PowerPC bugfixes:

- must clear carry bit when doing addic with a zero immediate value
- fix missing RETURN in micro-operation that would lead to random failures
  and crashes
- add USE_PRECISE_EMULATION compilation-time option to choose between
  getting exact floating point results and fast but less accurate computation.


git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@2526 c046a42c-6fe2-441c-8c8c-71466251a162

5 files changed, 110 insertions, 4 deletions

diff --git a/target-ppc/exec.h b/target-ppc/exec.h
index a10b62d25b..87e8c1836c 100644
--- a/target-ppc/exec.h
+++ b/target-ppc/exec.h
@@ -27,6 +27,10 @@
 #include "cpu.h"
 #include "exec-all.h"
 
+/* For normal operations, precise emulation should not be needed */
+//#define USE_PRECISE_EMULATION 1
+#define USE_PRECISE_EMULATION 0
+
 register struct CPUPPCState *env asm(AREG0);
 #if TARGET_LONG_BITS > HOST_LONG_BITS
 /* no registers can be used */
diff --git a/target-ppc/op.c b/target-ppc/op.c
index f13f6f5676..068ccb53b2 100644
--- a/target-ppc/op.c
+++ b/target-ppc/op.c
@@ -261,10 +261,15 @@ PPC_OP(load_xer_cr)
     RETURN();
 }
 
-PPC_OP(clear_xer_cr)
+PPC_OP(clear_xer_ov)
 {
     xer_so = 0;
     xer_ov = 0;
+    RETURN();
+}
+
+PPC_OP(clear_xer_ca)
+{
     xer_ca = 0;
     RETURN();
 }
@@ -714,6 +719,7 @@ void OPPROTO op_check_addo (void)
         xer_so = 1;
         xer_ov = 1;
     }
+    RETURN();
 }
 
 #if defined(TARGET_PPC64)
@@ -726,6 +732,7 @@ void OPPROTO op_check_addo_64 (void)
         xer_so = 1;
         xer_ov = 1;
     }
+    RETURN();
 }
 #endif
 
@@ -1643,16 +1650,24 @@ PPC_OP(fsel)
 /* fmadd - fmadd. */
 PPC_OP(fmadd)
 {
+#if USE_PRECISE_EMULATION
+    do_fmadd();
+#else
     FT0 = float64_mul(FT0, FT1, &env->fp_status);
     FT0 = float64_add(FT0, FT2, &env->fp_status);
+#endif
     RETURN();
 }
 
 /* fmsub - fmsub. */
 PPC_OP(fmsub)
 {
+#if USE_PRECISE_EMULATION
+    do_fmsub();
+#else
     FT0 = float64_mul(FT0, FT1, &env->fp_status);
     FT0 = float64_sub(FT0, FT2, &env->fp_status);
+#endif
     RETURN();
 }
 
@@ -2378,6 +2393,7 @@ void OPPROTO op_store_booke_tsr (void)
 void OPPROTO op_splatw_T1_64 (void)
 {
     T1_64 = (T1_64 << 32) | (T1_64 & 0x00000000FFFFFFFFULL);
+    RETURN();
 }
 
 void OPPROTO op_splatwi_T0_64 (void)
@@ -2385,6 +2401,7 @@ void OPPROTO op_splatwi_T0_64 (void)
     uint64_t tmp = PARAM1;
 
     T0_64 = (tmp << 32) | tmp;
+    RETURN();
 }
 
 void OPPROTO op_splatwi_T1_64 (void)
@@ -2392,6 +2409,7 @@ void OPPROTO op_splatwi_T1_64 (void)
     uint64_t tmp = PARAM1;
 
     T1_64 = (tmp << 32) | tmp;
+    RETURN();
 }
 
 void OPPROTO op_extsh_T1_64 (void)
diff --git a/target-ppc/op_helper.c b/target-ppc/op_helper.c
index 756d164dad..45e4dde9d3 100644
--- a/target-ppc/op_helper.c
+++ b/target-ppc/op_helper.c
@@ -615,11 +615,13 @@ void do_fctiw (void)
         uint64_t i;
     } p;
 
+    p.i = float64_to_int32(FT0, &env->fp_status);
+#if USE_PRECISE_EMULATION
     /* XXX: higher bits are not supposed to be significant.
      *     to make tests easier, return the same as a real PowerPC 750 (aka G3)
      */
-    p.i = float64_to_int32(FT0, &env->fp_status);
     p.i |= 0xFFF80000ULL << 32;
+#endif
     FT0 = p.d;
 }
 
