C_addx(CPUSPARCState *env)
-{
- return get_C_addx_icc(CC_DST, CC_SRC, CC_SRC2);
-}
-
static inline uint32_t get_V_tag_icc(target_ulong src1, target_ulong src2)
{
uint32_t ret = 0;
@@ -185,17 +110,6 @@ static inline uint32_t get_V_tag_icc(target_ulong src1, target_ulong src2)
return ret;
}
-static uint32_t compute_all_tadd(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_icc(CC_DST);
- ret |= get_C_add_icc(CC_DST, CC_SRC);
- ret |= get_V_add_icc(CC_DST, CC_SRC, CC_SRC2);
- ret |= get_V_tag_icc(CC_SRC, CC_SRC2);
- return ret;
-}
-
static uint32_t compute_all_taddtv(CPUSPARCState *env)
{
uint32_t ret;
@@ -361,9 +275,6 @@ typedef struct CCTable {
static const CCTable icc_table[CC_OP_NB] = {
/* CC_OP_DYNAMIC should never happen */
- [CC_OP_ADD] = { compute_all_add, compute_C_add },
- [CC_OP_ADDX] = { compute_all_addx, compute_C_addx },
- [CC_OP_TADD] = { compute_all_tadd, compute_C_add },
[CC_OP_TADDTV] = { compute_all_taddtv, compute_C_add },
[CC_OP_SUB] = { compute_all_sub, compute_C_sub },
[CC_OP_SUBX] = { compute_all_subx, compute_C_subx },
@@ -374,9 +285,6 @@ static const CCTable icc_table[CC_OP_NB] = {
#ifdef TARGET_SPARC64
static const CCTable xcc_table[CC_OP_NB] = {
/* CC_OP_DYNAMIC should never happen */
- [CC_OP_ADD] = { compute_all_add_xcc, compute_C_add_xcc },
- [CC_OP_ADDX] = { compute_all_addx_xcc, compute_C_addx_xcc },
- [CC_OP_TADD] = { compute_all_add_xcc, compute_C_add_xcc },
[CC_OP_TADDTV] = { compute_all_add_xcc, compute_C_add_xcc },
[CC_OP_SUB] = { compute_all_sub_xcc, compute_C_sub_xcc },
[CC_OP_SUBX] = { compute_all_subx_xcc, compute_C_subx_xcc },
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index fa4bad6d1f..cf121a237d 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -382,33 +382,71 @@ static void gen_goto_tb(DisasContext *s, int tb_num,
}
}
-static void gen_op_add_cc(TCGv dst, TCGv src1, TCGv src2)
+static TCGv gen_carry32(void)
{
- tcg_gen_mov_tl(cpu_cc_src, src1);
- tcg_gen_mov_tl(cpu_cc_src2, src2);
- tcg_gen_add_tl(cpu_cc_dst, cpu_cc_src, cpu_cc_src2);
- tcg_gen_mov_tl(dst, cpu_cc_dst);
+ if (TARGET_LONG_BITS == 64) {
+ TCGv t = tcg_temp_new();
+ tcg_gen_extract_tl(t, cpu_icc_C, 32, 1);
+ return t;
+ }
+ return cpu_icc_C;
}
-static TCGv_i32 gen_add32_carry32(void)
+static void gen_op_addcc_int(TCGv dst, TCGv src1, TCGv src2, TCGv cin)
{
- TCGv_i32 carry_32, cc_src1_32, cc_src2_32;
+ TCGv z = tcg_constant_tl(0);
- /* Carry is computed from a previous add: (dst < src) */
-#if TARGET_LONG_BITS == 64
- cc_src1_32 = tcg_temp_new_i32();
- cc_src2_32 = tcg_temp_new_i32();
- tcg_gen_extrl_i64_i32(cc_src1_32, cpu_cc_dst);
- tcg_gen_extrl_i64_i32(cc_src2_32, cpu_cc_src);
-#else
- cc_src1_32 = cpu_cc_dst;
- cc_src2_32 = cpu_cc_src;
-#endif
+ if (cin) {
+ tcg_gen_add2_tl(cpu_cc_N, cpu_cc_C, src1, z, cin, z);
+ tcg_gen_add2_tl(cpu_cc_N, cpu_cc_C, cpu_cc_N, cpu_cc_C, src2, z);
+ } else {
+ tcg_gen_add2_tl(cpu_cc_N, cpu_cc_C, src1, z, src2, z);
+ }
+ tcg_gen_xor_tl(cpu_cc_Z, src1, src2);
+ tcg_gen_xor_tl(cpu_cc_V, cpu_cc_N, src2);
+ tcg_gen_andc_tl(cpu_cc_V, cpu_cc_V, cpu_cc_Z);
+ if (TARGET_LONG_BITS == 64) {
+ /*
+ * Carry-in to bit 32 is result ^ src1 ^ src2.
+ * We already have the src xor term in Z, from computation of V.
+ */
+ tcg_gen_xor_tl(cpu_icc_C, cpu_cc_Z, cpu_cc_N);
+ tcg_gen_mov_tl(cpu_icc_Z, cpu_cc_N);
+ }
+ tcg_gen_mov_tl(cpu_cc_Z, cpu_cc_N);
+ tcg_gen_mov_tl(dst, cpu_cc_N);
+}
- carry_32 = tcg_temp_new_i32();
- tcg_gen_setcond_i32(TCG_COND_LTU, carry_32, cc_src1_32, cc_src2_32);
+static void gen_op_addcc(TCGv dst, TCGv src1, TCGv src2)
+{
+ gen_op_addcc_int(dst, src1, src2, NULL);
+}
- return carry_32;
+static void gen_op_taddcc(TCGv dst, TCGv src1, TCGv src2)
+{
+ TCGv t = tcg_temp_new();
+
+ /* Save the tag bits around modification of dst. */
+ tcg_gen_or_tl(t, src1, src2);
+
+ gen_op_addcc(dst, src1, src2);
+
+ /* Incorprate tag bits into icc.V */
+ tcg_gen_andi_tl(t, t, 3);
+ tcg_gen_neg_tl(t, t);
+ tcg_gen_ext32u_tl(t, t);
+ tcg_gen_or_tl(cpu_cc_V, cpu_cc_V, t);
+}
+
+static void gen_op_addc(TCGv dst, TCGv src1, TCGv src2)
+{
+ tcg_gen_add_tl(dst, src1, src2);
+ tcg_gen_add_tl(dst, dst, gen_carry32());
+}
+
+static void gen_op_addccc(TCGv dst, TCGv src1, TCGv src2)
+{
+ gen_op_addcc_int(dst, src1, src2, gen_carry32());
}
static TCGv_i32 gen_sub32_carry32(void)
@@ -432,89 +470,6 @@ static TCGv_i32 gen_sub32_carry32(void)
return carry_32;
}
-static void gen_op_addc_int(TCGv dst, TCGv src1, TCGv src2,
- TCGv_i32 carry_32, bool update_cc)
-{
- tcg_gen_add_tl(dst, src1, src2);
-
-#ifdef TARGET_SPARC64
- TCGv carry = tcg_temp_new();
- tcg_gen_extu_i32_tl(carry, carry_32);
- tcg_gen_add_tl(dst, dst, carry);
-#else
- tcg_gen_add_i32(dst, dst, carry_32);
-#endif
-
- if (update_cc) {
- tcg_debug_assert(dst == cpu_cc_dst);
- tcg_gen_mov_tl(cpu_cc_src, src1);
- tcg_gen_mov_tl(cpu_cc_src2, src2);
- }
-}
-
-static void gen_op_addc_int_add(TCGv dst, TCGv src1, TCGv src2, bool update_cc)
-{
- TCGv discard;
-
- if (TARGET_LONG_BITS == 64) {
- gen_op_addc_int(dst, src1, src2, gen_add32_carry32(), update_cc);
- return;
- }
-
- /*
- * We can re-use the host's hardware carry generation by using
- * an ADD2 opcode. We discard the low part of the output.
- * Ideally we'd combine this operation with the add that
- * generated the carry in the first place.
- */
- discard = tcg_temp_new();
- tcg_gen_add2_tl(discard, dst, cpu_cc_src, src1, cpu_cc_src2, src2);
-
- if (update_cc) {
- tcg_debug_assert(dst == cpu_cc_dst);
- tcg_gen_mov_tl(cpu_cc_src, src1);
- tcg_gen_mov_tl(cpu_cc_src2, src2);
- }
-}
-
-static void gen_op_addc_add(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_addc_int_add(dst, src1, src2, false);
-}
-
-static void gen_op_addccc_add(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_addc_int_add(dst, src1, src2, true);
-}
-
-static void gen_op_addc_sub(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_addc_int(dst, src1, src2, gen_sub32_carry32(), false);
-}
-
-static void gen_op_addccc_sub(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_addc_int(dst, src1, src2, gen_sub32_carry32(), true);
-}
-
-static void gen_op_addc_int_generic(TCGv dst, TCGv src1, TCGv src2,
- bool update_cc)
-{
- TCGv_i32 carry_32 = tcg_temp_new_i32();
- gen_helper_compute_C_icc(carry_32, tcg_env);
- gen_op_addc_int(dst, src1, src2, carry_32, update_cc);
-}
-
-static void gen_op_addc_generic(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_addc_int_generic(dst, src1, src2, false);
-}
-
-static void gen_op_addccc_generic(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_addc_int_generic(dst, src1, src2, true);
-}
-
static void gen_op_sub_cc(TCGv dst, TCGv src1, TCGv src2)
{
tcg_gen_mov_tl(cpu_cc_src, src1);
@@ -545,16 +500,6 @@ static void gen_op_subc_int(TCGv dst, TCGv src1, TCGv src2,
}
}
-static void gen_op_subc_add(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_subc_int(dst, src1, src2, gen_add32_carry32(), false);
-}
-
-static void gen_op_subccc_add(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_subc_int(dst, src1, src2, gen_add32_carry32(), true);
-}
-
static void gen_op_subc_int_sub(TCGv dst, TCGv src1, TCGv src2, bool update_cc)
{
TCGv discard;
@@ -609,39 +554,39 @@ static void gen_op_subccc_generic(TCGv dst, TCGv src1, TCGv src2)
static void gen_op_mulscc(TCGv dst, TCGv src1, TCGv src2)
{
- TCGv r_temp, zero, t0;
+ TCGv zero = tcg_constant_tl(0);
+ TCGv t_src1 = tcg_temp_new();
+ TCGv t_src2 = tcg_temp_new();
+ TCGv t0 = tcg_temp_new();
- r_temp = tcg_temp_new();
- t0 = tcg_temp_new();
+ tcg_gen_ext32u_tl(t_src1, src1);
+ tcg_gen_ext32u_tl(t_src2, src2);
- /* old op:
- if (!(env->y & 1))
- T1 = 0;
- */
- zero = tcg_constant_tl(0);
- tcg_gen_andi_tl(cpu_cc_src, src1, 0xffffffff);
- tcg_gen_andi_tl(r_temp, cpu_y, 0x1);
- tcg_gen_andi_tl(cpu_cc_src2, src2, 0xffffffff);
- tcg_gen_movcond_tl(TCG_COND_EQ, cpu_cc_src2, r_temp, zero,
- zero, cpu_cc_src2);
+ /*
+ * if (!(env->y & 1))
+ * src2 = 0;
+ */
+ tcg_gen_andi_tl(t0, cpu_y, 0x1);
+ tcg_gen_movcond_tl(TCG_COND_EQ, t_src2, t0, zero, zero, t_src2);
- // b2 = T0 & 1;
- // env->y = (b2 << 31) | (env->y >> 1);
+ /*
+ * b2 = src1 & 1;
+ * y = (b2 << 31) | (y >> 1);
+ */
tcg_gen_extract_tl(t0, cpu_y, 1, 31);
- tcg_gen_deposit_tl(cpu_y, t0, cpu_cc_src, 31, 1);
+ tcg_gen_deposit_tl(cpu_y, t0, src1, 31, 1);
// b1 = N ^ V;
tcg_gen_xor_tl(t0, cpu_cc_N, cpu_cc_V);
- // T0 = (b1 << 31) | (T0 >> 1);
- // src1 = T0;
+ /*
+ * src1 = (b1 << 31) | (src1 >> 1)
+ */
tcg_gen_andi_tl(t0, t0, 1u << 31);
- tcg_gen_shri_tl(cpu_cc_src, cpu_cc_src, 1);
- tcg_gen_or_tl(cpu_cc_src, cpu_cc_src, t0);
+ tcg_gen_shri_tl(t_src1, t_src1, 1);
+ tcg_gen_or_tl(t_src1, t_src1, t0);
- tcg_gen_add_tl(cpu_cc_dst, cpu_cc_src, cpu_cc_src2);
-
- tcg_gen_mov_tl(dst, cpu_cc_dst);
+ gen_op_addcc(dst, t_src1, t_src2);
}
static void gen_op_multiply(TCGv dst, TCGv src1, TCGv src2, int sign_ext)
@@ -3749,12 +3694,12 @@ static bool do_logic(DisasContext *dc, arg_r_r_ri_cc *a,
return do_arith_int(dc, a, CC_OP_FLAGS, func, funci, a->cc);
}
-TRANS(ADD, ALL, do_arith, a, CC_OP_ADD,
- tcg_gen_add_tl, tcg_gen_addi_tl, gen_op_add_cc)
+TRANS(ADD, ALL, do_arith, a, CC_OP_FLAGS,
+ tcg_gen_add_tl, tcg_gen_addi_tl, gen_op_addcc)
TRANS(SUB, ALL, do_arith, a, CC_OP_SUB,
tcg_gen_sub_tl, tcg_gen_subi_tl, gen_op_sub_cc)
-TRANS(TADDcc, ALL, do_arith, a, CC_OP_TADD, NULL, NULL, gen_op_add_cc)
+TRANS(TADDcc, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_taddcc)
TRANS(TSUBcc, ALL, do_arith, a, CC_OP_TSUB, NULL, NULL, gen_op_sub_cc)
TRANS(TADDccTV, ALL, do_arith, a, CC_OP_TADDTV, NULL, NULL, gen_op_taddcctv)
TRANS(TSUBccTV, ALL, do_arith, a, CC_OP_TSUBTV, NULL, NULL, gen_op_tsubcctv)
