summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorths <ths@c046a42c-6fe2-441c-8c8c-71466251a162>2008-06-19 18:35:02 +0000
committerths <ths@c046a42c-6fe2-441c-8c8c-71466251a162>2008-06-19 18:35:02 +0000
commita16336e4792c4f092bd2fdcd20c9d70c979b2b22 (patch)
treee68efa6208ef973029514d7e513f7ab6e2c03319
parent8f2ad0a3fc5e3569183d44bf1c7fcb95294be4c0 (diff)
downloadqemu-a16336e4792c4f092bd2fdcd20c9d70c979b2b22.tar.gz
Convert remaining MIPS FP instructions to TCG.
git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@4753 c046a42c-6fe2-441c-8c8c-71466251a162
-rw-r--r--target-mips/helper.h7
-rw-r--r--target-mips/op.c292
-rw-r--r--target-mips/op_helper.c87
-rw-r--r--target-mips/translate.c403
4 files changed, 426 insertions, 363 deletions
diff --git a/target-mips/helper.h b/target-mips/helper.h
index 1092f8733c..cba9cd7ebf 100644
--- a/target-mips/helper.h
+++ b/target-mips/helper.h
@@ -154,6 +154,7 @@ DEF_HELPER(void, do_float_mulr_ps, (void))
#define FOP_PROTO(op) \
DEF_HELPER(void, do_float_ ## op ## _s, (void)) \
DEF_HELPER(void, do_float_ ## op ## _d, (void))
+FOP_PROTO(sqrt)
FOP_PROTO(roundl)
FOP_PROTO(roundw)
FOP_PROTO(truncl)
@@ -174,6 +175,12 @@ FOP_PROTO(add)
FOP_PROTO(sub)
FOP_PROTO(mul)
FOP_PROTO(div)
+FOP_PROTO(abs)
+FOP_PROTO(chs)
+FOP_PROTO(muladd)
+FOP_PROTO(mulsub)
+FOP_PROTO(nmuladd)
+FOP_PROTO(nmulsub)
FOP_PROTO(recip1)
FOP_PROTO(recip2)
FOP_PROTO(rsqrt1)
diff --git a/target-mips/op.c b/target-mips/op.c
index 2c99ad754d..5e778f2b37 100644
--- a/target-mips/op.c
+++ b/target-mips/op.c
@@ -295,295 +295,3 @@ void op_mulshiu (void)
#else
# define DEBUG_FPU_STATE() do { } while(0)
#endif
-
-/* Float support.
- Single precition routines have a "s" suffix, double precision a
- "d" suffix, 32bit integer "w", 64bit integer "l", paired singe "ps",
- paired single lowwer "pl", paired single upper "pu". */
-
-#define FLOAT_OP(name, p) void OPPROTO op_float_##name##_##p(void)
-
-FLOAT_OP(pll, ps)
-{
- DT2 = ((uint64_t)WT0 << 32) | WT1;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(plu, ps)
-{
- DT2 = ((uint64_t)WT0 << 32) | WTH1;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(pul, ps)
-{
- DT2 = ((uint64_t)WTH0 << 32) | WT1;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(puu, ps)
-{
- DT2 = ((uint64_t)WTH0 << 32) | WTH1;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-
-FLOAT_OP(movf, d)
-{
- if (!(env->fpu->fcr31 & PARAM1))
- DT2 = DT0;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movf, s)
-{
- if (!(env->fpu->fcr31 & PARAM1))
- WT2 = WT0;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movf, ps)
-{
- unsigned int mask = GET_FP_COND (env->fpu) >> PARAM1;
- if (!(mask & 1))
- WT2 = WT0;
- if (!(mask & 2))
- WTH2 = WTH0;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movt, d)
-{
- if (env->fpu->fcr31 & PARAM1)
- DT2 = DT0;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movt, s)
-{
- if (env->fpu->fcr31 & PARAM1)
- WT2 = WT0;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movt, ps)
-{
- unsigned int mask = GET_FP_COND (env->fpu) >> PARAM1;
- if (mask & 1)
- WT2 = WT0;
- if (mask & 2)
- WTH2 = WTH0;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movz, d)
-{
- if (!T0)
- DT2 = DT0;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movz, s)
-{
- if (!T0)
- WT2 = WT0;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movz, ps)
-{
- if (!T0) {
- WT2 = WT0;
- WTH2 = WTH0;
- }
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movn, d)
-{
- if (T0)
- DT2 = DT0;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movn, s)
-{
- if (T0)
- WT2 = WT0;
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-FLOAT_OP(movn, ps)
-{
- if (T0) {
- WT2 = WT0;
- WTH2 = WTH0;
- }
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-
-/* ternary operations */
-#define FLOAT_TERNOP(name1, name2) \
-FLOAT_OP(name1 ## name2, d) \
-{ \
- FDT0 = float64_ ## name1 (FDT0, FDT1, &env->fpu->fp_status); \
- FDT2 = float64_ ## name2 (FDT0, FDT2, &env->fpu->fp_status); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-} \