@@ -3796,31 +3741,13 @@ static bool trans_OR(DisasContext *dc, arg_r_r_ri_cc *a)
static bool trans_ADDC(DisasContext *dc, arg_r_r_ri_cc *a)
{
- switch (dc->cc_op) {
- case CC_OP_ADD:
- case CC_OP_TADD:
- case CC_OP_TADDTV:
- return do_arith(dc, a, CC_OP_ADDX,
- gen_op_addc_add, NULL, gen_op_addccc_add);
- case CC_OP_SUB:
- case CC_OP_TSUB:
- case CC_OP_TSUBTV:
- return do_arith(dc, a, CC_OP_ADDX,
- gen_op_addc_sub, NULL, gen_op_addccc_sub);
- default:
- return do_arith(dc, a, CC_OP_ADDX,
- gen_op_addc_generic, NULL, gen_op_addccc_generic);
- }
+ update_psr(dc);
+ return do_arith(dc, a, CC_OP_FLAGS, gen_op_addc, NULL, gen_op_addccc);
}
static bool trans_SUBC(DisasContext *dc, arg_r_r_ri_cc *a)
{
switch (dc->cc_op) {
- case CC_OP_ADD:
- case CC_OP_TADD:
- case CC_OP_TADDTV:
- return do_arith(dc, a, CC_OP_SUBX,
- gen_op_subc_add, NULL, gen_op_subccc_add);
case CC_OP_SUB:
case CC_OP_TSUB:
case CC_OP_TSUBTV:
@@ -3835,7 +3762,7 @@ static bool trans_SUBC(DisasContext *dc, arg_r_r_ri_cc *a)
static bool trans_MULScc(DisasContext *dc, arg_r_r_ri_cc *a)
{
update_psr(dc);
- return do_arith(dc, a, CC_OP_ADD, NULL, NULL, gen_op_mulscc);
+ return do_arith(dc, a, CC_OP_FLAGS, NULL, NULL, gen_op_mulscc);
}
static bool gen_edge(DisasContext *dc, arg_r_r_r *a,
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 06/21] target/sparc: Remove CC_OP_SUB, CC_OP_SUBX, CC_OP_TSUB
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (4 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 05/21] target/sparc: Remove CC_OP_ADD, CC_OP_ADDX, CC_OP_TADD Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 07/21] target/sparc: Remove CC_OP_TADDTV, CC_OP_TSUBTV Richard Henderson
` (15 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
These are all related and implementable with common code.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/cpu.h | 3 -
target/sparc/cc_helper.c | 103 --------------------
target/sparc/translate.c | 203 +++++++++------------------------------
3 files changed, 45 insertions(+), 264 deletions(-)
diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 4ee8e2dc92..9884bd416a 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -151,9 +151,6 @@ enum {
CC_OP_DYNAMIC, /* must use dynamic code to get cc_op */
CC_OP_FLAGS, /* all cc are back in cc_*_[NZCV] registers */
CC_OP_TADDTV, /* modify all flags except V, CC_DST = res, CC_SRC = src1 */
- CC_OP_SUB, /* modify all flags, CC_DST = res, CC_SRC = src1 */
- CC_OP_SUBX, /* modify all flags, CC_DST = res, CC_SRC = src1 */
- CC_OP_TSUB, /* modify all flags, CC_DST = res, CC_SRC = src1 */
CC_OP_TSUBTV, /* modify all flags except V, CC_DST = res, CC_SRC = src1 */
CC_OP_NB,
};
diff --git a/target/sparc/cc_helper.c b/target/sparc/cc_helper.c
index 55bac722d2..20d451aa65 100644
--- a/target/sparc/cc_helper.c
+++ b/target/sparc/cc_helper.c
@@ -100,16 +100,6 @@ static uint32_t compute_C_add(CPUSPARCState *env)
return get_C_add_icc(CC_DST, CC_SRC);
}
-static inline uint32_t get_V_tag_icc(target_ulong src1, target_ulong src2)
-{
- uint32_t ret = 0;
-
- if ((src1 | src2) & 0x3) {
- ret = PSR_OVF;
- }
- return ret;
-}
-
static uint32_t compute_all_taddtv(CPUSPARCState *env)
{
uint32_t ret;
@@ -129,29 +119,6 @@ static inline uint32_t get_C_sub_icc(uint32_t src1, uint32_t src2)
return ret;
}
-static inline uint32_t get_C_subx_icc(uint32_t dst, uint32_t src1,
- uint32_t src2)
-{
- uint32_t ret = 0;
-
- if (((~src1 & src2) | (dst & (~src1 | src2))) & (1U << 31)) {
- ret = PSR_CARRY;
- }
- return ret;
-}
-
-static inline uint32_t get_V_sub_icc(uint32_t dst, uint32_t src1,
- uint32_t src2)
-{
- uint32_t ret = 0;
-
- if (((src1 ^ src2) & (src1 ^ dst)) & (1U << 31)) {
- ret = PSR_OVF;
- }
- return ret;
-}
-
-
#ifdef TARGET_SPARC64
static inline uint32_t get_C_sub_xcc(target_ulong src1, target_ulong src2)
{
@@ -163,17 +130,6 @@ static inline uint32_t get_C_sub_xcc(target_ulong src1, target_ulong src2)
return ret;
}
-static inline uint32_t get_C_subx_xcc(target_ulong dst, target_ulong src1,
- target_ulong src2)
-{
- uint32_t ret = 0;
-
- if (((~src1 & src2) | (dst & (~src1 | src2))) & (1ULL << 63)) {
- ret = PSR_CARRY;
- }
- return ret;
-}
-
static inline uint32_t get_V_sub_xcc(target_ulong dst, target_ulong src1,
target_ulong src2)
{
@@ -201,64 +157,11 @@ static uint32_t compute_C_sub_xcc(CPUSPARCState *env)
}
#endif
-static uint32_t compute_all_sub(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_icc(CC_DST);
- ret |= get_C_sub_icc(CC_SRC, CC_SRC2);
- ret |= get_V_sub_icc(CC_DST, CC_SRC, CC_SRC2);
- return ret;
-}
-
static uint32_t compute_C_sub(CPUSPARCState *env)
{
return get_C_sub_icc(CC_SRC, CC_SRC2);
}
-#ifdef TARGET_SPARC64
-static uint32_t compute_all_subx_xcc(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_xcc(CC_DST);
- ret |= get_C_subx_xcc(CC_DST, CC_SRC, CC_SRC2);
- ret |= get_V_sub_xcc(CC_DST, CC_SRC, CC_SRC2);
- return ret;
-}
-
-static uint32_t compute_C_subx_xcc(CPUSPARCState *env)
-{
- return get_C_subx_xcc(CC_DST, CC_SRC, CC_SRC2);
-}
-#endif
-
-static uint32_t compute_all_subx(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_icc(CC_DST);
- ret |= get_C_subx_icc(CC_DST, CC_SRC, CC_SRC2);
- ret |= get_V_sub_icc(CC_DST, CC_SRC, CC_SRC2);
- return ret;
-}
-
-static uint32_t compute_C_subx(CPUSPARCState *env)
-{
- return get_C_subx_icc(CC_DST, CC_SRC, CC_SRC2);
-}
-
-static uint32_t compute_all_tsub(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_icc(CC_DST);
- ret |= get_C_sub_icc(CC_SRC, CC_SRC2);
- ret |= get_V_sub_icc(CC_DST, CC_SRC, CC_SRC2);
- ret |= get_V_tag_icc(CC_SRC, CC_SRC2);
- return ret;
-}
-
static uint32_t compute_all_tsubtv(CPUSPARCState *env)
{
uint32_t ret;
@@ -276,9 +179,6 @@ typedef struct CCTable {
static const CCTable icc_table[CC_OP_NB] = {
/* CC_OP_DYNAMIC should never happen */
[CC_OP_TADDTV] = { compute_all_taddtv, compute_C_add },
- [CC_OP_SUB] = { compute_all_sub, compute_C_sub },
- [CC_OP_SUBX] = { compute_all_subx, compute_C_subx },
- [CC_OP_TSUB] = { compute_all_tsub, compute_C_sub },
[CC_OP_TSUBTV] = { compute_all_tsubtv, compute_C_sub },
};
@@ -286,9 +186,6 @@ static const CCTable icc_table[CC_OP_NB] = {
static const CCTable xcc_table[CC_OP_NB] = {
/* CC_OP_DYNAMIC should never happen */
[CC_OP_TADDTV] = { compute_all_add_xcc, compute_C_add_xcc },
- [CC_OP_SUB] = { compute_all_sub_xcc, compute_C_sub_xcc },
- [CC_OP_SUBX] = { compute_all_subx_xcc, compute_C_subx_xcc },
- [CC_OP_TSUB] = { compute_all_sub_xcc, compute_C_sub_xcc },
[CC_OP_TSUBTV] = { compute_all_sub_xcc, compute_C_sub_xcc },
};
#endif
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index cf121a237d..d119ce4c94 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -449,107 +449,58 @@ static void gen_op_addccc(TCGv dst, TCGv src1, TCGv src2)
gen_op_addcc_int(dst, src1, src2, gen_carry32());
}
-static TCGv_i32 gen_sub32_carry32(void)
+static void gen_op_subcc_int(TCGv dst, TCGv src1, TCGv src2, TCGv cin)
{
- TCGv_i32 carry_32, cc_src1_32, cc_src2_32;
+ TCGv z = tcg_constant_tl(0);
- /* Carry is computed from a previous borrow: (src1 < src2) */
-#if TARGET_LONG_BITS == 64
- cc_src1_32 = tcg_temp_new_i32();
- cc_src2_32 = tcg_temp_new_i32();
- tcg_gen_extrl_i64_i32(cc_src1_32, cpu_cc_src);
- tcg_gen_extrl_i64_i32(cc_src2_32, cpu_cc_src2);
-#else
- cc_src1_32 = cpu_cc_src;
- cc_src2_32 = cpu_cc_src2;
+ if (cin) {
+ tcg_gen_sub2_tl(cpu_cc_N, cpu_cc_C, src1, z, cin, z);
+ tcg_gen_sub2_tl(cpu_cc_N, cpu_cc_C, cpu_cc_N, cpu_cc_C, src2, z);
+ } else {
+ tcg_gen_sub2_tl(cpu_cc_N, cpu_cc_C, src1, z, src2, z);
+ }
+ tcg_gen_neg_tl(cpu_cc_C, cpu_cc_C);
+ tcg_gen_xor_tl(cpu_cc_Z, src1, src2);
+ tcg_gen_xor_tl(cpu_cc_V, cpu_cc_N, src1);
+ tcg_gen_and_tl(cpu_cc_V, cpu_cc_V, cpu_cc_Z);
+#ifdef TARGET_SPARC64
+ tcg_gen_xor_tl(cpu_icc_C, cpu_cc_Z, cpu_cc_N);
+ tcg_gen_mov_tl(cpu_icc_Z, cpu_cc_N);
#endif
-
- carry_32 = tcg_temp_new_i32();
- tcg_gen_setcond_i32(TCG_COND_LTU, carry_32, cc_src1_32, cc_src2_32);
-
- return carry_32;
+ tcg_gen_mov_tl(cpu_cc_Z, cpu_cc_N);
+ tcg_gen_mov_tl(dst, cpu_cc_N);
}
-static void gen_op_sub_cc(TCGv dst, TCGv src1, TCGv src2)
+static void gen_op_subcc(TCGv dst, TCGv src1, TCGv src2)
{
- tcg_gen_mov_tl(cpu_cc_src, src1);
- tcg_gen_mov_tl(cpu_cc_src2, src2);
- tcg_gen_sub_tl(cpu_cc_dst, cpu_cc_src, cpu_cc_src2);
- tcg_gen_mov_tl(dst, cpu_cc_dst);
+ gen_op_subcc_int(dst, src1, src2, NULL);
}
-static void gen_op_subc_int(TCGv dst, TCGv src1, TCGv src2,
- TCGv_i32 carry_32, bool update_cc)
+static void gen_op_tsubcc(TCGv dst, TCGv src1, TCGv src2)
{
- TCGv carry;
+ TCGv t = tcg_temp_new();
-#if TARGET_LONG_BITS == 64
- carry = tcg_temp_new();
- tcg_gen_extu_i32_i64(carry, carry_32);
-#else
- carry = carry_32;
-#endif
+ /* Save the tag bits around modification of dst. */
+ tcg_gen_or_tl(t, src1, src2);
+ gen_op_subcc(dst, src1, src2);
+
+ /* Incorprate tag bits into icc.V */
+ tcg_gen_andi_tl(t, t, 3);
+ tcg_gen_neg_tl(t, t);
+ tcg_gen_ext32u_tl(t, t);
+ tcg_gen_or_tl(cpu_cc_V, cpu_cc_V, t);
+}
+
+static void gen_op_subc(TCGv dst, TCGv src1, TCGv src2)
+{
tcg_gen_sub_tl(dst, src1, src2);
- tcg_gen_sub_tl(dst, dst, carry);
-
- if (update_cc) {
- tcg_debug_assert(dst == cpu_cc_dst);
- tcg_gen_mov_tl(cpu_cc_src, src1);
- tcg_gen_mov_tl(cpu_cc_src2, src2);
- }
+ tcg_gen_sub_tl(dst, dst, gen_carry32());
}
-static void gen_op_subc_int_sub(TCGv dst, TCGv src1, TCGv src2, bool update_cc)
+static void gen_op_subccc(TCGv dst, TCGv src1, TCGv src2)
{
- TCGv discard;
-
- if (TARGET_LONG_BITS == 64) {
- gen_op_subc_int(dst, src1, src2, gen_sub32_carry32(), update_cc);
- return;
- }
-
- /*
- * We can re-use the host's hardware carry generation by using
- * a SUB2 opcode. We discard the low part of the output.