-FLOAT_OP(name1 ## name2, s) \
-{ \
- FST0 = float32_ ## name1 (FST0, FST1, &env->fpu->fp_status); \
- FST2 = float32_ ## name2 (FST0, FST2, &env->fpu->fp_status); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-} \
-FLOAT_OP(name1 ## name2, ps) \
-{ \
- FST0 = float32_ ## name1 (FST0, FST1, &env->fpu->fp_status); \
- FSTH0 = float32_ ## name1 (FSTH0, FSTH1, &env->fpu->fp_status); \
- FST2 = float32_ ## name2 (FST0, FST2, &env->fpu->fp_status); \
- FSTH2 = float32_ ## name2 (FSTH0, FSTH2, &env->fpu->fp_status); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-}
-FLOAT_TERNOP(mul, add)
-FLOAT_TERNOP(mul, sub)
-#undef FLOAT_TERNOP
-
-/* negated ternary operations */
-#define FLOAT_NTERNOP(name1, name2) \
-FLOAT_OP(n ## name1 ## name2, d) \
-{ \
- FDT0 = float64_ ## name1 (FDT0, FDT1, &env->fpu->fp_status); \
- FDT2 = float64_ ## name2 (FDT0, FDT2, &env->fpu->fp_status); \
- FDT2 = float64_chs(FDT2); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-} \
-FLOAT_OP(n ## name1 ## name2, s) \
-{ \
- FST0 = float32_ ## name1 (FST0, FST1, &env->fpu->fp_status); \
- FST2 = float32_ ## name2 (FST0, FST2, &env->fpu->fp_status); \
- FST2 = float32_chs(FST2); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-} \
-FLOAT_OP(n ## name1 ## name2, ps) \
-{ \
- FST0 = float32_ ## name1 (FST0, FST1, &env->fpu->fp_status); \
- FSTH0 = float32_ ## name1 (FSTH0, FSTH1, &env->fpu->fp_status); \
- FST2 = float32_ ## name2 (FST0, FST2, &env->fpu->fp_status); \
- FSTH2 = float32_ ## name2 (FSTH0, FSTH2, &env->fpu->fp_status); \
- FST2 = float32_chs(FST2); \
- FSTH2 = float32_chs(FSTH2); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-}
-FLOAT_NTERNOP(mul, add)
-FLOAT_NTERNOP(mul, sub)
-#undef FLOAT_NTERNOP
-
-/* unary operations, modifying fp status */
-#define FLOAT_UNOP(name) \
-FLOAT_OP(name, d) \
-{ \
- FDT2 = float64_ ## name(FDT0, &env->fpu->fp_status); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-} \
-FLOAT_OP(name, s) \
-{ \
- FST2 = float32_ ## name(FST0, &env->fpu->fp_status); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-}
-FLOAT_UNOP(sqrt)
-#undef FLOAT_UNOP
-
-/* unary operations, not modifying fp status */
-#define FLOAT_UNOP(name) \
-FLOAT_OP(name, d) \
-{ \
- FDT2 = float64_ ## name(FDT0); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-} \
-FLOAT_OP(name, s) \
-{ \
- FST2 = float32_ ## name(FST0); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-} \
-FLOAT_OP(name, ps) \
-{ \
- FST2 = float32_ ## name(FST0); \
- FSTH2 = float32_ ## name(FSTH0); \
- DEBUG_FPU_STATE(); \
- FORCE_RET(); \
-}
-FLOAT_UNOP(abs)
-FLOAT_UNOP(chs)
-#undef FLOAT_UNOP
-
-FLOAT_OP(alnv, ps)
-{
- switch (T0 & 0x7) {
- case 0:
- FST2 = FST0;
- FSTH2 = FSTH0;
- break;
- case 4:
-#ifdef TARGET_WORDS_BIGENDIAN
- FSTH2 = FST0;
- FST2 = FSTH1;
-#else
- FSTH2 = FST1;
- FST2 = FSTH0;
-#endif
- break;
- default: /* unpredictable */
- break;
- }
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-
-void op_bc1f (void)
-{
- T0 = !!(~GET_FP_COND(env->fpu) & (0x1 << PARAM1));
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-void op_bc1any2f (void)
-{
- T0 = !!(~GET_FP_COND(env->fpu) & (0x3 << PARAM1));
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-void op_bc1any4f (void)
-{
- T0 = !!(~GET_FP_COND(env->fpu) & (0xf << PARAM1));
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-
-void op_bc1t (void)
-{
- T0 = !!(GET_FP_COND(env->fpu) & (0x1 << PARAM1));
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-void op_bc1any2t (void)
-{
- T0 = !!(GET_FP_COND(env->fpu) & (0x3 << PARAM1));
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
-void op_bc1any4t (void)
-{
- T0 = !!(GET_FP_COND(env->fpu) & (0xf << PARAM1));
- DEBUG_FPU_STATE();
- FORCE_RET();
-}
diff --git a/target-mips/op_helper.c b/target-mips/op_helper.c
index c5a4c93698..696295b904 100644
--- a/target-mips/op_helper.c
+++ b/target-mips/op_helper.c
@@ -1703,8 +1703,26 @@ static always_inline void update_fcr31(void)
UPDATE_FP_FLAGS(env->fpu->fcr31, tmp);
}
+/* Float support.