- */
- discard = tcg_temp_new();
- tcg_gen_sub2_tl(discard, dst, cpu_cc_src, src1, cpu_cc_src2, src2);
-
- if (update_cc) {
- tcg_debug_assert(dst == cpu_cc_dst);
- tcg_gen_mov_tl(cpu_cc_src, src1);
- tcg_gen_mov_tl(cpu_cc_src2, src2);
- }
-}
-
-static void gen_op_subc_sub(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_subc_int_sub(dst, src1, src2, false);
-}
-
-static void gen_op_subccc_sub(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_subc_int_sub(dst, src1, src2, true);
-}
-
-static void gen_op_subc_int_generic(TCGv dst, TCGv src1, TCGv src2,
- bool update_cc)
-{
- TCGv_i32 carry_32 = tcg_temp_new_i32();
-
- gen_helper_compute_C_icc(carry_32, tcg_env);
- gen_op_subc_int(dst, src1, src2, carry_32, update_cc);
-}
-
-static void gen_op_subc_generic(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_subc_int_generic(dst, src1, src2, false);
-}
-
-static void gen_op_subccc_generic(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_op_subc_int_generic(dst, src1, src2, true);
+ gen_op_subcc_int(dst, src1, src2, gen_carry32());
}
static void gen_op_mulscc(TCGv dst, TCGv src1, TCGv src2)
@@ -1097,65 +1048,11 @@ static void gen_op_next_insn(void)
static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
DisasContext *dc)
{
- static int subcc_cond[16] = {
- TCG_COND_NEVER,
- TCG_COND_EQ,
- TCG_COND_LE,
- TCG_COND_LT,
- TCG_COND_LEU,
- TCG_COND_LTU,
- -1, /* neg */
- -1, /* overflow */
- TCG_COND_ALWAYS,
- TCG_COND_NE,
- TCG_COND_GT,
- TCG_COND_GE,
- TCG_COND_GTU,
- TCG_COND_GEU,
- -1, /* pos */
- -1, /* no overflow */
- };
-
TCGv t1, t2;
cmp->is_bool = false;
switch (dc->cc_op) {
- case CC_OP_SUB:
- switch (cond) {
- case 6: /* neg */
- case 14: /* pos */
- cmp->cond = (cond == 6 ? TCG_COND_LT : TCG_COND_GE);
- cmp->c2 = tcg_constant_tl(0);
- if (TARGET_LONG_BITS == 32 || xcc) {
- cmp->c1 = cpu_cc_dst;
- } else {
- cmp->c1 = t1 = tcg_temp_new();
- tcg_gen_ext32s_tl(t1, cpu_cc_dst);
- }
- return;
-
- case 7: /* overflow */
- case 15: /* !overflow */
- break;
-
- default:
- cmp->cond = subcc_cond[cond];
- if (TARGET_LONG_BITS == 32 || xcc) {
- cmp->c1 = cpu_cc_src;
- cmp->c2 = cpu_cc_src2;
- } else {
- /* Note that sign-extension works for unsigned compares as
- long as both operands are sign-extended. */
- cmp->c1 = t1 = tcg_temp_new();
- tcg_gen_ext32s_tl(t1, cpu_cc_src);
- cmp->c2 = t2 = tcg_temp_new();
- tcg_gen_ext32s_tl(t2, cpu_cc_src2);
- }
- return;
- }
- break;
-
default:
gen_helper_compute_psr(tcg_env);
dc->cc_op = CC_OP_FLAGS;
@@ -3696,11 +3593,11 @@ static bool do_logic(DisasContext *dc, arg_r_r_ri_cc *a,
TRANS(ADD, ALL, do_arith, a, CC_OP_FLAGS,
tcg_gen_add_tl, tcg_gen_addi_tl, gen_op_addcc)
-TRANS(SUB, ALL, do_arith, a, CC_OP_SUB,
- tcg_gen_sub_tl, tcg_gen_subi_tl, gen_op_sub_cc)
+TRANS(SUB, ALL, do_arith, a, CC_OP_FLAGS,
+ tcg_gen_sub_tl, tcg_gen_subi_tl, gen_op_subcc)
TRANS(TADDcc, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_taddcc)
-TRANS(TSUBcc, ALL, do_arith, a, CC_OP_TSUB, NULL, NULL, gen_op_sub_cc)
+TRANS(TSUBcc, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_tsubcc)
TRANS(TADDccTV, ALL, do_arith, a, CC_OP_TADDTV, NULL, NULL, gen_op_taddcctv)
TRANS(TSUBccTV, ALL, do_arith, a, CC_OP_TSUBTV, NULL, NULL, gen_op_tsubcctv)
@@ -3747,16 +3644,8 @@ static bool trans_ADDC(DisasContext *dc, arg_r_r_ri_cc *a)
static bool trans_SUBC(DisasContext *dc, arg_r_r_ri_cc *a)
{
- switch (dc->cc_op) {
- case CC_OP_SUB:
- case CC_OP_TSUB:
- case CC_OP_TSUBTV:
- return do_arith(dc, a, CC_OP_SUBX,
- gen_op_subc_sub, NULL, gen_op_subccc_sub);
- default:
- return do_arith(dc, a, CC_OP_SUBX,
- gen_op_subc_generic, NULL, gen_op_subccc_generic);
- }
+ update_psr(dc);
+ return do_arith(dc, a, CC_OP_FLAGS, gen_op_subc, NULL, gen_op_subccc);
}
static bool trans_MULScc(DisasContext *dc, arg_r_r_ri_cc *a)
@@ -3777,11 +3666,9 @@ static bool gen_edge(DisasContext *dc, arg_r_r_r *a,
s2 = gen_load_gpr(dc, a->rs2);
if (cc) {
- tcg_gen_mov_tl(cpu_cc_src, s1);
- tcg_gen_mov_tl(cpu_cc_src2, s2);
- tcg_gen_sub_tl(cpu_cc_dst, s1, s2);
- tcg_gen_movi_i32(cpu_cc_op, CC_OP_SUB);
- dc->cc_op = CC_OP_SUB;
+ gen_op_subcc(cpu_cc_N, s1, s2);
+ tcg_gen_movi_i32(cpu_cc_op, CC_OP_FLAGS);
+ dc->cc_op = CC_OP_FLAGS;
}
/*
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 07/21] target/sparc: Remove CC_OP_TADDTV, CC_OP_TSUBTV
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (5 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 06/21] target/sparc: Remove CC_OP_SUB, CC_OP_SUBX, CC_OP_TSUB Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 08/21] target/sparc: Remove CC_OP leftovers Richard Henderson
` (14 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/cpu.h | 2 -
target/sparc/cc_helper.c | 190 +--------------------------------------
target/sparc/helper.c | 36 ++++++--
target/sparc/translate.c | 4 +-
4 files changed, 32 insertions(+), 200 deletions(-)
diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 9884bd416a..a7999eaab5 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -150,8 +150,6 @@ enum {
enum {
CC_OP_DYNAMIC, /* must use dynamic code to get cc_op */
CC_OP_FLAGS, /* all cc are back in cc_*_[NZCV] registers */
- CC_OP_TADDTV, /* modify all flags except V, CC_DST = res, CC_SRC = src1 */
- CC_OP_TSUBTV, /* modify all flags except V, CC_DST = res, CC_SRC = src1 */
CC_OP_NB,
};
diff --git a/target/sparc/cc_helper.c b/target/sparc/cc_helper.c
index 20d451aa65..05f1479aea 100644
--- a/target/sparc/cc_helper.c
+++ b/target/sparc/cc_helper.c
@@ -21,198 +21,12 @@
#include "cpu.h"
#include "exec/helper-proto.h"
-static inline uint32_t get_NZ_icc(int32_t dst)
-{
- uint32_t ret = 0;
-
- if (dst == 0) {
- ret = PSR_ZERO;
- } else if (dst < 0) {
- ret = PSR_NEG;
- }
- return ret;
-}
-
-#ifdef TARGET_SPARC64
-static inline uint32_t get_NZ_xcc(target_long dst)
-{
- uint32_t ret = 0;
-
- if (!dst) {
- ret = PSR_ZERO;
- } else if (dst < 0) {
- ret = PSR_NEG;
- }
- return ret;
-}
-#endif
-
-static inline uint32_t get_C_add_icc(uint32_t dst, uint32_t src1)
-{
- uint32_t ret = 0;
-
- if (dst < src1) {
- ret = PSR_CARRY;
- }
- return ret;
-}
-
-#ifdef TARGET_SPARC64
-static inline uint32_t get_C_add_xcc(target_ulong dst, target_ulong src1)
-{
- uint32_t ret = 0;
-
- if (dst < src1) {
- ret = PSR_CARRY;
- }
- return ret;
-}
-
-static inline uint32_t get_V_add_xcc(target_ulong dst, target_ulong src1,
- target_ulong src2)
-{
- uint32_t ret = 0;
-
- if (((src1 ^ src2 ^ -1) & (src1 ^ dst)) & (1ULL << 63)) {
- ret = PSR_OVF;
- }
- return ret;
-}
-
-static uint32_t compute_all_add_xcc(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_xcc(CC_DST);
- ret |= get_C_add_xcc(CC_DST, CC_SRC);
- ret |= get_V_add_xcc(CC_DST, CC_SRC, CC_SRC2);
- return ret;
-}
-
-static uint32_t compute_C_add_xcc(CPUSPARCState *env)
-{
- return get_C_add_xcc(CC_DST, CC_SRC);
-}
-#endif
-
-static uint32_t compute_C_add(CPUSPARCState *env)
-{
- return get_C_add_icc(CC_DST, CC_SRC);
-}
-
-static uint32_t compute_all_taddtv(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_icc(CC_DST);
- ret |= get_C_add_icc(CC_DST, CC_SRC);
- return ret;
-}
-
-static inline uint32_t get_C_sub_icc(uint32_t src1, uint32_t src2)
-{
- uint32_t ret = 0;
-
- if (src1 < src2) {
- ret = PSR_CARRY;
- }
- return ret;
-}
-
-#ifdef TARGET_SPARC64
-static inline uint32_t get_C_sub_xcc(target_ulong src1, target_ulong src2)
-{
- uint32_t ret = 0;
-
- if (src1 < src2) {
- ret = PSR_CARRY;
- }
- return ret;
-}
-
-static inline uint32_t get_V_sub_xcc(target_ulong dst, target_ulong src1,
- target_ulong src2)
-{
- uint32_t ret = 0;
-
- if (((src1 ^ src2) & (src1 ^ dst)) & (1ULL << 63)) {
- ret = PSR_OVF;
- }
- return ret;
-}
-
-static uint32_t compute_all_sub_xcc(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_xcc(CC_DST);
- ret |= get_C_sub_xcc(CC_SRC, CC_SRC2);
- ret |= get_V_sub_xcc(CC_DST, CC_SRC, CC_SRC2);
- return ret;
-}
-
-static uint32_t compute_C_sub_xcc(CPUSPARCState *env)
-{
- return get_C_sub_xcc(CC_SRC, CC_SRC2);
-}
-#endif
-
-static uint32_t compute_C_sub(CPUSPARCState *env)
-{
- return get_C_sub_icc(CC_SRC, CC_SRC2);
-}
-
-static uint32_t compute_all_tsubtv(CPUSPARCState *env)
-{
- uint32_t ret;
-
- ret = get_NZ_icc(CC_DST);
- ret |= get_C_sub_icc(CC_SRC, CC_SRC2);
- return ret;
-}
-
-typedef struct CCTable {
- uint32_t (*compute_all)(CPUSPARCState *env); /* return all the flags */
- uint32_t (*compute_c)(CPUSPARCState *env); /* return the C flag */
-} CCTable;
-
-static const CCTable icc_table[CC_OP_NB] = {
- /* CC_OP_DYNAMIC should never happen */
- [CC_OP_TADDTV] = { compute_all_taddtv, compute_C_add },
- [CC_OP_TSUBTV] = { compute_all_tsubtv, compute_C_sub },
-};
-
-#ifdef TARGET_SPARC64
-static const CCTable xcc_table[CC_OP_NB] = {
- /* CC_OP_DYNAMIC should never happen */
- [CC_OP_TADDTV] = { compute_all_add_xcc, compute_C_add_xcc },
- [CC_OP_TSUBTV] = { compute_all_sub_xcc, compute_C_sub_xcc },
-};
-#endif
-
void helper_compute_psr(CPUSPARCState *env)
{
if (CC_OP == CC_OP_FLAGS) {
return;
}
-
- uint32_t icc = icc_table[CC_OP].compute_all(env);
-#ifdef TARGET_SPARC64
- uint32_t xcc = xcc_table[CC_OP].compute_all(env);
-
- env->cc_N = deposit64(-(icc & PSR_NEG), 32, 32, -(xcc & PSR_NEG));
- env->cc_V = deposit64(-(icc & PSR_OVF), 32, 32, -(xcc & PSR_OVF));
- env->icc_C = (uint64_t)icc << (32 - PSR_CARRY_SHIFT);
- env->xcc_C = (xcc >> PSR_CARRY_SHIFT) & 1;
- env->xcc_Z = ~xcc & PSR_ZERO;
-#else
- env->cc_N = -(icc & PSR_NEG);
- env->cc_V = -(icc & PSR_OVF);
- env->icc_C = (icc >> PSR_CARRY_SHIFT) & 1;
-#endif
- env->icc_Z = ~icc & PSR_ZERO;
-
- CC_OP = CC_OP_FLAGS;
+ g_assert_not_reached();
}
uint32_t helper_compute_C_icc(CPUSPARCState *env)
@@ -224,5 +38,5 @@ uint32_t helper_compute_C_icc(CPUSPARCState *env)
return env->icc_C;
#endif
}
- return icc_table[CC_OP].compute_c(env) >> PSR_CARRY_SHIFT;
+ g_assert_not_reached();
}
diff --git a/target/sparc/helper.c b/target/sparc/helper.c
index 53eec693dd..6117e99b55 100644
--- a/target/sparc/helper.c
+++ b/target/sparc/helper.c
@@ -156,7 +156,7 @@ uint64_t helper_udivx(CPUSPARCState *env, uint64_t a, uint64_t b)
target_ulong helper_taddcctv(CPUSPARCState *env, target_ulong src1,
target_ulong src2)
{
- target_ulong dst;
+ target_ulong dst, v;
/* Tag overflow occurs if either input has bits 0 or 1 set. */
if ((src1 | src2) & 3) {
@@ -166,13 +166,23 @@ target_ulong helper_taddcctv(CPUSPARCState *env, target_ulong src1,
dst = src1 + src2;
/* Tag overflow occurs if the addition overflows. */
- if (~(src1 ^ src2) & (src1 ^ dst) & (1u << 31)) {
+ v = ~(src1 ^ src2) & (src1 ^ dst);
+ if (v & (1u << 31)) {
goto tag_overflow;
}
/* Only modify the CC after any exceptions have been generated. */
- env->cc_src = src1;
- env->cc_src2 = src2;
+ env->cc_V = v;
+ env->cc_N = dst;
+ env->icc_Z = dst;
+#ifdef TARGET_SPARC64
+ env->xcc_Z = dst;
+ env->icc_C = dst ^ src1 ^ src2;
+ env->xcc_C = dst < src1;
+#else
+ env->icc_C = dst < src1;
+#endif
+
return dst;
tag_overflow:
@@ -182,7 +192,7 @@ target_ulong helper_taddcctv(CPUSPARCState *env, target_ulong src1,
target_ulong helper_tsubcctv(CPUSPARCState *env, target_ulong src1,
target_ulong src2)
{
- target_ulong dst;
+ target_ulong dst, v;
/* Tag overflow occurs if either input has bits 0 or 1 set. */
if ((src1 | src2) & 3) {
@@ -192,13 +202,23 @@ target_ulong helper_tsubcctv(CPUSPARCState *env, target_ulong src1,
dst = src1 - src2;
/* Tag overflow occurs if the subtraction overflows. */
- if ((src1 ^ src2) & (src1 ^ dst) & (1u << 31)) {
+ v = (src1 ^ src2) & (src1 ^ dst);
+ if (v & (1u << 31)) {
goto tag_overflow;
}
/* Only modify the CC after any exceptions have been generated. */
- env->cc_src = src1;
- env->cc_src2 = src2;
+ env->cc_V = v;
+ env->cc_N = dst;
+ env->icc_Z = dst;
+#ifdef TARGET_SPARC64
+ env->xcc_Z = dst;
+ env->icc_C = dst ^ src1 ^ src2;
+ env->xcc_C = src1 < src2;
+#else
+ env->icc_C = src1 < src2;
+#endif
+
return dst;
tag_overflow:
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index d119ce4c94..7703166ebd 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -3598,8 +3598,8 @@ TRANS(SUB, ALL, do_arith, a, CC_OP_FLAGS,
TRANS(TADDcc, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_taddcc)
TRANS(TSUBcc, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_tsubcc)
-TRANS(TADDccTV, ALL, do_arith, a, CC_OP_TADDTV, NULL, NULL, gen_op_taddcctv)
-TRANS(TSUBccTV, ALL, do_arith, a, CC_OP_TSUBTV, NULL, NULL, gen_op_tsubcctv)
+TRANS(TADDccTV, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_taddcctv)
+TRANS(TSUBccTV, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_tsubcctv)
TRANS(AND, ALL, do_logic, a, tcg_gen_and_tl, tcg_gen_andi_tl)
TRANS(XOR, ALL, do_logic, a, tcg_gen_xor_tl, tcg_gen_xori_tl)
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 08/21] target/sparc: Remove CC_OP leftovers
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (6 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 07/21] target/sparc: Remove CC_OP_TADDTV, CC_OP_TSUBTV Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 09/21] target/sparc: Remove DisasCompare.is_bool Richard Henderson
` (13 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
All instructions have been converted to generate
full condition codes explicitly.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/cpu.h | 21 -------
target/sparc/helper.h | 2 -
linux-user/sparc/cpu_loop.c | 5 --
target/sparc/cc_helper.c | 42 -------------
target/sparc/cpu.c | 1 -
target/sparc/int32_helper.c | 5 --
target/sparc/int64_helper.c | 5 --
target/sparc/translate.c | 115 ++++++++----------------------------
target/sparc/win_helper.c | 7 ---
target/sparc/meson.build | 1 -
10 files changed, 26 insertions(+), 178 deletions(-)
delete mode 100644 target/sparc/cc_helper.c
diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index a7999eaab5..3e361a5b75 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -137,22 +137,6 @@ enum {
#define PSR_CWP 0x1f
#endif
-#define CC_SRC (env->cc_src)
-#define CC_SRC2 (env->cc_src2)
-#define CC_DST (env->cc_dst)
-#define CC_OP (env->cc_op)
-
-/* Even though lazy evaluation of CPU condition codes tends to be less
- * important on RISC systems where condition codes are only updated
- * when explicitly requested, SPARC uses it to update 32-bit and 64-bit
- * condition codes.