+ Single precition routines have a "s" suffix, double precision a
+ "d" suffix, 32bit integer "w", 64bit integer "l", paired single "ps",
+ paired single lower "pl", paired single upper "pu". */
+
#define FLOAT_OP(name, p) void do_float_##name##_##p(void)
+/* unary operations, modifying fp status */
+#define FLOAT_UNOP(name) \
+FLOAT_OP(name, d) \
+{ \
+ FDT2 = float64_ ## name(FDT0, &env->fpu->fp_status); \
+} \
+FLOAT_OP(name, s) \
+{ \
+ FST2 = float32_ ## name(FST0, &env->fpu->fp_status); \
+}
+FLOAT_UNOP(sqrt)
+#undef FLOAT_UNOP
+
FLOAT_OP(cvtd, s)
{
set_float_exception_flags(0, &env->fpu->fp_status);
@@ -1943,6 +1961,25 @@ FLOAT_OP(floorw, s)
WT2 = FLOAT_SNAN32;
}
+/* unary operations, not modifying fp status */
+#define FLOAT_UNOP(name) \
+FLOAT_OP(name, d) \
+{ \
+ FDT2 = float64_ ## name(FDT0); \
+} \
+FLOAT_OP(name, s) \
+{ \
+ FST2 = float32_ ## name(FST0); \
+} \
+FLOAT_OP(name, ps) \
+{ \
+ FST2 = float32_ ## name(FST0); \
+ FSTH2 = float32_ ## name(FSTH0); \
+}
+FLOAT_UNOP(abs)
+FLOAT_UNOP(chs)
+#undef FLOAT_UNOP
+
/* MIPS specific unary operations */
FLOAT_OP(recip, d)
{
@@ -2051,6 +2088,56 @@ FLOAT_BINOP(mul)
FLOAT_BINOP(div)
#undef FLOAT_BINOP
+/* ternary operations */
+#define FLOAT_TERNOP(name1, name2) \
+FLOAT_OP(name1 ## name2, d) \
+{ \
+ FDT0 = float64_ ## name1 (FDT0, FDT1, &env->fpu->fp_status); \
+ FDT2 = float64_ ## name2 (FDT0, FDT2, &env->fpu->fp_status); \
+} \
+FLOAT_OP(name1 ## name2, s) \
+{ \
+ FST0 = float32_ ## name1 (FST0, FST1, &env->fpu->fp_status); \
+ FST2 = float32_ ## name2 (FST0, FST2, &env->fpu->fp_status); \
+} \
+FLOAT_OP(name1 ## name2, ps) \
+{ \
+ FST0 = float32_ ## name1 (FST0, FST1, &env->fpu->fp_status); \
+ FSTH0 = float32_ ## name1 (FSTH0, FSTH1, &env->fpu->fp_status); \
+ FST2 = float32_ ## name2 (FST0, FST2, &env->fpu->fp_status); \
+ FSTH2 = float32_ ## name2 (FSTH0, FSTH2, &env->fpu->fp_status); \
+}
+FLOAT_TERNOP(mul, add)
+FLOAT_TERNOP(mul, sub)
+#undef FLOAT_TERNOP
+
+/* negated ternary operations */
+#define FLOAT_NTERNOP(name1, name2) \
+FLOAT_OP(n ## name1 ## name2, d) \
+{ \
+ FDT0 = float64_ ## name1 (FDT0, FDT1, &env->fpu->fp_status); \
+ FDT2 = float64_ ## name2 (FDT0, FDT2, &env->fpu->fp_status); \
+ FDT2 = float64_chs(FDT2); \
+} \
+FLOAT_OP(n ## name1 ## name2, s) \
+{ \
+ FST0 = float32_ ## name1 (FST0, FST1, &env->fpu->fp_status); \
+ FST2 = float32_ ## name2 (FST0, FST2, &env->fpu->fp_status); \
+ FST2 = float32_chs(FST2); \
+} \
+FLOAT_OP(n ## name1 ## name2, ps) \
+{ \
+ FST0 = float32_ ## name1 (FST0, FST1, &env->fpu->fp_status); \
+ FSTH0 = float32_ ## name1 (FSTH0, FSTH1, &env->fpu->fp_status); \
+ FST2 = float32_ ## name2 (FST0, FST2, &env->fpu->fp_status); \
+ FSTH2 = float32_ ## name2 (FSTH0, FSTH2, &env->fpu->fp_status); \
+ FST2 = float32_chs(FST2); \
+ FSTH2 = float32_chs(FSTH2); \
+}
+FLOAT_NTERNOP(mul, add)
+FLOAT_NTERNOP(mul, sub)
+#undef FLOAT_NTERNOP
+
/* MIPS specific binary operations */
FLOAT_OP(recip2, d)
{
diff --git a/target-mips/translate.c b/target-mips/translate.c
index b7e3967ceb..56984147e6 100644
--- a/target-mips/translate.c
+++ b/target-mips/translate.c
@@ -630,6 +630,21 @@ static inline void gen_store_fpr32h (TCGv t, int reg)
tcg_gen_st_i32(t, current_fpu, 8 * reg + 4 * !FP_ENDIAN_IDX);
}
+static inline void get_fp_cond (TCGv t)
+{
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+ TCGv r_tmp2 = tcg_temp_new(TCG_TYPE_I32);
+
+ tcg_gen_ld_i32(r_tmp1, current_fpu, offsetof(CPUMIPSFPUContext, fcr31));
+ tcg_gen_shri_i32(r_tmp2, r_tmp1, 24);
+ tcg_gen_andi_i32(r_tmp2, r_tmp2, 0xfe);
+ tcg_gen_shri_i32(r_tmp1, r_tmp1, 23);
+ tcg_gen_andi_i32(r_tmp1, r_tmp1, 0x1);
+ tcg_gen_or_i32(t, r_tmp1, r_tmp2);
+ tcg_temp_free(r_tmp1);
+ tcg_temp_free(r_tmp2);
+}
+
#define FOP_CONDS(type, fmt) \