- */
-enum {
- CC_OP_DYNAMIC, /* must use dynamic code to get cc_op */
- CC_OP_FLAGS, /* all cc are back in cc_*_[NZCV] registers */
- CC_OP_NB,
-};
-
/* Trap base register */
#define TBR_BASE_MASK 0xfffff000
@@ -474,11 +458,6 @@ struct CPUArchState {
target_ulong xcc_C;
#endif
- /* emulator internal flags handling */
- target_ulong cc_src, cc_src2;
- target_ulong cc_dst;
- uint32_t cc_op;
-
target_ulong cond; /* conditional branch result (XXX: save it in a
temporary register when possible) */
diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index a7b0079c3b..decd94c0d6 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -148,5 +148,3 @@ VIS_CMPHELPER(cmpne)
#undef F_HELPER_0_1
#undef VIS_HELPER
#undef VIS_CMPHELPER
-DEF_HELPER_1(compute_psr, void, env)
-DEF_HELPER_FLAGS_1(compute_C_icc, TCG_CALL_NO_WG_SE, i32, env)
diff --git a/linux-user/sparc/cpu_loop.c b/linux-user/sparc/cpu_loop.c
index c1a2362041..3c1bde00dd 100644
--- a/linux-user/sparc/cpu_loop.c
+++ b/linux-user/sparc/cpu_loop.c
@@ -222,11 +222,6 @@ void cpu_loop (CPUSPARCState *env)
cpu_exec_end(cs);
process_queued_cpu_work(cs);
- /* Compute PSR before exposing state. */
- if (env->cc_op != CC_OP_FLAGS) {
- cpu_get_psr(env);
- }
-
switch (trapnr) {
case TARGET_TT_SYSCALL:
ret = do_syscall (env, env->gregs[1],
diff --git a/target/sparc/cc_helper.c b/target/sparc/cc_helper.c
deleted file mode 100644
index 05f1479aea..0000000000
--- a/target/sparc/cc_helper.c
+++ /dev/null
@@ -1,42 +0,0 @@
-/*
- * Helpers for lazy condition code handling
- *
- * Copyright (c) 2003-2005 Fabrice Bellard
- *
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, see <http://www.gnu.org/licenses/>.
- */
-
-#include "qemu/osdep.h"
-#include "cpu.h"
-#include "exec/helper-proto.h"
-
-void helper_compute_psr(CPUSPARCState *env)
-{
- if (CC_OP == CC_OP_FLAGS) {
- return;
- }
- g_assert_not_reached();
-}
-
-uint32_t helper_compute_C_icc(CPUSPARCState *env)
-{
- if (CC_OP == CC_OP_FLAGS) {
-#ifdef TARGET_SPARC64
- return extract64(env->icc_C, 32, 1);
-#else
- return env->icc_C;
-#endif
- }
- g_assert_not_reached();
-}
diff --git a/target/sparc/cpu.c b/target/sparc/cpu.c
index bb1a155510..befa7fc4eb 100644
--- a/target/sparc/cpu.c
+++ b/target/sparc/cpu.c
@@ -46,7 +46,6 @@ static void sparc_cpu_reset_hold(Object *obj)
env->wim = 1;
#endif
env->regwptr = env->regbase + (env->cwp * 16);
- CC_OP = CC_OP_FLAGS;
#if defined(CONFIG_USER_ONLY)
#ifdef TARGET_SPARC64
env->cleanwin = env->nwindows - 2;
diff --git a/target/sparc/int32_helper.c b/target/sparc/int32_helper.c
index 82e8418e46..1563613582 100644
--- a/target/sparc/int32_helper.c
+++ b/target/sparc/int32_helper.c
@@ -103,11 +103,6 @@ void sparc_cpu_do_interrupt(CPUState *cs)
CPUSPARCState *env = &cpu->env;
int cwp, intno = cs->exception_index;
- /* Compute PSR before exposing state. */
- if (env->cc_op != CC_OP_FLAGS) {
- cpu_get_psr(env);
- }
-
if (qemu_loglevel_mask(CPU_LOG_INT)) {
static int count;
const char *name;
diff --git a/target/sparc/int64_helper.c b/target/sparc/int64_helper.c
index 793e57c536..1b4155f5f3 100644
--- a/target/sparc/int64_helper.c
+++ b/target/sparc/int64_helper.c
@@ -135,11 +135,6 @@ void sparc_cpu_do_interrupt(CPUState *cs)
int intno = cs->exception_index;
trap_state *tsptr;
- /* Compute PSR before exposing state. */
- if (env->cc_op != CC_OP_FLAGS) {
- cpu_get_psr(env);
- }
-
#ifdef DEBUG_PCALL
if (qemu_loglevel_mask(CPU_LOG_INT)) {
static int count;
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 7703166ebd..7c4fcf8326 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -105,8 +105,6 @@
/* global register indexes */
static TCGv_ptr cpu_regwptr;
-static TCGv cpu_cc_src, cpu_cc_src2, cpu_cc_dst;
-static TCGv_i32 cpu_cc_op;
static TCGv cpu_fsr, cpu_pc, cpu_npc;
static TCGv cpu_regs[32];
static TCGv cpu_y;
@@ -172,7 +170,6 @@ typedef struct DisasContext {
#endif
#endif
- uint32_t cc_op; /* current CC operation */
sparc_def_t *def;
#ifdef TARGET_SPARC64
int fprs_dirty;
@@ -962,14 +959,6 @@ static void save_npc(DisasContext *dc)
}
}
-static void update_psr(DisasContext *dc)
-{
- if (dc->cc_op != CC_OP_FLAGS) {
- dc->cc_op = CC_OP_FLAGS;
- gen_helper_compute_psr(tcg_env);
- }
-}
-
static void save_state(DisasContext *dc)
{
tcg_gen_movi_tl(cpu_pc, dc->pc);
@@ -1048,20 +1037,9 @@ static void gen_op_next_insn(void)
static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
DisasContext *dc)
{
- TCGv t1, t2;
+ TCGv t1;
cmp->is_bool = false;
-
- switch (dc->cc_op) {
- default:
- gen_helper_compute_psr(tcg_env);
- dc->cc_op = CC_OP_FLAGS;
- break;
-
- case CC_OP_FLAGS:
- break;
- }
-
cmp->c1 = t1 = tcg_temp_new();
cmp->c2 = tcg_constant_tl(0);
@@ -2739,7 +2717,6 @@ TRANS(RDASR17, ASR17, do_rd_special, true, a->rd, do_rd_leon3_config)
static TCGv do_rdccr(DisasContext *dc, TCGv dst)
{
- update_psr(dc);
gen_helper_rdccr(dst, tcg_env);
return dst;
}
@@ -2852,7 +2829,6 @@ TRANS(RDSTRAND_STATUS, HYPV, do_rd_special, true, a->rd, do_rdstrand_status)
static TCGv do_rdpsr(DisasContext *dc, TCGv dst)
{
- update_psr(dc);
gen_helper_rdpsr(dst, tcg_env);
return dst;
}
@@ -3257,8 +3233,6 @@ TRANS(WRPOWERDOWN, POWERDOWN, do_wr_special, a, supervisor(dc), do_wrpowerdown)
static void do_wrpsr(DisasContext *dc, TCGv src)
{
gen_helper_wrpsr(tcg_env, src);
- tcg_gen_movi_i32(cpu_cc_op, CC_OP_FLAGS);
- dc->cc_op = CC_OP_FLAGS;
dc->base.is_jmp = DISAS_EXIT;
}
@@ -3522,7 +3496,7 @@ static bool trans_NOP(DisasContext *dc, arg_NOP *a)
TRANS(NOP_v7, 32, trans_NOP, a)
TRANS(NOP_v9, 64, trans_NOP, a)
-static bool do_arith_int(DisasContext *dc, arg_r_r_ri_cc *a, int cc_op,
+static bool do_arith_int(DisasContext *dc, arg_r_r_ri_cc *a,
void (*func)(TCGv, TCGv, TCGv),
void (*funci)(TCGv, TCGv, target_long),
bool logic_cc)
@@ -3536,8 +3510,6 @@ static bool do_arith_int(DisasContext *dc, arg_r_r_ri_cc *a, int cc_op,
if (logic_cc) {
dst = cpu_cc_N;
- } else if (a->cc && cc_op > CC_OP_FLAGS) {
- dst = cpu_cc_dst;
} else {
dst = gen_dest_gpr(dc, a->rd);
}
@@ -3564,42 +3536,36 @@ static bool do_arith_int(DisasContext *dc, arg_r_r_ri_cc *a, int cc_op,
}
gen_store_gpr(dc, a->rd, dst);
-
- if (a->cc) {
- tcg_gen_movi_i32(cpu_cc_op, cc_op);
- dc->cc_op = cc_op;
- }
return advance_pc(dc);
}
-static bool do_arith(DisasContext *dc, arg_r_r_ri_cc *a, int cc_op,
+static bool do_arith(DisasContext *dc, arg_r_r_ri_cc *a,
void (*func)(TCGv, TCGv, TCGv),
void (*funci)(TCGv, TCGv, target_long),
void (*func_cc)(TCGv, TCGv, TCGv))
{
if (a->cc) {
- assert(cc_op >= 0);
- return do_arith_int(dc, a, cc_op, func_cc, NULL, false);
+ return do_arith_int(dc, a, func_cc, NULL, false);
}
- return do_arith_int(dc, a, cc_op, func, funci, false);
+ return do_arith_int(dc, a, func, funci, false);
}
static bool do_logic(DisasContext *dc, arg_r_r_ri_cc *a,
void (*func)(TCGv, TCGv, TCGv),
void (*funci)(TCGv, TCGv, target_long))
{
- return do_arith_int(dc, a, CC_OP_FLAGS, func, funci, a->cc);
+ return do_arith_int(dc, a, func, funci, a->cc);
}
-TRANS(ADD, ALL, do_arith, a, CC_OP_FLAGS,
- tcg_gen_add_tl, tcg_gen_addi_tl, gen_op_addcc)
-TRANS(SUB, ALL, do_arith, a, CC_OP_FLAGS,
- tcg_gen_sub_tl, tcg_gen_subi_tl, gen_op_subcc)
+TRANS(ADD, ALL, do_arith, a, tcg_gen_add_tl, tcg_gen_addi_tl, gen_op_addcc)
+TRANS(SUB, ALL, do_arith, a, tcg_gen_sub_tl, tcg_gen_subi_tl, gen_op_subcc)
+TRANS(ADDC, ALL, do_arith, a, gen_op_addc, NULL, gen_op_addccc)
+TRANS(SUBC, ALL, do_arith, a, gen_op_subc, NULL, gen_op_subccc)
-TRANS(TADDcc, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_taddcc)
-TRANS(TSUBcc, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_tsubcc)
-TRANS(TADDccTV, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_taddcctv)
-TRANS(TSUBccTV, ALL, do_arith, a, CC_OP_FLAGS, NULL, NULL, gen_op_tsubcctv)
+TRANS(TADDcc, ALL, do_arith, a, NULL, NULL, gen_op_taddcc)
+TRANS(TSUBcc, ALL, do_arith, a, NULL, NULL, gen_op_tsubcc)
+TRANS(TADDccTV, ALL, do_arith, a, NULL, NULL, gen_op_taddcctv)
+TRANS(TSUBccTV, ALL, do_arith, a, NULL, NULL, gen_op_tsubcctv)
TRANS(AND, ALL, do_logic, a, tcg_gen_and_tl, tcg_gen_andi_tl)
TRANS(XOR, ALL, do_logic, a, tcg_gen_xor_tl, tcg_gen_xori_tl)
@@ -3607,17 +3573,18 @@ TRANS(ANDN, ALL, do_logic, a, tcg_gen_andc_tl, NULL)
TRANS(ORN, ALL, do_logic, a, tcg_gen_orc_tl, NULL)
TRANS(XORN, ALL, do_logic, a, tcg_gen_eqv_tl, NULL)
-TRANS(MULX, 64, do_arith, a, -1, tcg_gen_mul_tl, tcg_gen_muli_tl, NULL)
+TRANS(MULX, 64, do_arith, a, tcg_gen_mul_tl, tcg_gen_muli_tl, NULL)
TRANS(UMUL, MUL, do_logic, a, gen_op_umul, NULL)
TRANS(SMUL, MUL, do_logic, a, gen_op_smul, NULL)
+TRANS(MULScc, ALL, do_arith, a, NULL, NULL, gen_op_mulscc)
-TRANS(UDIVX, 64, do_arith, a, -1, gen_op_udivx, NULL, NULL)
-TRANS(SDIVX, 64, do_arith, a, -1, gen_op_sdivx, NULL, NULL)
-TRANS(UDIV, DIV, do_arith, a, CC_OP_FLAGS, gen_op_udiv, NULL, gen_op_udivcc)
-TRANS(SDIV, DIV, do_arith, a, CC_OP_FLAGS, gen_op_sdiv, NULL, gen_op_sdivcc)
+TRANS(UDIVX, 64, do_arith, a, gen_op_udivx, NULL, NULL)
+TRANS(SDIVX, 64, do_arith, a, gen_op_sdivx, NULL, NULL)
+TRANS(UDIV, DIV, do_arith, a, gen_op_udiv, NULL, gen_op_udivcc)
+TRANS(SDIV, DIV, do_arith, a, gen_op_sdiv, NULL, gen_op_sdivcc)
/* TODO: Should have feature bit -- comes in with UltraSparc T2. */
-TRANS(POPC, 64, do_arith, a, -1, gen_op_popc, NULL, NULL)
+TRANS(POPC, 64, do_arith, a, gen_op_popc, NULL, NULL)
static bool trans_OR(DisasContext *dc, arg_r_r_ri_cc *a)
{
@@ -3636,24 +3603,6 @@ static bool trans_OR(DisasContext *dc, arg_r_r_ri_cc *a)
return do_logic(dc, a, tcg_gen_or_tl, tcg_gen_ori_tl);
}
-static bool trans_ADDC(DisasContext *dc, arg_r_r_ri_cc *a)
-{
- update_psr(dc);
- return do_arith(dc, a, CC_OP_FLAGS, gen_op_addc, NULL, gen_op_addccc);
-}
-
-static bool trans_SUBC(DisasContext *dc, arg_r_r_ri_cc *a)
-{
- update_psr(dc);
- return do_arith(dc, a, CC_OP_FLAGS, gen_op_subc, NULL, gen_op_subccc);
-}
-
-static bool trans_MULScc(DisasContext *dc, arg_r_r_ri_cc *a)
-{
- update_psr(dc);
- return do_arith(dc, a, CC_OP_FLAGS, NULL, NULL, gen_op_mulscc);
-}
-
static bool gen_edge(DisasContext *dc, arg_r_r_r *a,
int width, bool cc, bool left)
{
@@ -3667,8 +3616,6 @@ static bool gen_edge(DisasContext *dc, arg_r_r_r *a,
if (cc) {
gen_op_subcc(cpu_cc_N, s1, s2);
- tcg_gen_movi_i32(cpu_cc_op, CC_OP_FLAGS);
- dc->cc_op = CC_OP_FLAGS;
}
/*
@@ -5080,7 +5027,6 @@ static void sparc_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs)
dc->pc = dc->base.pc_first;
dc->npc = (target_ulong)dc->base.tb->cs_base;
- dc->cc_op = CC_OP_DYNAMIC;
dc->mem_idx = dc->base.tb->flags & TB_FLAG_MMU_MASK;
dc->def = &env->def;
dc->fpu_enabled = tb_fpu_enabled(dc->base.tb->flags);
@@ -5269,13 +5215,6 @@ void sparc_tcg_init(void)
"f48", "f50", "f52", "f54", "f56", "f58", "f60", "f62",
};
- static const struct { TCGv_i32 *ptr; int off; const char *name; } r32[] = {
-#ifdef TARGET_SPARC64
- { &cpu_fprs, offsetof(CPUSPARCState, fprs), "fprs" },
-#endif
- { &cpu_cc_op, offsetof(CPUSPARCState, cc_op), "cc_op" },
- };
-
static const struct { TCGv *ptr; int off; const char *name; } rtl[] = {
#ifdef TARGET_SPARC64
{ &cpu_gsr, offsetof(CPUSPARCState, gsr), "gsr" },
@@ -5287,9 +5226,6 @@ void sparc_tcg_init(void)
{ &cpu_icc_Z, offsetof(CPUSPARCState, icc_Z), "icc_Z" },
{ &cpu_icc_C, offsetof(CPUSPARCState, icc_C), "icc_C" },
{ &cpu_cond, offsetof(CPUSPARCState, cond), "cond" },
- { &cpu_cc_src, offsetof(CPUSPARCState, cc_src), "cc_src" },
- { &cpu_cc_src2, offsetof(CPUSPARCState, cc_src2), "cc_src2" },
- { &cpu_cc_dst, offsetof(CPUSPARCState, cc_dst), "cc_dst" },
{ &cpu_fsr, offsetof(CPUSPARCState, fsr), "fsr" },
{ &cpu_pc, offsetof(CPUSPARCState, pc), "pc" },
{ &cpu_npc, offsetof(CPUSPARCState, npc), "npc" },
@@ -5303,10 +5239,6 @@ void sparc_tcg_init(void)
offsetof(CPUSPARCState, regwptr),
"regwptr");
- for (i = 0; i < ARRAY_SIZE(r32); ++i) {
- *r32[i].ptr = tcg_global_mem_new_i32(tcg_env, r32[i].off, r32[i].name);
- }
-
for (i = 0; i < ARRAY_SIZE(rtl); ++i) {
*rtl[i].ptr = tcg_global_mem_new(tcg_env, rtl[i].off, rtl[i].name);
}
@@ -5329,6 +5261,11 @@ void sparc_tcg_init(void)
offsetof(CPUSPARCState, fpr[i]),
fregnames[i]);
}
+
+#ifdef TARGET_SPARC64
+ cpu_fprs = tcg_global_mem_new_i32(tcg_env,
+ offsetof(CPUSPARCState, fprs), "fprs");
+#endif
}
void sparc_restore_state_to_opc(CPUState *cs,
diff --git a/target/sparc/win_helper.c b/target/sparc/win_helper.c
index f0ff6bf5db..16d1c70fe7 100644
--- a/target/sparc/win_helper.c
+++ b/target/sparc/win_helper.c
@@ -55,8 +55,6 @@ target_ulong cpu_get_psr(CPUSPARCState *env)
{
target_ulong icc = 0;
- helper_compute_psr(env);
-
icc |= ((int32_t)env->cc_N < 0) << PSR_NEG_SHIFT;
icc |= ((int32_t)env->cc_V < 0) << PSR_OVF_SHIFT;
icc |= ((int32_t)env->icc_Z == 0) << PSR_ZERO_SHIFT;