static GenOpFunc1 * fcmp ## type ## _ ## fmt ## _table[16] = { \
do_cmp ## type ## _ ## fmt ## _f, \
@@ -5541,38 +5556,170 @@ static void gen_compute_branch1 (CPUState *env, DisasContext *ctx, uint32_t op,
switch (op) {
case OPC_BC1F:
- gen_op_bc1f(cc);
+ {
+ int l1 = gen_new_label();
+ int l2 = gen_new_label();
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+
+ get_fp_cond(r_tmp1);
+ tcg_gen_ext_i32_tl(cpu_T[0], r_tmp1);
+ tcg_temp_free(r_tmp1);
+ tcg_gen_not_tl(cpu_T[0], cpu_T[0]);
+ tcg_gen_movi_tl(cpu_T[1], 0x1 << cc);
+ tcg_gen_and_tl(cpu_T[0], cpu_T[0], cpu_T[1]);
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_movi_tl(cpu_T[0], 0);
+ tcg_gen_br(l2);
+ gen_set_label(l1);
+ tcg_gen_movi_tl(cpu_T[0], 1);
+ gen_set_label(l2);
+ }
opn = "bc1f";
goto not_likely;
case OPC_BC1FL:
- gen_op_bc1f(cc);
+ {
+ int l1 = gen_new_label();
+ int l2 = gen_new_label();
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+
+ get_fp_cond(r_tmp1);
+ tcg_gen_ext_i32_tl(cpu_T[0], r_tmp1);
+ tcg_temp_free(r_tmp1);
+ tcg_gen_not_tl(cpu_T[0], cpu_T[0]);
+ tcg_gen_movi_tl(cpu_T[1], 0x1 << cc);
+ tcg_gen_and_tl(cpu_T[0], cpu_T[0], cpu_T[1]);
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_movi_tl(cpu_T[0], 0);
+ tcg_gen_br(l2);
+ gen_set_label(l1);
+ tcg_gen_movi_tl(cpu_T[0], 1);
+ gen_set_label(l2);
+ }
opn = "bc1fl";
goto likely;
case OPC_BC1T:
- gen_op_bc1t(cc);
+ {
+ int l1 = gen_new_label();
+ int l2 = gen_new_label();
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+
+ get_fp_cond(r_tmp1);
+ tcg_gen_ext_i32_tl(cpu_T[0], r_tmp1);
+ tcg_temp_free(r_tmp1);
+ tcg_gen_movi_tl(cpu_T[1], 0x1 << cc);
+ tcg_gen_and_tl(cpu_T[0], cpu_T[0], cpu_T[1]);
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_movi_tl(cpu_T[0], 0);
+ tcg_gen_br(l2);
+ gen_set_label(l1);
+ tcg_gen_movi_tl(cpu_T[0], 1);
+ gen_set_label(l2);
+ }
opn = "bc1t";
goto not_likely;
case OPC_BC1TL:
- gen_op_bc1t(cc);
+ {
+ int l1 = gen_new_label();
+ int l2 = gen_new_label();
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+
+ get_fp_cond(r_tmp1);
+ tcg_gen_ext_i32_tl(cpu_T[0], r_tmp1);
+ tcg_temp_free(r_tmp1);
+ tcg_gen_movi_tl(cpu_T[1], 0x1 << cc);
+ tcg_gen_and_tl(cpu_T[0], cpu_T[0], cpu_T[1]);
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_movi_tl(cpu_T[0], 0);
+ tcg_gen_br(l2);
+ gen_set_label(l1);
+ tcg_gen_movi_tl(cpu_T[0], 1);
+ gen_set_label(l2);
+ }
opn = "bc1tl";
likely:
ctx->hflags |= MIPS_HFLAG_BL;
tcg_gen_st_tl(cpu_T[0], cpu_env, offsetof(CPUState, bcond));
break;
case OPC_BC1FANY2:
- gen_op_bc1any2f(cc);
+ {
+ int l1 = gen_new_label();
+ int l2 = gen_new_label();
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+
+ get_fp_cond(r_tmp1);
+ tcg_gen_ext_i32_tl(cpu_T[0], r_tmp1);
+ tcg_temp_free(r_tmp1);
+ tcg_gen_not_tl(cpu_T[0], cpu_T[0]);
+ tcg_gen_movi_tl(cpu_T[1], 0x3 << cc);
+ tcg_gen_and_tl(cpu_T[0], cpu_T[0], cpu_T[1]);
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_movi_tl(cpu_T[0], 0);
+ tcg_gen_br(l2);
+ gen_set_label(l1);
+ tcg_gen_movi_tl(cpu_T[0], 1);
+ gen_set_label(l2);
+ }
opn = "bc1any2f";
goto not_likely;
case OPC_BC1TANY2:
- gen_op_bc1any2t(cc);
+ {
+ int l1 = gen_new_label();
+ int l2 = gen_new_label();
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+
+ get_fp_cond(r_tmp1);
+ tcg_gen_ext_i32_tl(cpu_T[0], r_tmp1);
+ tcg_temp_free(r_tmp1);
+ tcg_gen_movi_tl(cpu_T[1], 0x3 << cc);
+ tcg_gen_and_tl(cpu_T[0], cpu_T[0], cpu_T[1]);
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_movi_tl(cpu_T[0], 0);
+ tcg_gen_br(l2);
+ gen_set_label(l1);
+ tcg_gen_movi_tl(cpu_T[0], 1);
+ gen_set_label(l2);
+ }
opn = "bc1any2t";
goto not_likely;
case OPC_BC1FANY4:
- gen_op_bc1any4f(cc);
+ {
+ int l1 = gen_new_label();
+ int l2 = gen_new_label();
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+
+ get_fp_cond(r_tmp1);