@@ -103,7 +101,6 @@ void cpu_put_psr_raw(CPUSPARCState *env, target_ulong val)
env->psrps = (val & PSR_PS) ? 1 : 0;
env->psret = (val & PSR_ET) ? 1 : 0;
#endif
- env->cc_op = CC_OP_FLAGS;
#if !defined(TARGET_SPARC64)
cpu_set_cwp(env, val & PSR_CWP);
#endif
@@ -272,8 +269,6 @@ target_ulong cpu_get_ccr(CPUSPARCState *env)
{
target_ulong ccr = 0;
- helper_compute_psr(env);
-
ccr |= (env->icc_C >> 32) & 1;
ccr |= ((int32_t)env->cc_V < 0) << 1;
ccr |= ((int32_t)env->icc_Z == 0) << 2;
@@ -295,8 +290,6 @@ void cpu_put_ccr(CPUSPARCState *env, target_ulong val)
env->xcc_C = (val >> 4) & 1;
env->icc_Z = ~val & 0x04;
env->xcc_Z = ~val & 0x40;
-
- CC_OP = CC_OP_FLAGS;
}
target_ulong cpu_get_cwp64(CPUSPARCState *env)
diff --git a/target/sparc/meson.build b/target/sparc/meson.build
index c316773db6..46289c8669 100644
--- a/target/sparc/meson.build
+++ b/target/sparc/meson.build
@@ -3,7 +3,6 @@ gen = decodetree.process('insns.decode')
sparc_ss = ss.source_set()
sparc_ss.add(gen)
sparc_ss.add(files(
- 'cc_helper.c',
'cpu.c',
'fop_helper.c',
'gdbstub.c',
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 09/21] target/sparc: Remove DisasCompare.is_bool
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (7 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 08/21] target/sparc: Remove CC_OP leftovers Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 10/21] target/sparc: Change DisasCompare.c2 to int Richard Henderson
` (12 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Since we're going to feed cpu_cond to another comparison, we don't
reqire a boolean value -- anything non-zero is sufficient.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 22 +++++++---------------
1 file changed, 7 insertions(+), 15 deletions(-)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 7c4fcf8326..464f1607e3 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -180,7 +180,6 @@ typedef struct DisasContext {
typedef struct {
TCGCond cond;
- bool is_bool;
TCGv c1, c2;
} DisasCompare;
@@ -1039,7 +1038,6 @@ static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
{
TCGv t1;
- cmp->is_bool = false;
cmp->c1 = t1 = tcg_temp_new();
cmp->c2 = tcg_constant_tl(0);
@@ -1104,7 +1102,6 @@ static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
case 0x5: /* ltu: C */
cmp->cond = TCG_COND_NE;
- cmp->is_bool = true;
if (TARGET_LONG_BITS == 32 || xcc) {
tcg_gen_mov_tl(t1, cpu_cc_C);
} else {
@@ -1132,7 +1129,6 @@ static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
}
if (cond & 8) {
cmp->cond = tcg_invert_cond(cmp->cond);
- cmp->is_bool = false;
}
}
@@ -1143,7 +1139,6 @@ static void gen_fcompare(DisasCompare *cmp, unsigned int cc, unsigned int cond)
/* For now we still generate a straight boolean result. */
cmp->cond = TCG_COND_NE;
- cmp->is_bool = true;
cmp->c1 = r_dst = tcg_temp_new();
cmp->c2 = tcg_constant_tl(0);
@@ -1230,7 +1225,6 @@ static const TCGCond gen_tcg_cond_reg[8] = {
static void gen_compare_reg(DisasCompare *cmp, int cond, TCGv r_src)
{
cmp->cond = tcg_invert_cond(gen_tcg_cond_reg[cond]);
- cmp->is_bool = false;
cmp->c1 = r_src;
cmp->c2 = tcg_constant_tl(0);
}
@@ -2232,18 +2226,14 @@ static void gen_fmovs(DisasContext *dc, DisasCompare *cmp, int rd, int rs)
{
#ifdef TARGET_SPARC64
TCGv_i32 c32, zero, dst, s1, s2;
+ TCGv_i64 c64 = tcg_temp_new_i64();
/* We have two choices here: extend the 32 bit data and use movcond_i64,
or fold the comparison down to 32 bits and use movcond_i32. Choose
the later. */
c32 = tcg_temp_new_i32();
- if (cmp->is_bool) {
- tcg_gen_extrl_i64_i32(c32, cmp->c1);
- } else {
- TCGv_i64 c64 = tcg_temp_new_i64();
- tcg_gen_setcond_i64(cmp->cond, c64, cmp->c1, cmp->c2);
- tcg_gen_extrl_i64_i32(c32, c64);
- }
+ tcg_gen_setcond_i64(cmp->cond, c64, cmp->c1, cmp->c2);
+ tcg_gen_extrl_i64_i32(c32, c64);
s1 = gen_load_fpr_F(dc, rs);
s2 = gen_load_fpr_F(dc, rd);
@@ -2445,8 +2435,10 @@ static bool advance_jump_cond(DisasContext *dc, DisasCompare *cmp,
dc->jump_pc[0] = dest;
dc->jump_pc[1] = npc + 4;
dc->npc = JUMP_PC;
- if (cmp->is_bool) {
- tcg_gen_mov_tl(cpu_cond, cmp->c1);
+
+ /* The condition for cpu_cond is always NE -- normalize. */
+ if (cmp->cond == TCG_COND_NE) {
+ tcg_gen_xor_tl(cpu_cond, cmp->c1, cmp->c2);
} else {
tcg_gen_setcond_tl(cmp->cond, cpu_cond, cmp->c1, cmp->c2);
}
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 10/21] target/sparc: Change DisasCompare.c2 to int
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (8 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 09/21] target/sparc: Remove DisasCompare.is_bool Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 11/21] target/sparc: Always copy conditions into a new temporary Richard Henderson
` (11 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
We don't require c2 to be variable, so emphasize that.
We don't currently require c2 to be non-zero, but that will change.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 33 ++++++++++++++++++---------------
1 file changed, 18 insertions(+), 15 deletions(-)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 464f1607e3..a405512e6c 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -180,7 +180,8 @@ typedef struct DisasContext {
typedef struct {
TCGCond cond;
- TCGv c1, c2;
+ TCGv c1;
+ int c2;
} DisasCompare;
// This function uses non-native bit order
@@ -1039,12 +1040,12 @@ static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
TCGv t1;
cmp->c1 = t1 = tcg_temp_new();
- cmp->c2 = tcg_constant_tl(0);
+ cmp->c2 = 0;
switch (cond & 7) {
case 0x0: /* never */
cmp->cond = TCG_COND_NEVER;
- cmp->c1 = cmp->c2;
+ cmp->c1 = tcg_constant_tl(0);
break;
case 0x1: /* eq: Z */
@@ -1140,7 +1141,7 @@ static void gen_fcompare(DisasCompare *cmp, unsigned int cc, unsigned int cond)
/* For now we still generate a straight boolean result. */
cmp->cond = TCG_COND_NE;
cmp->c1 = r_dst = tcg_temp_new();
- cmp->c2 = tcg_constant_tl(0);
+ cmp->c2 = 0;
switch (cc) {
default:
@@ -1226,7 +1227,7 @@ static void gen_compare_reg(DisasCompare *cmp, int cond, TCGv r_src)
{
cmp->cond = tcg_invert_cond(gen_tcg_cond_reg[cond]);
cmp->c1 = r_src;
- cmp->c2 = tcg_constant_tl(0);
+ cmp->c2 = 0;
}
static void gen_op_clear_ieee_excp_and_FTT(void)
@@ -2232,7 +2233,7 @@ static void gen_fmovs(DisasContext *dc, DisasCompare *cmp, int rd, int rs)
or fold the comparison down to 32 bits and use movcond_i32. Choose
the later. */
c32 = tcg_temp_new_i32();
- tcg_gen_setcond_i64(cmp->cond, c64, cmp->c1, cmp->c2);
+ tcg_gen_setcondi_i64(cmp->cond, c64, cmp->c1, cmp->c2);
tcg_gen_extrl_i64_i32(c32, c64);
s1 = gen_load_fpr_F(dc, rs);
@@ -2252,7 +2253,7 @@ static void gen_fmovd(DisasContext *dc, DisasCompare *cmp, int rd, int rs)
{
#ifdef TARGET_SPARC64
TCGv_i64 dst = gen_dest_fpr_D(dc, rd);
- tcg_gen_movcond_i64(cmp->cond, dst, cmp->c1, cmp->c2,
+ tcg_gen_movcond_i64(cmp->cond, dst, cmp->c1, tcg_constant_tl(cmp->c2),
gen_load_fpr_D(dc, rs),
gen_load_fpr_D(dc, rd));
gen_store_fpr_D(dc, rd, dst);
@@ -2266,10 +2267,11 @@ static void gen_fmovq(DisasContext *dc, DisasCompare *cmp, int rd, int rs)
#ifdef TARGET_SPARC64
int qd = QFPREG(rd);
int qs = QFPREG(rs);
+ TCGv c2 = tcg_constant_tl(cmp->c2);
- tcg_gen_movcond_i64(cmp->cond, cpu_fpr[qd / 2], cmp->c1, cmp->c2,
+ tcg_gen_movcond_i64(cmp->cond, cpu_fpr[qd / 2], cmp->c1, c2,
cpu_fpr[qs / 2], cpu_fpr[qd / 2]);
- tcg_gen_movcond_i64(cmp->cond, cpu_fpr[qd / 2 + 1], cmp->c1, cmp->c2,
+ tcg_gen_movcond_i64(cmp->cond, cpu_fpr[qd / 2 + 1], cmp->c1, c2,
cpu_fpr[qs / 2 + 1], cpu_fpr[qd / 2 + 1]);
gen_update_fprs_dirty(dc, qd);
@@ -2409,7 +2411,7 @@ static bool advance_jump_cond(DisasContext *dc, DisasCompare *cmp,
if (annul) {
TCGLabel *l1 = gen_new_label();
- tcg_gen_brcond_tl(tcg_invert_cond(cmp->cond), cmp->c1, cmp->c2, l1);
+ tcg_gen_brcondi_tl(tcg_invert_cond(cmp->cond), cmp->c1, cmp->c2, l1);
gen_goto_tb(dc, 0, npc, dest);
gen_set_label(l1);
gen_goto_tb(dc, 1, npc + 4, npc + 8);
@@ -2423,7 +2425,7 @@ static bool advance_jump_cond(DisasContext *dc, DisasCompare *cmp,
tcg_gen_mov_tl(cpu_pc, cpu_npc);
tcg_gen_addi_tl(cpu_npc, cpu_npc, 4);
tcg_gen_movcond_tl(cmp->cond, cpu_npc,
- cmp->c1, cmp->c2,
+ cmp->c1, tcg_constant_tl(cmp->c2),
tcg_constant_tl(dest), cpu_npc);
dc->pc = npc;
break;
@@ -2438,9 +2440,9 @@ static bool advance_jump_cond(DisasContext *dc, DisasCompare *cmp,
/* The condition for cpu_cond is always NE -- normalize. */
if (cmp->cond == TCG_COND_NE) {
- tcg_gen_xor_tl(cpu_cond, cmp->c1, cmp->c2);
+ tcg_gen_xori_tl(cpu_cond, cmp->c1, cmp->c2);
} else {
- tcg_gen_setcond_tl(cmp->cond, cpu_cond, cmp->c1, cmp->c2);
+ tcg_gen_setcondi_tl(cmp->cond, cpu_cond, cmp->c1, cmp->c2);
}
}
}
@@ -2612,7 +2614,7 @@ static bool do_tcc(DisasContext *dc, int cond, int cc,
flush_cond(dc);
lab = delay_exceptionv(dc, trap);
gen_compare(&cmp, cc, cond, dc);
- tcg_gen_brcond_tl(cmp.cond, cmp.c1, cmp.c2, lab);
+ tcg_gen_brcondi_tl(cmp.cond, cmp.c1, cmp.c2, lab);
return advance_pc(dc);
}
@@ -3849,8 +3851,9 @@ static TCGv gen_rs2_or_imm(DisasContext *dc, bool imm, int rs2_or_imm)
static bool do_mov_cond(DisasContext *dc, DisasCompare *cmp, int rd, TCGv src2)
{
TCGv dst = gen_load_gpr(dc, rd);
+ TCGv c2 = tcg_constant_tl(cmp->c2);
- tcg_gen_movcond_tl(cmp->cond, dst, cmp->c1, cmp->c2, src2, dst);
+ tcg_gen_movcond_tl(cmp->cond, dst, cmp->c1, c2, src2, dst);
gen_store_gpr(dc, rd, dst);
return advance_pc(dc);
}
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 11/21] target/sparc: Always copy conditions into a new temporary
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (9 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 10/21] target/sparc: Change DisasCompare.c2 to int Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 12/21] target/sparc: Do flush_cond in advance_jump_cond Richard Henderson
` (10 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
This will allow the condition to live across changes to
the global cc variables.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index a405512e6c..dd6d43d1f1 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -1226,8 +1226,9 @@ static const TCGCond gen_tcg_cond_reg[8] = {
static void gen_compare_reg(DisasCompare *cmp, int cond, TCGv r_src)
{
cmp->cond = tcg_invert_cond(gen_tcg_cond_reg[cond]);
- cmp->c1 = r_src;
+ cmp->c1 = tcg_temp_new();
cmp->c2 = 0;
+ tcg_gen_mov_tl(cmp->c1, r_src);
}
static void gen_op_clear_ieee_excp_and_FTT(void)
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 12/21] target/sparc: Do flush_cond in advance_jump_cond
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (10 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 11/21] target/sparc: Always copy conditions into a new temporary Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 13/21] target/sparc: Merge gen_branch2 into advance_pc Richard Henderson
` (9 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Do this here instead of in each caller.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 10 ++++------
1 file changed, 4 insertions(+), 6 deletions(-)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index dd6d43d1f1..2e7deb5e33 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -2407,7 +2407,10 @@ static bool advance_jump_uncond_always(DisasContext *dc, bool annul,
static bool advance_jump_cond(DisasContext *dc, DisasCompare *cmp,
bool annul, target_ulong dest)
{
- target_ulong npc = dc->npc;
+ target_ulong npc;
+
+ flush_cond(dc);
+ npc = dc->npc;
if (annul) {
TCGLabel *l1 = gen_new_label();
@@ -2481,8 +2484,6 @@ static bool do_bpcc(DisasContext *dc, arg_bcc *a)
case 0x8:
return advance_jump_uncond_always(dc, a->a, target);
default:
- flush_cond(dc);
-
gen_compare(&cmp, a->cc, a->cond, dc);
return advance_jump_cond(dc, &cmp, a->a, target);
}
@@ -2505,8 +2506,6 @@ static bool do_fbpfcc(DisasContext *dc, arg_bcc *a)
case 0x8:
return advance_jump_uncond_always(dc, a->a, target);
default:
- flush_cond(dc);
-
gen_fcompare(&cmp, a->cc, a->cond);
return advance_jump_cond(dc, &cmp, a->a, target);
}
@@ -2527,7 +2526,6 @@ static bool trans_BPr(DisasContext *dc, arg_BPr *a)
return false;
}
- flush_cond(dc);
gen_compare_reg(&cmp, a->cond, gen_load_gpr(dc, a->rs1));
return advance_jump_cond(dc, &cmp, a->a, target);
}
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 13/21] target/sparc: Merge gen_branch2 into advance_pc
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (11 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 12/21] target/sparc: Do flush_cond in advance_jump_cond Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 14/21] target/sparc: Merge advance_jump_uncond_{never, always} into advance_jump_cond Richard Henderson