+ tcg_gen_ext_i32_tl(cpu_T[0], r_tmp1);
+ tcg_temp_free(r_tmp1);
+ tcg_gen_not_tl(cpu_T[0], cpu_T[0]);
+ tcg_gen_movi_tl(cpu_T[1], 0xf << cc);
+ tcg_gen_and_tl(cpu_T[0], cpu_T[0], cpu_T[1]);
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_movi_tl(cpu_T[0], 0);
+ tcg_gen_br(l2);
+ gen_set_label(l1);
+ tcg_gen_movi_tl(cpu_T[0], 1);
+ gen_set_label(l2);
+ }
opn = "bc1any4f";
goto not_likely;
case OPC_BC1TANY4:
- gen_op_bc1any4t(cc);
+ {
+ int l1 = gen_new_label();
+ int l2 = gen_new_label();
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+
+ get_fp_cond(r_tmp1);
+ tcg_gen_ext_i32_tl(cpu_T[0], r_tmp1);
+ tcg_temp_free(r_tmp1);
+ tcg_gen_movi_tl(cpu_T[1], 0xf << cc);
+ tcg_gen_and_tl(cpu_T[0], cpu_T[0], cpu_T[1]);
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_movi_tl(cpu_T[0], 0);
+ tcg_gen_br(l2);
+ gen_set_label(l1);
+ tcg_gen_movi_tl(cpu_T[0], 1);
+ gen_set_label(l2);
+ }
opn = "bc1any4t";
not_likely:
ctx->hflags |= MIPS_HFLAG_BC;
@@ -5685,23 +5832,83 @@ static void gen_movci (DisasContext *ctx, int rd, int rs, int cc, int tf)
gen_store_gpr(cpu_T[0], rd);
}
-#define GEN_MOVCF(fmt) \
-static void glue(gen_movcf_, fmt) (DisasContext *ctx, int cc, int tf) \
-{ \
- uint32_t ccbit; \
- \
- if (cc) { \
- ccbit = 1 << (24 + cc); \
- } else \
- ccbit = 1 << 23; \
- if (!tf) \
- glue(gen_op_float_movf_, fmt)(ccbit); \
- else \
- glue(gen_op_float_movt_, fmt)(ccbit); \
+static inline void gen_movcf_s (int cc, int tf)
+{
+ uint32_t ccbit;
+ int cond;
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+ int l1 = gen_new_label();
+
+ if (cc)
+ ccbit = 1 << (24 + cc);
+ else
+ ccbit = 1 << 23;
+
+ if (tf)
+ cond = TCG_COND_EQ;
+ else
+ cond = TCG_COND_NE;
+
+ tcg_gen_ld_i32(r_tmp1, current_fpu, offsetof(CPUMIPSFPUContext, fcr31));
+ tcg_gen_andi_i32(r_tmp1, r_tmp1, ccbit);
+ tcg_gen_brcondi_i32(cond, r_tmp1, 0, l1);
+ tcg_gen_movi_i32(fpu32_T[2], fpu32_T[0]);
+ gen_set_label(l1);
+ tcg_temp_free(r_tmp1);
}
-GEN_MOVCF(d);
-GEN_MOVCF(s);
-#undef GEN_MOVCF
+
+static inline void gen_movcf_d (int cc, int tf)
+{
+ uint32_t ccbit;
+ int cond;
+ TCGv r_tmp1 = tcg_temp_new(TCG_TYPE_I32);
+ int l1 = gen_new_label();
+
+ if (cc)
+ ccbit = 1 << (24 + cc);
+ else
+ ccbit = 1 << 23;
+
+ if (tf)
+ cond = TCG_COND_EQ;
+ else
+ cond = TCG_COND_NE;
+
+ tcg_gen_ld_i32(r_tmp1, current_fpu, offsetof(CPUMIPSFPUContext, fcr31));
+ tcg_gen_andi_i32(r_tmp1, r_tmp1, ccbit);
+ tcg_gen_brcondi_i32(cond, r_tmp1, 0, l1);
+ tcg_gen_movi_i64(fpu64_T[2], fpu64_T[0]);
+ gen_set_label(l1);
+ tcg_temp_free(r_tmp1);
+}
+
+static inline void gen_movcf_ps (int cc, int tf)
+{
+ int cond;
+ TCGv r_tmp1 = tcg_temp_local_new(TCG_TYPE_I32);
+ TCGv r_tmp2 = tcg_temp_local_new(TCG_TYPE_I32);
+ int l1 = gen_new_label();
+ int l2 = gen_new_label();
+
+ if (tf)
+ cond = TCG_COND_EQ;
+ else
+ cond = TCG_COND_NE;
+
+ get_fp_cond(r_tmp1);
+ tcg_gen_shri_i32(r_tmp1, r_tmp1, cc);
+ tcg_gen_andi_i32(r_tmp2, r_tmp1, 0x1);
+ tcg_gen_brcondi_i32(cond, r_tmp2, 0, l1);
+ tcg_gen_movi_i32(fpu32_T[2], fpu32_T[0]);
+ gen_set_label(l1);
+ tcg_gen_andi_i32(r_tmp2, r_tmp1, 0x2);
+ tcg_gen_brcondi_i32(cond, r_tmp2, 0, l2);
+ tcg_gen_movi_i32(fpu32h_T[2], fpu32h_T[0]);
+ gen_set_label(l2);
+ tcg_temp_free(r_tmp1);
+ tcg_temp_free(r_tmp2);
+}
+
static void gen_farith (DisasContext *ctx, uint32_t op1,
int ft, int fs, int fd, int cc)
@@ -5781,13 +5988,13 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
break;
case FOP(4, 16):
gen_load_fpr32(fpu32_T[0], fs);
- gen_op_float_sqrt_s();
+ tcg_gen_helper_0_0(do_float_sqrt_s);
gen_store_fpr32(fpu32_T[2], fd);
opn = "sqrt.s";
break;
case FOP(5, 16):
gen_load_fpr32(fpu32_T[0], fs);
- gen_op_float_abs_s();
+ tcg_gen_helper_0_0(do_float_abs_s);
gen_store_fpr32(fpu32_T[2], fd);
opn = "abs.s";
break;
@@ -5798,7 +6005,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
break;
case FOP(7, 16):
gen_load_fpr32(fpu32_T[0], fs);
- gen_op_float_chs_s();
+ tcg_gen_helper_0_0(do_float_chs_s);
gen_store_fpr32(fpu32_T[2], fd);
opn = "neg.s";
break;
@@ -5855,10 +6062,9 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
opn = "floor.w.s";
break;
case FOP(17, 16):
- gen_load_gpr(cpu_T[0], ft);
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32(fpu32_T[2], fd);
- gen_movcf_s(ctx, (ft >> 2) & 0x7, ft & 0x1);
+ gen_movcf_s((ft >> 2) & 0x7, ft & 0x1);
gen_store_fpr32(fpu32_T[2], fd);
opn = "movcf.s";
break;
@@ -5866,7 +6072,13 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
gen_load_gpr(cpu_T[0], ft);
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32(fpu32_T[2], fd);
- gen_op_float_movz_s();
+ {
+ int l1 = gen_new_label();
+
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_mov_i32(fpu32_T[2], fpu32_T[0]);
+ gen_set_label(l1);
+ }
gen_store_fpr32(fpu32_T[2], fd);
opn = "movz.s";
break;
@@ -5874,7 +6086,13 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
gen_load_gpr(cpu_T[0], ft);
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32(fpu32_T[2], fd);
- gen_op_float_movn_s();
+ {
+ int l1 = gen_new_label();
+
+ tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_T[0], 0, l1);
+ tcg_gen_mov_i32(fpu32_T[2], fpu32_T[0]);
+ gen_set_label(l1);
+ }
gen_store_fpr32(fpu32_T[2], fd);
opn = "movn.s";
break;
@@ -6019,14 +6237,14 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
case FOP(4, 17):
check_cp1_registers(ctx, fs | fd);
gen_load_fpr64(ctx, fpu64_T[0], fs);
- gen_op_float_sqrt_d();
+ tcg_gen_helper_0_0(do_float_sqrt_d);
gen_store_fpr64(ctx, fpu64_T[2], fd);
opn = "sqrt.d";
break;
case FOP(5, 17):
check_cp1_registers(ctx, fs | fd);
gen_load_fpr64(ctx, fpu64_T[0], fs);
- gen_op_float_abs_d();
+ tcg_gen_helper_0_0(do_float_abs_d);
gen_store_fpr64(ctx, fpu64_T[2], fd);
opn = "abs.d";
break;
@@ -6039,7 +6257,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
case FOP(7, 17):
check_cp1_registers(ctx, fs | fd);
gen_load_fpr64(ctx, fpu64_T[0], fs);
- gen_op_float_chs_d();
+ tcg_gen_helper_0_0(do_float_chs_d);
gen_store_fpr64(ctx, fpu64_T[2], fd);
opn = "neg.d";
break;
@@ -6100,10 +6318,9 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
opn = "floor.w.d";
break;
case FOP(17, 17):
- gen_load_gpr(cpu_T[0], ft);
gen_load_fpr64(ctx, fpu64_T[0], fs);
gen_load_fpr64(ctx, fpu64_T[2], fd);
- gen_movcf_d(ctx, (ft >> 2) & 0x7, ft & 0x1);
+ gen_movcf_d((ft >> 2) & 0x7, ft & 0x1);
gen_store_fpr64(ctx, fpu64_T[2], fd);
opn = "movcf.d";
break;
@@ -6111,7 +6328,13 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
gen_load_gpr(cpu_T[0], ft);
gen_load_fpr64(ctx, fpu64_T[0], fs);
gen_load_fpr64(ctx, fpu64_T[2], fd);
- gen_op_float_movz_d();
+ {
+ int l1 = gen_new_label();
+
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_mov_i64(fpu64_T[2], fpu64_T[0]);
+ gen_set_label(l1);
+ }
gen_store_fpr64(ctx, fpu64_T[2], fd);
opn = "movz.d";
break;
@@ -6119,7 +6342,13 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
gen_load_gpr(cpu_T[0], ft);
gen_load_fpr64(ctx, fpu64_T[0], fs);
gen_load_fpr64(ctx, fpu64_T[2], fd);
- gen_op_float_movn_d();
+ {
+ int l1 = gen_new_label();
+
+ tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_T[0], 0, l1);
+ tcg_gen_mov_i64(fpu64_T[2], fpu64_T[0]);
+ gen_set_label(l1);
+ }
gen_store_fpr64(ctx, fpu64_T[2], fd);
opn = "movn.d";
break;
@@ -6290,7 +6519,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
check_cp1_64bitmode(ctx);
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32h(fpu32h_T[0], fs);
- gen_op_float_abs_ps();
+ tcg_gen_helper_0_0(do_float_abs_ps);
gen_store_fpr32(fpu32_T[2], fd);
gen_store_fpr32h(fpu32h_T[2], fd);
opn = "abs.ps";