` (8 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
The function had only one caller. Canonicalize the cpu_cond
test to TCG_COND_NE, the "natural" sense of its value.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 28 ++++++++++++++--------------
1 file changed, 14 insertions(+), 14 deletions(-)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 2e7deb5e33..e134ba8821 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -908,19 +908,6 @@ static void gen_op_eval_fbo(TCGv dst, TCGv src, unsigned int fcc_offset)
tcg_gen_xori_tl(dst, dst, 0x1);
}
-static void gen_branch2(DisasContext *dc, target_ulong pc1,
- target_ulong pc2, TCGv r_cond)
-{
- TCGLabel *l1 = gen_new_label();
-
- tcg_gen_brcondi_tl(TCG_COND_EQ, r_cond, 0, l1);
-
- gen_goto_tb(dc, 0, pc1, pc1 + 4);
-
- gen_set_label(l1);
- gen_goto_tb(dc, 1, pc2, pc2 + 4);
-}
-
static void gen_generic_branch(DisasContext *dc)
{
TCGv npc0 = tcg_constant_tl(dc->jump_pc[0]);
@@ -2352,6 +2339,8 @@ static int extract_qfpreg(DisasContext *dc, int x)
/* Default case for non jump instructions. */
static bool advance_pc(DisasContext *dc)
{
+ TCGLabel *l1;
+
if (dc->npc & 3) {
switch (dc->npc) {
case DYNAMIC_PC:
@@ -2359,11 +2348,22 @@ static bool advance_pc(DisasContext *dc)
dc->pc = dc->npc;
gen_op_next_insn();
break;
+
case JUMP_PC:
/* we can do a static jump */
- gen_branch2(dc, dc->jump_pc[0], dc->jump_pc[1], cpu_cond);
+ l1 = gen_new_label();
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_cond, 0, l1);
+
+ /* jump not taken */
+ gen_goto_tb(dc, 1, dc->jump_pc[1], dc->jump_pc[1] + 4);
+
+ /* jump taken */
+ gen_set_label(l1);
+ gen_goto_tb(dc, 0, dc->jump_pc[0], dc->jump_pc[0] + 4);
+
dc->base.is_jmp = DISAS_NORETURN;
break;
+
default:
g_assert_not_reached();
}
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 14/21] target/sparc: Merge advance_jump_uncond_{never, always} into advance_jump_cond
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (12 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 13/21] target/sparc: Merge gen_branch2 into advance_pc Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 15/21] target/sparc: Pass displacement to advance_jump_cond Richard Henderson
` (7 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Handle these via TCG_COND_{ALWAYS,NEVER}.
Allow dc->npc to be variable, using gen_mov_pc_npc.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 74 ++++++++++++++++------------------------
1 file changed, 30 insertions(+), 44 deletions(-)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index e134ba8821..cbee5435a3 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -2378,37 +2378,37 @@ static bool advance_pc(DisasContext *dc)
* Major opcodes 00 and 01 -- branches, call, and sethi
*/
-static bool advance_jump_uncond_never(DisasContext *dc, bool annul)
-{
- if (annul) {
- dc->pc = dc->npc + 4;
- dc->npc = dc->pc + 4;
- } else {
- dc->pc = dc->npc;
- dc->npc = dc->pc + 4;
- }
- return true;
-}
-
-static bool advance_jump_uncond_always(DisasContext *dc, bool annul,
- target_ulong dest)
-{
- if (annul) {
- dc->pc = dest;
- dc->npc = dest + 4;
- } else {
- dc->pc = dc->npc;
- dc->npc = dest;
- tcg_gen_mov_tl(cpu_pc, cpu_npc);
- }
- return true;
-}
-
static bool advance_jump_cond(DisasContext *dc, DisasCompare *cmp,
bool annul, target_ulong dest)
{
target_ulong npc;
+ if (cmp->cond == TCG_COND_ALWAYS) {
+ if (annul) {
+ dc->pc = dest;
+ dc->npc = dest + 4;
+ } else {
+ gen_mov_pc_npc(dc);
+ dc->npc = dest;
+ }
+ return true;
+ }
+
+ if (cmp->cond == TCG_COND_NEVER) {
+ npc = dc->npc;
+ if (npc & 3) {
+ gen_mov_pc_npc(dc);
+ if (annul) {
+ tcg_gen_addi_tl(cpu_pc, cpu_pc, 4);
+ }
+ tcg_gen_addi_tl(cpu_npc, cpu_pc, 4);
+ } else {
+ dc->pc = npc + (annul ? 4 : 0);
+ dc->npc = dc->pc + 4;
+ }
+ return true;
+ }
+
flush_cond(dc);
npc = dc->npc;
@@ -2478,15 +2478,8 @@ static bool do_bpcc(DisasContext *dc, arg_bcc *a)
target_long target = address_mask_i(dc, dc->pc + a->i * 4);
DisasCompare cmp;
- switch (a->cond) {
- case 0x0:
- return advance_jump_uncond_never(dc, a->a);
- case 0x8:
- return advance_jump_uncond_always(dc, a->a, target);
- default:
- gen_compare(&cmp, a->cc, a->cond, dc);
- return advance_jump_cond(dc, &cmp, a->a, target);
- }
+ gen_compare(&cmp, a->cc, a->cond, dc);
+ return advance_jump_cond(dc, &cmp, a->a, target);
}
TRANS(Bicc, ALL, do_bpcc, a)
@@ -2500,15 +2493,8 @@ static bool do_fbpfcc(DisasContext *dc, arg_bcc *a)
if (gen_trap_ifnofpu(dc)) {
return true;
}
- switch (a->cond) {
- case 0x0:
- return advance_jump_uncond_never(dc, a->a);
- case 0x8:
- return advance_jump_uncond_always(dc, a->a, target);
- default:
- gen_fcompare(&cmp, a->cc, a->cond);
- return advance_jump_cond(dc, &cmp, a->a, target);
- }
+ gen_fcompare(&cmp, a->cc, a->cond);
+ return advance_jump_cond(dc, &cmp, a->a, target);
}
TRANS(FBPfcc, 64, do_fbpfcc, a)
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 15/21] target/sparc: Pass displacement to advance_jump_cond
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (13 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 14/21] target/sparc: Merge advance_jump_uncond_{never, always} into advance_jump_cond Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 16/21] target/sparc: Merge gen_op_next_insn into only caller Richard Henderson
` (6 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 12 +++++-------
1 file changed, 5 insertions(+), 7 deletions(-)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index cbee5435a3..1233911b69 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -2379,8 +2379,9 @@ static bool advance_pc(DisasContext *dc)
*/
static bool advance_jump_cond(DisasContext *dc, DisasCompare *cmp,
- bool annul, target_ulong dest)
+ bool annul, int disp)
{
+ target_ulong dest = address_mask_i(dc, dc->pc + disp * 4);
target_ulong npc;
if (cmp->cond == TCG_COND_ALWAYS) {
@@ -2475,11 +2476,10 @@ static bool gen_trap_float128(DisasContext *dc)
static bool do_bpcc(DisasContext *dc, arg_bcc *a)
{
- target_long target = address_mask_i(dc, dc->pc + a->i * 4);
DisasCompare cmp;
gen_compare(&cmp, a->cc, a->cond, dc);
- return advance_jump_cond(dc, &cmp, a->a, target);
+ return advance_jump_cond(dc, &cmp, a->a, a->i);
}
TRANS(Bicc, ALL, do_bpcc, a)
@@ -2487,14 +2487,13 @@ TRANS(BPcc, 64, do_bpcc, a)
static bool do_fbpfcc(DisasContext *dc, arg_bcc *a)
{
- target_long target = address_mask_i(dc, dc->pc + a->i * 4);
DisasCompare cmp;
if (gen_trap_ifnofpu(dc)) {
return true;
}
gen_fcompare(&cmp, a->cc, a->cond);
- return advance_jump_cond(dc, &cmp, a->a, target);
+ return advance_jump_cond(dc, &cmp, a->a, a->i);
}
TRANS(FBPfcc, 64, do_fbpfcc, a)
@@ -2502,7 +2501,6 @@ TRANS(FBfcc, ALL, do_fbpfcc, a)
static bool trans_BPr(DisasContext *dc, arg_BPr *a)
{
- target_long target = address_mask_i(dc, dc->pc + a->i * 4);
DisasCompare cmp;
if (!avail_64(dc)) {
@@ -2513,7 +2511,7 @@ static bool trans_BPr(DisasContext *dc, arg_BPr *a)
}
gen_compare_reg(&cmp, a->cond, gen_load_gpr(dc, a->rs1));
- return advance_jump_cond(dc, &cmp, a->a, target);
+ return advance_jump_cond(dc, &cmp, a->a, a->i);
}
static bool trans_CALL(DisasContext *dc, arg_CALL *a)
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 16/21] target/sparc: Merge gen_op_next_insn into only caller
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (14 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 15/21] target/sparc: Pass displacement to advance_jump_cond Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 17/21] target/sparc: Record entire jump condition in DisasContext Richard Henderson
` (5 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 9 ++-------
1 file changed, 2 insertions(+), 7 deletions(-)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 1233911b69..0bbe4cff3b 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -1015,12 +1015,6 @@ static void gen_mov_pc_npc(DisasContext *dc)
}
}
-static void gen_op_next_insn(void)
-{
- tcg_gen_mov_tl(cpu_pc, cpu_npc);
- tcg_gen_addi_tl(cpu_npc, cpu_npc, 4);
-}
-
static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
DisasContext *dc)
{
@@ -2346,7 +2340,8 @@ static bool advance_pc(DisasContext *dc)
case DYNAMIC_PC:
case DYNAMIC_PC_LOOKUP:
dc->pc = dc->npc;
- gen_op_next_insn();
+ tcg_gen_mov_tl(cpu_pc, cpu_npc);
+ tcg_gen_addi_tl(cpu_npc, cpu_npc, 4);
break;
case JUMP_PC:
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 17/21] target/sparc: Record entire jump condition in DisasContext
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (15 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 16/21] target/sparc: Merge gen_op_next_insn into only caller Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 18/21] target/sparc: Discard cpu_cond at the end of each insn Richard Henderson
` (4 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Use the original condition instead of consuming cpu_cond,
which will now only be live along exception paths.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 27 ++++++++++++++++-----------
1 file changed, 16 insertions(+), 11 deletions(-)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 0bbe4cff3b..5c9a3d45fa 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -146,6 +146,12 @@ static TCGv_i64 cpu_fpr[TARGET_DPREGS];
# define env64_field_offsetof(X) ({ qemu_build_not_reached(); 0; })
#endif
+typedef struct DisasCompare {
+ TCGCond cond;
+ TCGv c1;
+ int c2;
+} DisasCompare;
+
typedef struct DisasDelayException {
struct DisasDelayException *next;
TCGLabel *lab;
@@ -159,7 +165,11 @@ typedef struct DisasContext {
DisasContextBase base;
target_ulong pc; /* current Program Counter: integer or DYNAMIC_PC */
target_ulong npc; /* next PC: integer or DYNAMIC_PC or JUMP_PC */
- target_ulong jump_pc[2]; /* used when JUMP_PC pc value is used */
+
+ /* Used when JUMP_PC value is used. */
+ DisasCompare jump;
+ target_ulong jump_pc[2];
+
int mem_idx;
bool fpu_enabled;
bool address_mask_32bit;
@@ -178,12 +188,6 @@ typedef struct DisasContext {
DisasDelayException *delay_excp_list;
} DisasContext;
-typedef struct {
- TCGCond cond;
- TCGv c1;
- int c2;
-} DisasCompare;
-
// This function uses non-native bit order
#define GET_FIELD(X, FROM, TO) \
((X) >> (31 - (TO)) & ((1 << ((TO) - (FROM) + 1)) - 1))
@@ -912,9 +916,9 @@ static void gen_generic_branch(DisasContext *dc)
{
TCGv npc0 = tcg_constant_tl(dc->jump_pc[0]);
TCGv npc1 = tcg_constant_tl(dc->jump_pc[1]);
- TCGv zero = tcg_constant_tl(0);
+ TCGv c2 = tcg_constant_tl(dc->jump.c2);
- tcg_gen_movcond_tl(TCG_COND_NE, cpu_npc, cpu_cond, zero, npc0, npc1);
+ tcg_gen_movcond_tl(dc->jump.cond, cpu_npc, dc->jump.c1, c2, npc0, npc1);
}
/* call this function before using the condition register as it may
@@ -2347,7 +2351,7 @@ static bool advance_pc(DisasContext *dc)
case JUMP_PC:
/* we can do a static jump */
l1 = gen_new_label();
- tcg_gen_brcondi_tl(TCG_COND_NE, cpu_cond, 0, l1);
+ tcg_gen_brcondi_tl(dc->jump.cond, dc->jump.c1, dc->jump.c2, l1);
/* jump not taken */
gen_goto_tb(dc, 1, dc->jump_pc[1], dc->jump_pc[1] + 4);
@@ -2434,9 +2438,10 @@ static bool advance_jump_cond(DisasContext *dc, DisasCompare *cmp,
}
} else {
dc->pc = npc;
+ dc->npc = JUMP_PC;
+ dc->jump = *cmp;
dc->jump_pc[0] = dest;
dc->jump_pc[1] = npc + 4;
- dc->npc = JUMP_PC;
/* The condition for cpu_cond is always NE -- normalize. */
if (cmp->cond == TCG_COND_NE) {
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 18/21] target/sparc: Discard cpu_cond at the end of each insn
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (16 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 17/21] target/sparc: Record entire jump condition in DisasContext Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 19/21] target/sparc: Implement UDIVX and SDIVX inline Richard Henderson
` (3 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
If the insn raises no exceptions, there will be no path in which
cpu_cond is used, and so the computation may be optimized away.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 27 +++++++++++++++++++++++++++
1 file changed, 27 insertions(+)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 5c9a3d45fa..3564c6032e 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -171,6 +171,7 @@ typedef struct DisasContext {
target_ulong jump_pc[2];
int mem_idx;
+ bool cpu_cond_live;
bool fpu_enabled;
bool address_mask_32bit;
#ifndef CONFIG_USER_ONLY
@@ -912,6 +913,19 @@ static void gen_op_eval_fbo(TCGv dst, TCGv src, unsigned int fcc_offset)
tcg_gen_xori_tl(dst, dst, 0x1);
}
+static void finishing_insn(DisasContext *dc)
+{
+ /*
+ * From here, there is no future path through an unwinding exception.