@@ -6307,22 +6536,18 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
check_cp1_64bitmode(ctx);
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32h(fpu32h_T[0], fs);
- gen_op_float_chs_ps();
+ tcg_gen_helper_0_0(do_float_chs_ps);
gen_store_fpr32(fpu32_T[2], fd);
gen_store_fpr32h(fpu32h_T[2], fd);
opn = "neg.ps";
break;
case FOP(17, 22):
check_cp1_64bitmode(ctx);
- gen_load_gpr(cpu_T[0], ft);
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32h(fpu32h_T[0], fs);
gen_load_fpr32(fpu32_T[2], fd);
gen_load_fpr32h(fpu32h_T[2], fd);
- if (ft & 0x1)
- gen_op_float_movt_ps ((ft >> 2) & 0x7);
- else
- gen_op_float_movf_ps ((ft >> 2) & 0x7);
+ gen_movcf_ps((ft >> 2) & 0x7, ft & 0x1);
gen_store_fpr32(fpu32_T[2], fd);
gen_store_fpr32h(fpu32h_T[2], fd);
opn = "movcf.ps";
@@ -6334,7 +6559,14 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
gen_load_fpr32h(fpu32h_T[0], fs);
gen_load_fpr32(fpu32_T[2], fd);
gen_load_fpr32h(fpu32h_T[2], fd);
- gen_op_float_movz_ps();
+ {
+ int l1 = gen_new_label();
+
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_mov_i32(fpu32_T[2], fpu32_T[0]);
+ tcg_gen_mov_i32(fpu32h_T[2], fpu32h_T[0]);
+ gen_set_label(l1);
+ }
gen_store_fpr32(fpu32_T[2], fd);
gen_store_fpr32h(fpu32h_T[2], fd);
opn = "movz.ps";
@@ -6346,7 +6578,14 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
gen_load_fpr32h(fpu32h_T[0], fs);
gen_load_fpr32(fpu32_T[2], fd);
gen_load_fpr32h(fpu32h_T[2], fd);
- gen_op_float_movn_ps();
+ {
+ int l1 = gen_new_label();
+
+ tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_T[0], 0, l1);
+ tcg_gen_mov_i32(fpu32_T[2], fpu32_T[0]);
+ tcg_gen_mov_i32(fpu32h_T[2], fpu32h_T[0]);
+ gen_set_label(l1);
+ }
gen_store_fpr32(fpu32_T[2], fd);
gen_store_fpr32h(fpu32h_T[2], fd);
opn = "movn.ps";
@@ -6440,32 +6679,32 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
check_cp1_64bitmode(ctx);
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32(fpu32_T[1], ft);
- gen_op_float_pll_ps();
- gen_store_fpr64(ctx, fpu64_T[2], fd);
+ gen_store_fpr32h(fpu32_T[0], fd);
+ gen_store_fpr32(fpu32_T[1], fd);
opn = "pll.ps";
break;
case FOP(45, 22):
check_cp1_64bitmode(ctx);
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32h(fpu32h_T[1], ft);
- gen_op_float_plu_ps();
- gen_store_fpr64(ctx, fpu64_T[2], fd);
+ gen_store_fpr32(fpu32h_T[1], fd);
+ gen_store_fpr32h(fpu32_T[0], fd);
opn = "plu.ps";
break;
case FOP(46, 22):
check_cp1_64bitmode(ctx);
gen_load_fpr32h(fpu32h_T[0], fs);
gen_load_fpr32(fpu32_T[1], ft);
- gen_op_float_pul_ps();
- gen_store_fpr64(ctx, fpu64_T[2], fd);
+ gen_store_fpr32(fpu32_T[1], fd);
+ gen_store_fpr32h(fpu32h_T[0], fd);
opn = "pul.ps";
break;
case FOP(47, 22):
check_cp1_64bitmode(ctx);
gen_load_fpr32h(fpu32h_T[0], fs);
gen_load_fpr32h(fpu32h_T[1], ft);
- gen_op_float_puu_ps();
- gen_store_fpr64(ctx, fpu64_T[2], fd);
+ gen_store_fpr32(fpu32h_T[1], fd);
+ gen_store_fpr32h(fpu32h_T[0], fd);
opn = "puu.ps";
break;
case FOP(48, 22):
@@ -6595,10 +6834,32 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
case OPC_ALNV_PS:
check_cp1_64bitmode(ctx);
gen_load_gpr(cpu_T[0], fr);
- gen_load_fpr64(ctx, fpu64_T[0], fs);
- gen_load_fpr64(ctx, fpu64_T[1], ft);
- gen_op_float_alnv_ps();
- gen_store_fpr64(ctx, fpu64_T[2], fd);
+ tcg_gen_andi_tl(cpu_T[0], cpu_T[0], 0x7);
+ gen_load_fpr32(fpu32_T[0], fs);
+ gen_load_fpr32h(fpu32h_T[0], fs);
+ gen_load_fpr32(fpu32_T[1], ft);
+ gen_load_fpr32h(fpu32h_T[1], ft);
+ {
+ int l1 = gen_new_label();
+ int l2 = gen_new_label();
+
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 0, l1);
+ tcg_gen_mov_i32(fpu32_T[2], fpu32_T[0]);
+ tcg_gen_mov_i32(fpu32h_T[2], fpu32h_T[0]);
+ tcg_gen_br(l2);
+ gen_set_label(l1);
+ tcg_gen_brcondi_tl(TCG_COND_NE, cpu_T[0], 4, l2);
+#ifdef TARGET_WORDS_BIGENDIAN
+ tcg_gen_mov_i32(fpu32h_T[2], fpu32_T[0]);
+ tcg_gen_mov_i32(fpu32_T[2], fpu32h_T[1]);
+#else
+ tcg_gen_mov_i32(fpu32h_T[2], fpu32_T[1]);
+ tcg_gen_mov_i32(fpu32_T[2], fpu32h_T[0]);
+#endif
+ gen_set_label(l2);
+ }
+ gen_store_fpr32(fpu32_T[2], fd);
+ gen_store_fpr32h(fpu32h_T[2], fd);
opn = "alnv.ps";
break;
case OPC_MADD_S:
@@ -6606,7 +6867,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32(fpu32_T[1], ft);
gen_load_fpr32(fpu32_T[2], fr);
- gen_op_float_muladd_s();
+ tcg_gen_helper_0_0(do_float_muladd_s);
gen_store_fpr32(fpu32_T[2], fd);
opn = "madd.s";
break;
@@ -6616,7 +6877,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr64(ctx, fpu64_T[0], fs);
gen_load_fpr64(ctx, fpu64_T[1], ft);
gen_load_fpr64(ctx, fpu64_T[2], fr);
- gen_op_float_muladd_d();
+ tcg_gen_helper_0_0(do_float_muladd_d);
gen_store_fpr64(ctx, fpu64_T[2], fd);
opn = "madd.d";
break;
@@ -6628,7 +6889,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr32h(fpu32h_T[1], ft);
gen_load_fpr32(fpu32_T[2], fr);
gen_load_fpr32h(fpu32h_T[2], fr);
- gen_op_float_muladd_ps();
+ tcg_gen_helper_0_0(do_float_muladd_ps);
gen_store_fpr32(fpu32_T[2], fd);
gen_store_fpr32h(fpu32h_T[2], fd);
opn = "madd.ps";
@@ -6638,7 +6899,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32(fpu32_T[1], ft);
gen_load_fpr32(fpu32_T[2], fr);
- gen_op_float_mulsub_s();
+ tcg_gen_helper_0_0(do_float_mulsub_s);
gen_store_fpr32(fpu32_T[2], fd);
opn = "msub.s";
break;
@@ -6648,7 +6909,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr64(ctx, fpu64_T[0], fs);
gen_load_fpr64(ctx, fpu64_T[1], ft);
gen_load_fpr64(ctx, fpu64_T[2], fr);
- gen_op_float_mulsub_d();
+ tcg_gen_helper_0_0(do_float_mulsub_d);
gen_store_fpr64(ctx, fpu64_T[2], fd);
opn = "msub.d";
break;
@@ -6660,7 +6921,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr32h(fpu32h_T[1], ft);
gen_load_fpr32(fpu32_T[2], fr);
gen_load_fpr32h(fpu32h_T[2], fr);
- gen_op_float_mulsub_ps();
+ tcg_gen_helper_0_0(do_float_mulsub_ps);
gen_store_fpr32(fpu32_T[2], fd);
gen_store_fpr32h(fpu32h_T[2], fd);
opn = "msub.ps";
@@ -6670,7 +6931,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32(fpu32_T[1], ft);
gen_load_fpr32(fpu32_T[2], fr);
- gen_op_float_nmuladd_s();
+ tcg_gen_helper_0_0(do_float_nmuladd_s);
gen_store_fpr32(fpu32_T[2], fd);
opn = "nmadd.s";
break;
@@ -6680,7 +6941,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr64(ctx, fpu64_T[0], fs);
gen_load_fpr64(ctx, fpu64_T[1], ft);
gen_load_fpr64(ctx, fpu64_T[2], fr);
- gen_op_float_nmuladd_d();
+ tcg_gen_helper_0_0(do_float_nmuladd_d);
gen_store_fpr64(ctx, fpu64_T[2], fd);
opn = "nmadd.d";
break;
@@ -6692,7 +6953,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr32h(fpu32h_T[1], ft);
gen_load_fpr32(fpu32_T[2], fr);
gen_load_fpr32h(fpu32h_T[2], fr);
- gen_op_float_nmuladd_ps();
+ tcg_gen_helper_0_0(do_float_nmuladd_ps);
gen_store_fpr32(fpu32_T[2], fd);
gen_store_fpr32h(fpu32h_T[2], fd);
opn = "nmadd.ps";
@@ -6702,7 +6963,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr32(fpu32_T[0], fs);
gen_load_fpr32(fpu32_T[1], ft);
gen_load_fpr32(fpu32_T[2], fr);
- gen_op_float_nmulsub_s();
+ tcg_gen_helper_0_0(do_float_nmulsub_s);
gen_store_fpr32(fpu32_T[2], fd);
opn = "nmsub.s";
break;
@@ -6712,7 +6973,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr64(ctx, fpu64_T[0], fs);
gen_load_fpr64(ctx, fpu64_T[1], ft);
gen_load_fpr64(ctx, fpu64_T[2], fr);
- gen_op_float_nmulsub_d();
+ tcg_gen_helper_0_0(do_float_nmulsub_d);
gen_store_fpr64(ctx, fpu64_T[2], fd);
opn = "nmsub.d";
break;
@@ -6724,7 +6985,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc,
gen_load_fpr32h(fpu32h_T[1], ft);
gen_load_fpr32(fpu32_T[2], fr);
gen_load_fpr32h(fpu32h_T[2], fr);
- gen_op_float_nmulsub_ps();
+ tcg_gen_helper_0_0(do_float_nmulsub_ps);
gen_store_fpr32(fpu32_T[2], fd);
gen_store_fpr32h(fpu32h_T[2], fd);
opn = "nmsub.ps";