+ * If the current insn cannot raise an exception, the computation of
+ * cpu_cond may be able to be elided.
+ */
+ if (dc->cpu_cond_live) {
+ tcg_gen_discard_tl(cpu_cond);
+ dc->cpu_cond_live = false;
+ }
+}
+
static void gen_generic_branch(DisasContext *dc)
{
TCGv npc0 = tcg_constant_tl(dc->jump_pc[0]);
@@ -958,6 +972,7 @@ static void save_state(DisasContext *dc)
static void gen_exception(DisasContext *dc, int which)
{
+ finishing_insn(dc);
save_state(dc);
gen_helper_raise_exception(tcg_env, tcg_constant_i32(which));
dc->base.is_jmp = DISAS_NORETURN;
@@ -999,6 +1014,8 @@ static void gen_check_align(DisasContext *dc, TCGv addr, int mask)
static void gen_mov_pc_npc(DisasContext *dc)
{
+ finishing_insn(dc);
+
if (dc->npc & 3) {
switch (dc->npc) {
case JUMP_PC:
@@ -2339,6 +2356,8 @@ static bool advance_pc(DisasContext *dc)
{
TCGLabel *l1;
+ finishing_insn(dc);
+
if (dc->npc & 3) {
switch (dc->npc) {
case DYNAMIC_PC:
@@ -2383,6 +2402,8 @@ static bool advance_jump_cond(DisasContext *dc, DisasCompare *cmp,
target_ulong dest = address_mask_i(dc, dc->pc + disp * 4);
target_ulong npc;
+ finishing_insn(dc);
+
if (cmp->cond == TCG_COND_ALWAYS) {
if (annul) {
dc->pc = dest;
@@ -2449,6 +2470,7 @@ static bool advance_jump_cond(DisasContext *dc, DisasCompare *cmp,
} else {
tcg_gen_setcondi_tl(cmp->cond, cpu_cond, cmp->c1, cmp->c2);
}
+ dc->cpu_cond_live = true;
}
}
return true;
@@ -2585,6 +2607,8 @@ static bool do_tcc(DisasContext *dc, int cond, int cc,
tcg_gen_addi_i32(trap, trap, TT_TRAP);
}
+ finishing_insn(dc);
+
/* Trap always. */
if (cond == 8) {
save_state(dc);
@@ -3201,6 +3225,7 @@ TRANS(WRSTICK_CMPR, 64, do_wr_special, a, supervisor(dc), do_wrstick_cmpr)
static void do_wrpowerdown(DisasContext *dc, TCGv src)
{
+ finishing_insn(dc);
save_state(dc);
gen_helper_power_down(tcg_env);
}
@@ -5080,6 +5105,8 @@ static void sparc_tr_tb_stop(DisasContextBase *dcbase, CPUState *cs)
DisasDelayException *e, *e_next;
bool may_lookup;
+ finishing_insn(dc);
+
switch (dc->base.is_jmp) {
case DISAS_NEXT:
case DISAS_TOO_MANY:
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 19/21] target/sparc: Implement UDIVX and SDIVX inline
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (17 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 18/21] target/sparc: Discard cpu_cond at the end of each insn Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 20/21] target/sparc: Implement UDIV inline Richard Henderson
` (2 subsequent siblings)
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/helper.h | 4 --
target/sparc/insns.decode | 4 +-
target/sparc/helper.c | 24 ---------
target/sparc/translate.c | 109 +++++++++++++++++++++++++++++++++-----
4 files changed, 97 insertions(+), 44 deletions(-)
diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index decd94c0d6..55eff66283 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -31,10 +31,6 @@ DEF_HELPER_FLAGS_3(udiv, TCG_CALL_NO_WG, i64, env, tl, tl)
DEF_HELPER_FLAGS_3(sdiv, TCG_CALL_NO_WG, i64, env, tl, tl)
DEF_HELPER_3(taddcctv, tl, env, tl, tl)
DEF_HELPER_3(tsubcctv, tl, env, tl, tl)
-#ifdef TARGET_SPARC64
-DEF_HELPER_FLAGS_3(sdivx, TCG_CALL_NO_WG, s64, env, s64, s64)
-DEF_HELPER_FLAGS_3(udivx, TCG_CALL_NO_WG, i64, env, i64, i64)
-#endif
#if !defined(CONFIG_USER_ONLY) || defined(TARGET_SPARC64)
DEF_HELPER_FLAGS_4(ld_asi, TCG_CALL_NO_WG, i64, env, tl, int, i32)
DEF_HELPER_FLAGS_5(st_asi, TCG_CALL_NO_WG, void, env, tl, i64, int, i32)
diff --git a/target/sparc/insns.decode b/target/sparc/insns.decode
index 0552f1447d..52f54b87cc 100644
--- a/target/sparc/insns.decode
+++ b/target/sparc/insns.decode
@@ -182,8 +182,8 @@ UMUL 10 ..... 0.1010 ..... . ............. @r_r_ri_cc
SMUL 10 ..... 0.1011 ..... . ............. @r_r_ri_cc
MULScc 10 ..... 100100 ..... . ............. @r_r_ri_cc1
-UDIVX 10 ..... 001101 ..... . ............. @r_r_ri_cc0
-SDIVX 10 ..... 101101 ..... . ............. @r_r_ri_cc0
+UDIVX 10 ..... 001101 ..... . ............. @r_r_ri
+SDIVX 10 ..... 101101 ..... . ............. @r_r_ri
UDIV 10 ..... 0.1110 ..... . ............. @r_r_ri_cc
SDIV 10 ..... 0.1111 ..... . ............. @r_r_ri_cc
diff --git a/target/sparc/helper.c b/target/sparc/helper.c
index 6117e99b55..bd10b60e4b 100644
--- a/target/sparc/helper.c
+++ b/target/sparc/helper.c
@@ -129,30 +129,6 @@ uint64_t helper_sdiv(CPUSPARCState *env, target_ulong a, target_ulong b)
return (uint32_t)r;
}
-#ifdef TARGET_SPARC64
-int64_t helper_sdivx(CPUSPARCState *env, int64_t a, int64_t b)
-{
- if (b == 0) {
- /* Raise divide by zero trap. */
- cpu_raise_exception_ra(env, TT_DIV_ZERO, GETPC());
- } else if (b == -1) {
- /* Avoid overflow trap with i386 divide insn. */
- return -a;
- } else {
- return a / b;
- }
-}
-
-uint64_t helper_udivx(CPUSPARCState *env, uint64_t a, uint64_t b)
-{
- if (b == 0) {
- /* Raise divide by zero trap. */
- cpu_raise_exception_ra(env, TT_DIV_ZERO, GETPC());
- }
- return a / b;
-}
-#endif
-
target_ulong helper_taddcctv(CPUSPARCState *env, target_ulong src1,
target_ulong src2)
{
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 3564c6032e..95cc4c71f4 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -51,12 +51,10 @@
# define gen_helper_restored(E) qemu_build_not_reached()
# define gen_helper_retry(E) qemu_build_not_reached()
# define gen_helper_saved(E) qemu_build_not_reached()
-# define gen_helper_sdivx(D, E, A, B) qemu_build_not_reached()
# define gen_helper_set_softint(E, S) qemu_build_not_reached()
# define gen_helper_tick_get_count(D, E, T, C) qemu_build_not_reached()
# define gen_helper_tick_set_count(P, S) qemu_build_not_reached()
# define gen_helper_tick_set_limit(P, S) qemu_build_not_reached()
-# define gen_helper_udivx(D, E, A, B) qemu_build_not_reached()
# define gen_helper_wrccr(E, S) qemu_build_not_reached()
# define gen_helper_wrcwp(E, S) qemu_build_not_reached()
# define gen_helper_wrgl(E, S) qemu_build_not_reached()
@@ -579,16 +577,6 @@ static void gen_op_smul(TCGv dst, TCGv src1, TCGv src2)
gen_op_multiply(dst, src1, src2, 1);
}
-static void gen_op_udivx(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_helper_udivx(dst, tcg_env, src1, src2);
-}
-
-static void gen_op_sdivx(TCGv dst, TCGv src1, TCGv src2)
-{
- gen_helper_sdivx(dst, tcg_env, src1, src2);
-}
-
static void gen_op_udiv(TCGv dst, TCGv src1, TCGv src2)
{
#ifdef TARGET_SPARC64
@@ -3580,8 +3568,6 @@ TRANS(UMUL, MUL, do_logic, a, gen_op_umul, NULL)
TRANS(SMUL, MUL, do_logic, a, gen_op_smul, NULL)
TRANS(MULScc, ALL, do_arith, a, NULL, NULL, gen_op_mulscc)
-TRANS(UDIVX, 64, do_arith, a, gen_op_udivx, NULL, NULL)
-TRANS(SDIVX, 64, do_arith, a, gen_op_sdivx, NULL, NULL)
TRANS(UDIV, DIV, do_arith, a, gen_op_udiv, NULL, gen_op_udivcc)
TRANS(SDIV, DIV, do_arith, a, gen_op_sdiv, NULL, gen_op_sdivcc)
@@ -3605,6 +3591,101 @@ static bool trans_OR(DisasContext *dc, arg_r_r_ri_cc *a)
return do_logic(dc, a, tcg_gen_or_tl, tcg_gen_ori_tl);
}
+static bool trans_UDIVX(DisasContext *dc, arg_r_r_ri *a)
+{
+ TCGv dst, src1, src2;
+
+ if (!avail_64(dc)) {
+ return false;
+ }
+ /* For simplicity, we under-decoded the rs2 form. */
+ if (!a->imm && a->rs2_or_imm & ~0x1f) {
+ return false;
+ }
+
+ if (unlikely(a->rs2_or_imm == 0)) {
+ gen_exception(dc, TT_DIV_ZERO);
+ return true;
+ }
+
+ if (a->imm) {
+ src2 = tcg_constant_tl(a->rs2_or_imm);
+ } else {
+ TCGLabel *lab;
+
+ finishing_insn(dc);
+ flush_cond(dc);
+
+ lab = delay_exception(dc, TT_DIV_ZERO);
+ src2 = cpu_regs[a->rs2_or_imm];
+ tcg_gen_brcondi_tl(TCG_COND_EQ, src2, 0, lab);
+ }
+
+ dst = gen_dest_gpr(dc, a->rd);
+ src1 = gen_load_gpr(dc, a->rs1);
+
+ tcg_gen_divu_tl(dst, src1, src2);
+ gen_store_gpr(dc, a->rd, dst);
+ return advance_pc(dc);
+}
+
+static bool trans_SDIVX(DisasContext *dc, arg_r_r_ri *a)
+{
+ TCGv dst, src1, src2;
+
+ if (!avail_64(dc)) {
+ return false;
+ }
+ /* For simplicity, we under-decoded the rs2 form. */
+ if (!a->imm && a->rs2_or_imm & ~0x1f) {
+ return false;
+ }
+
+ if (unlikely(a->rs2_or_imm == 0)) {
+ gen_exception(dc, TT_DIV_ZERO);
+ return true;
+ }
+
+ dst = gen_dest_gpr(dc, a->rd);
+ src1 = gen_load_gpr(dc, a->rs1);
+
+ if (a->imm) {
+ if (unlikely(a->rs2_or_imm == -1)) {
+ tcg_gen_neg_tl(dst, src1);
+ gen_store_gpr(dc, a->rd, dst);
+ return advance_pc(dc);
+ }
+ src2 = tcg_constant_tl(a->rs2_or_imm);
+ } else {
+ TCGLabel *lab;
+ TCGv t1, t2;
+
+ finishing_insn(dc);
+ flush_cond(dc);
+
+ lab = delay_exception(dc, TT_DIV_ZERO);
+ src2 = cpu_regs[a->rs2_or_imm];
+ tcg_gen_brcondi_tl(TCG_COND_EQ, src2, 0, lab);
+
+ /*
+ * Need to avoid INT64_MIN / -1, which will trap on x86 host.
+ * Set SRC2 to 1 as a new divisor, to produce the correct result.
+ */
+ t1 = tcg_temp_new();
+ t2 = tcg_temp_new();
+ tcg_gen_setcondi_tl(TCG_COND_EQ, t1, src1, (target_long)INT64_MIN);
+ tcg_gen_setcondi_tl(TCG_COND_EQ, t2, src2, -1);
+ tcg_gen_and_tl(t1, t1, t2);
+ tcg_gen_movcond_tl(TCG_COND_NE, t1, t1, tcg_constant_tl(0),
+ tcg_constant_tl(1), src2);
+ src2 = t1;
+ }
+
+ tcg_gen_div_tl(dst, src1, src2);
+ gen_store_gpr(dc, a->rd, dst);
+ return advance_pc(dc);
+}
+
static bool gen_edge(DisasContext *dc, arg_r_r_r *a,
int width, bool cc, bool left)
{
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 20/21] target/sparc: Implement UDIV inline
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (18 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 19/21] target/sparc: Implement UDIVX and SDIVX inline Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-01 4:11 ` [PATCH v2 21/21] target/sparc: Check for invalid cond in gen_compare_reg Richard Henderson
2023-11-05 13:22 ` [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Mark Cave-Ayland
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/insns.decode | 3 +-
target/sparc/translate.c | 67 +++++++++++++++++++++++++++++++--------
2 files changed, 56 insertions(+), 14 deletions(-)
diff --git a/target/sparc/insns.decode b/target/sparc/insns.decode
index 52f54b87cc..2d26404cb2 100644
--- a/target/sparc/insns.decode
+++ b/target/sparc/insns.decode
@@ -184,7 +184,8 @@ MULScc 10 ..... 100100 ..... . ............. @r_r_ri_cc1
UDIVX 10 ..... 001101 ..... . ............. @r_r_ri
SDIVX 10 ..... 101101 ..... . ............. @r_r_ri
-UDIV 10 ..... 0.1110 ..... . ............. @r_r_ri_cc
+UDIV 10 ..... 001110 ..... . ............. @r_r_ri
+UDIVcc 10 ..... 011110 ..... . ............. @r_r_ri_cc1
SDIV 10 ..... 0.1111 ..... . ............. @r_r_ri_cc
TADDcc 10 ..... 100000 ..... . ............. @r_r_ri_cc1
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 95cc4c71f4..4b7d943bae 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -577,18 +577,6 @@ static void gen_op_smul(TCGv dst, TCGv src1, TCGv src2)
gen_op_multiply(dst, src1, src2, 1);
}
-static void gen_op_udiv(TCGv dst, TCGv src1, TCGv src2)
-{
-#ifdef TARGET_SPARC64
- gen_helper_udiv(dst, tcg_env, src1, src2);
- tcg_gen_ext32u_tl(dst, dst);
-#else
- TCGv_i64 t64 = tcg_temp_new_i64();
- gen_helper_udiv(t64, tcg_env, src1, src2);
- tcg_gen_trunc_i64_tl(dst, t64);
-#endif
-}
-
static void gen_op_sdiv(TCGv dst, TCGv src1, TCGv src2)
{
#ifdef TARGET_SPARC64
@@ -3568,7 +3556,7 @@ TRANS(UMUL, MUL, do_logic, a, gen_op_umul, NULL)
TRANS(SMUL, MUL, do_logic, a, gen_op_smul, NULL)
TRANS(MULScc, ALL, do_arith, a, NULL, NULL, gen_op_mulscc)
-TRANS(UDIV, DIV, do_arith, a, gen_op_udiv, NULL, gen_op_udivcc)
+TRANS(UDIVcc, DIV, do_arith, a, NULL, NULL, gen_op_udivcc)
TRANS(SDIV, DIV, do_arith, a, gen_op_sdiv, NULL, gen_op_sdivcc)
/* TODO: Should have feature bit -- comes in with UltraSparc T2. */
@@ -3591,6 +3579,59 @@ static bool trans_OR(DisasContext *dc, arg_r_r_ri_cc *a)
return do_logic(dc, a, tcg_gen_or_tl, tcg_gen_ori_tl);
}
+static bool trans_UDIV(DisasContext *dc, arg_r_r_ri *a)
+{
+ TCGv_i64 t1, t2;
+ TCGv dst;
+
+ if (!avail_DIV(dc)) {
+ return false;
+ }
+ /* For simplicity, we under-decoded the rs2 form. */
+ if (!a->imm && a->rs2_or_imm & ~0x1f) {
+ return false;
+ }
+
+ if (unlikely(a->rs2_or_imm == 0)) {
+ gen_exception(dc, TT_DIV_ZERO);
+ return true;
+ }
+
+ if (a->imm) {
+ t2 = tcg_constant_i64((uint32_t)a->rs2_or_imm);
+ } else {
+ TCGLabel *lab;
+ TCGv_i32 n2;
+
+ finishing_insn(dc);
+ flush_cond(dc);
+
+ n2 = tcg_temp_new_i32();
+ tcg_gen_trunc_tl_i32(n2, cpu_regs[a->rs2_or_imm]);
+
+ lab = delay_exception(dc, TT_DIV_ZERO);
+ tcg_gen_brcondi_i32(TCG_COND_EQ, n2, 0, lab);
+
+ t2 = tcg_temp_new_i64();
+#ifdef TARGET_SPARC64
+ tcg_gen_ext32u_i64(t2, cpu_regs[a->rs2_or_imm]);
+#else
+ tcg_gen_extu_i32_i64(t2, cpu_regs[a->rs2_or_imm]);
+#endif
+ }
+
+ t1 = tcg_temp_new_i64();
+ tcg_gen_concat_tl_i64(t1, gen_load_gpr(dc, a->rs1), cpu_y);
+
+ tcg_gen_divu_i64(t1, t1, t2);
+ tcg_gen_umin_i64(t1, t1, tcg_constant_i64(UINT32_MAX));
+
+ dst = gen_dest_gpr(dc, a->rd);
+ tcg_gen_trunc_i64_tl(dst, t1);
+ gen_store_gpr(dc, a->rd, dst);
+ return advance_pc(dc);
+}
+
static bool trans_UDIVX(DisasContext *dc, arg_r_r_ri *a)
{
TCGv dst, src1, src2;
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* [PATCH v2 21/21] target/sparc: Check for invalid cond in gen_compare_reg
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (19 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 20/21] target/sparc: Implement UDIV inline Richard Henderson
@ 2023-11-01 4:11 ` Richard Henderson
2023-11-05 13:22 ` [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Mark Cave-Ayland
21 siblings, 0 replies; 23+ messages in thread
From: Richard Henderson @ 2023-11-01 4:11 UTC (permalink / raw)
To: qemu-devel; +Cc: mark.cave-ayland
Consolidate the test here; drop the "inverted logic".
Fix MOVr and FMOVR, which were missing the invalid test.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
target/sparc/translate.c | 45 +++++++++++++++++++++++-----------------
1 file changed, 26 insertions(+), 19 deletions(-)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 4b7d943bae..6fc333a6b8 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -1189,24 +1189,29 @@ static void gen_fcompare(DisasCompare *cmp, unsigned int cc, unsigned int cond)
}
}
-// Inverted logic
-static const TCGCond gen_tcg_cond_reg[8] = {
- TCG_COND_NEVER, /* reserved */
- TCG_COND_NE,
- TCG_COND_GT,
- TCG_COND_GE,
- TCG_COND_NEVER, /* reserved */
- TCG_COND_EQ,
- TCG_COND_LE,
- TCG_COND_LT,
-};
-
-static void gen_compare_reg(DisasCompare *cmp, int cond, TCGv r_src)
+static bool gen_compare_reg(DisasCompare *cmp, int cond, TCGv r_src)
{
- cmp->cond = tcg_invert_cond(gen_tcg_cond_reg[cond]);
+ static const TCGCond cond_reg[4] = {
+ TCG_COND_NEVER, /* reserved */
+ TCG_COND_EQ,
+ TCG_COND_LE,
+ TCG_COND_LT,
+ };
+ TCGCond tcond;
+
+ if ((cond & 3) == 0) {
+ return false;
+ }
+ tcond = cond_reg[cond & 3];
+ if (cond & 4) {
+ tcond = tcg_invert_cond(tcond);
+ }
+
+ cmp->cond = tcond;
cmp->c1 = tcg_temp_new();
cmp->c2 = 0;
tcg_gen_mov_tl(cmp->c1, r_src);
+ return true;
}
static void gen_op_clear_ieee_excp_and_FTT(void)
@@ -2504,11 +2509,9 @@ static bool trans_BPr(DisasContext *dc, arg_BPr *a)
if (!avail_64(dc)) {
return false;
}
- if (gen_tcg_cond_reg[a->cond] == TCG_COND_NEVER) {
+ if (!gen_compare_reg(&cmp, a->cond, gen_load_gpr(dc, a->rs1))) {
return false;
}
-
- gen_compare_reg(&cmp, a->cond, gen_load_gpr(dc, a->rs1));
return advance_jump_cond(dc, &cmp, a->a, a->i);
}
@@ -4020,7 +4023,9 @@ static bool trans_MOVR(DisasContext *dc, arg_MOVR *a)
if (src2 == NULL) {
return false;
}
- gen_compare_reg(&cmp, a->cond, gen_load_gpr(dc, a->rs1));
+ if (!gen_compare_reg(&cmp, a->cond, gen_load_gpr(dc, a->rs1))) {
+ return false;
+ }
return do_mov_cond(dc, &cmp, a->rd, src2);
}
@@ -5007,6 +5012,9 @@ static bool do_fmovr(DisasContext *dc, arg_FMOVRs *a, bool is_128,
{
DisasCompare cmp;
+ if (!gen_compare_reg(&cmp, a->cond, gen_load_gpr(dc, a->rs1))) {
+ return false;
+ }
if (gen_trap_ifnofpu(dc)) {
return true;
}
@@ -5015,7 +5023,6 @@ static bool do_fmovr(DisasContext *dc, arg_FMOVRs *a, bool is_128,
}
gen_op_clear_ieee_excp_and_FTT();
- gen_compare_reg(&cmp, a->cond, gen_load_gpr(dc, a->rs1));
func(dc, &cmp, a->rd, a->rs2);
return advance_pc(dc);
}
--
2.34.1
^ permalink raw reply related [flat|nested] 23+ messages in thread
* Re: [PATCH v2 00/21] target/sparc: Cleanup condition codes etc
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
` (20 preceding siblings ...)
2023-11-01 4:11 ` [PATCH v2 21/21] target/sparc: Check for invalid cond in gen_compare_reg Richard Henderson
@ 2023-11-05 13:22 ` Mark Cave-Ayland
21 siblings, 0 replies; 23+ messages in thread
From: Mark Cave-Ayland @ 2023-11-05 13:22 UTC (permalink / raw)
To: Richard Henderson, qemu-devel
On 01/11/2023 04:11, Richard Henderson wrote:
> This was part of my guess for some of the performance problems.
>
> I saw compute_all_sub quite high in the profile at some point, and I
> believe that the test case has a partially rotated loop such that "cmp"
> is in a delay slot, and so the gen_compare fast path for CC_OP_SUB is
> not visible to the conditional branch that uses the output of the compare.
> Which means that helper_compute_psr gets called more often that we'd like.
>
> Since almost all Sparc instructions that set cc also have a version of
> the instruction that does not set cc, we can trust that the compiler
> has only used the cc-setting version when it is actually required.
> Thus, unlike CISC processors, there is very little scope for optimization
> of the flags -- we might as well compute them immediately.
>
> Move away from CC_OP to explicit computation of conditions. This is
> modeled on target/arm for the (mostly) separate representation of the bits.
> We can pack icc.[NV] and xcc.[NV] into the same target_ulong, but Z and C
> cannot share. (For "normal" setting of Z, we could share, but it is
> possible to set xcc.Z and !icc.Z via explicit write to %ccr, and for
> that we have to have two variables.)
>
> After removing CC_OP, clean up the handling of conditions so that we can
> minimize additional setcond required for env->cond.
>
> Finally, inline some division, which can make use of the new out-of-line
> exception path, which means we can expand UDIVX and SDIVX with very few
> host insns. The 64/32 UDIV insn needs only a few more. Leave UDIVcc and
> SDIV* out of line, as the overflow and saturation computation in these
> cases is really too large to inline.
>
> r~
I've tested this series by running through my OpenBIOS boot tests for SPARC32 and
SPARC64 and didn't spot any obvious regressions, so:
Tested-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Acked-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
> Richard Henderson (21):
> target/sparc: Introduce cpu_put_psr_icc
> target/sparc: Split psr and xcc into components
> target/sparc: Remove CC_OP_LOGIC
> target/sparc: Remove CC_OP_DIV
> target/sparc: Remove CC_OP_ADD, CC_OP_ADDX, CC_OP_TADD
> target/sparc: Remove CC_OP_SUB, CC_OP_SUBX, CC_OP_TSUB
> target/sparc: Remove CC_OP_TADDTV, CC_OP_TSUBTV
> target/sparc: Remove CC_OP leftovers
> target/sparc: Remove DisasCompare.is_bool
> target/sparc: Change DisasCompare.c2 to int
> target/sparc: Always copy conditions into a new temporary
> target/sparc: Do flush_cond in advance_jump_cond
> target/sparc: Merge gen_branch2 into advance_pc
> target/sparc: Merge advance_jump_uncond_{never,always} into
> advance_jump_cond
> target/sparc: Pass displacement to advance_jump_cond
> target/sparc: Merge gen_op_next_insn into only caller
> target/sparc: Record entire jump condition in DisasContext
> target/sparc: Discard cpu_cond at the end of each insn
> target/sparc: Implement UDIVX and SDIVX inline
> target/sparc: Implement UDIV inline
> target/sparc: Check for invalid cond in gen_compare_reg
>
> linux-user/sparc/target_cpu.h | 17 +-
> target/sparc/cpu.h | 58 +-
> target/sparc/helper.h | 12 +-
> target/sparc/insns.decode | 7 +-
> linux-user/sparc/cpu_loop.c | 11 +-
> linux-user/sparc/signal.c | 2 +-
> target/sparc/cc_helper.c | 471 ------------
> target/sparc/cpu.c | 1 -
> target/sparc/helper.c | 171 ++---
> target/sparc/int32_helper.c | 5 -
> target/sparc/int64_helper.c | 5 -
> target/sparc/machine.c | 45 +-
> target/sparc/translate.c | 1333 ++++++++++++++-------------------
> target/sparc/win_helper.c | 56 +-
> target/sparc/meson.build | 1 -
> 15 files changed, 789 insertions(+), 1406 deletions(-)
> delete mode 100644 target/sparc/cc_helper.c
>
ATB,
Mark.
^ permalink raw reply [flat|nested] 23+ messages in thread
end of thread, other threads:[~2023-11-05 13:24 UTC | newest]
Thread overview: 23+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2023-11-01 4:11 [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Richard Henderson
2023-11-01 4:11 ` [PATCH v2 01/21] target/sparc: Introduce cpu_put_psr_icc Richard Henderson
2023-11-01 4:11 ` [PATCH v2 02/21] target/sparc: Split psr and xcc into components Richard Henderson
2023-11-01 4:11 ` [PATCH v2 03/21] target/sparc: Remove CC_OP_LOGIC Richard Henderson
2023-11-01 4:11 ` [PATCH v2 04/21] target/sparc: Remove CC_OP_DIV Richard Henderson
2023-11-01 4:11 ` [PATCH v2 05/21] target/sparc: Remove CC_OP_ADD, CC_OP_ADDX, CC_OP_TADD Richard Henderson
2023-11-01 4:11 ` [PATCH v2 06/21] target/sparc: Remove CC_OP_SUB, CC_OP_SUBX, CC_OP_TSUB Richard Henderson
2023-11-01 4:11 ` [PATCH v2 07/21] target/sparc: Remove CC_OP_TADDTV, CC_OP_TSUBTV Richard Henderson
2023-11-01 4:11 ` [PATCH v2 08/21] target/sparc: Remove CC_OP leftovers Richard Henderson
2023-11-01 4:11 ` [PATCH v2 09/21] target/sparc: Remove DisasCompare.is_bool Richard Henderson
2023-11-01 4:11 ` [PATCH v2 10/21] target/sparc: Change DisasCompare.c2 to int Richard Henderson
2023-11-01 4:11 ` [PATCH v2 11/21] target/sparc: Always copy conditions into a new temporary Richard Henderson
2023-11-01 4:11 ` [PATCH v2 12/21] target/sparc: Do flush_cond in advance_jump_cond Richard Henderson
2023-11-01 4:11 ` [PATCH v2 13/21] target/sparc: Merge gen_branch2 into advance_pc Richard Henderson
2023-11-01 4:11 ` [PATCH v2 14/21] target/sparc: Merge advance_jump_uncond_{never, always} into advance_jump_cond Richard Henderson
2023-11-01 4:11 ` [PATCH v2 15/21] target/sparc: Pass displacement to advance_jump_cond Richard Henderson
2023-11-01 4:11 ` [PATCH v2 16/21] target/sparc: Merge gen_op_next_insn into only caller Richard Henderson
2023-11-01 4:11 ` [PATCH v2 17/21] target/sparc: Record entire jump condition in DisasContext Richard Henderson
2023-11-01 4:11 ` [PATCH v2 18/21] target/sparc: Discard cpu_cond at the end of each insn Richard Henderson
2023-11-01 4:11 ` [PATCH v2 19/21] target/sparc: Implement UDIVX and SDIVX inline Richard Henderson
2023-11-01 4:11 ` [PATCH v2 20/21] target/sparc: Implement UDIV inline Richard Henderson
2023-11-01 4:11 ` [PATCH v2 21/21] target/sparc: Check for invalid cond in gen_compare_reg Richard Henderson
2023-11-05 13:22 ` [PATCH v2 00/21] target/sparc: Cleanup condition codes etc Mark Cave-Ayland
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).