@@ -630,26 +632,96 @@ void do_fctiwz (void)
         uint64_t i;
     } p;
 
+    p.i = float64_to_int32_round_to_zero(FT0, &env->fp_status);
+#if USE_PRECISE_EMULATION
     /* XXX: higher bits are not supposed to be significant.
      *     to make tests easier, return the same as a real PowerPC 750 (aka G3)
      */
-    p.i = float64_to_int32_round_to_zero(FT0, &env->fp_status);
     p.i |= 0xFFF80000ULL << 32;
+#endif
     FT0 = p.d;
 }
 
+#if USE_PRECISE_EMULATION
+void do_fmadd (void)
+{
+#ifdef FLOAT128
+    float128 ft0_128, ft1_128;
+
+    ft0_128 = float64_to_float128(FT0, &env->fp_status);
+    ft1_128 = float64_to_float128(FT1, &env->fp_status);
+    ft0_128 = float128_mul(ft0_128, ft1_128, &env->fp_status);
+    ft1_128 = float64_to_float128(FT2, &env->fp_status);
+    ft0_128 = float128_add(ft0_128, ft1_128, &env->fp_status);
+    FT0 = float128_to_float64(ft0_128, &env->fp_status);
+#else
+    /* This is OK on x86 hosts */
+    FT0 = (FT0 * FT1) + FT2;
+#endif
+}
+
+void do_fmsub (void)
+{
+#ifdef FLOAT128
+    float128 ft0_128, ft1_128;
+
+    ft0_128 = float64_to_float128(FT0, &env->fp_status);
+    ft1_128 = float64_to_float128(FT1, &env->fp_status);
+    ft0_128 = float128_mul(ft0_128, ft1_128, &env->fp_status);
+    ft1_128 = float64_to_float128(FT2, &env->fp_status);
+    ft0_128 = float128_sub(ft0_128, ft1_128, &env->fp_status);
+    FT0 = float128_to_float64(ft0_128, &env->fp_status);
+#else
+    /* This is OK on x86 hosts */
+    FT0 = (FT0 * FT1) - FT2;
+#endif
+}
+#endif /* USE_PRECISE_EMULATION */
+
 void do_fnmadd (void)
 {
+#if USE_PRECISE_EMULATION
+#ifdef FLOAT128
+    float128 ft0_128, ft1_128;
+
+    ft0_128 = float64_to_float128(FT0, &env->fp_status);
+    ft1_128 = float64_to_float128(FT1, &env->fp_status);
+    ft0_128 = float128_mul(ft0_128, ft1_128, &env->fp_status);
+    ft1_128 = float64_to_float128(FT2, &env->fp_status);
+    ft0_128 = float128_add(ft0_128, ft1_128, &env->fp_status);
+    FT0 = float128_to_float64(ft0_128, &env->fp_status);
+#else
+    /* This is OK on x86 hosts */
+    FT0 = (FT0 * FT1) + FT2;
+#endif
+#else
     FT0 = float64_mul(FT0, FT1, &env->fp_status);
     FT0 = float64_add(FT0, FT2, &env->fp_status);
+#endif
     if (likely(!isnan(FT0)))
         FT0 = float64_chs(FT0);
 }
 
 void do_fnmsub (void)
 {
+#if USE_PRECISE_EMULATION
+#ifdef FLOAT128
+    float128 ft0_128, ft1_128;
+
+    ft0_128 = float64_to_float128(FT0, &env->fp_status);
+    ft1_128 = float64_to_float128(FT1, &env->fp_status);
+    ft0_128 = float128_mul(ft0_128, ft1_128, &env->fp_status);
+    ft1_128 = float64_to_float128(FT2, &env->fp_status);
+    ft0_128 = float128_sub(ft0_128, ft1_128, &env->fp_status);
+    FT0 = float128_to_float64(ft0_128, &env->fp_status);
+#else
+    /* This is OK on x86 hosts */
+    FT0 = (FT0 * FT1) - FT2;
+#endif
+#else
     FT0 = float64_mul(FT0, FT1, &env->fp_status);
     FT0 = float64_sub(FT0, FT2, &env->fp_status);
+#endif
     if (likely(!isnan(FT0)))
         FT0 = float64_chs(FT0);
 }
@@ -667,7 +739,12 @@ void do_fres (void)
     } p;
 
     if (likely(isnormal(FT0))) {
+#if USE_PRECISE_EMULATION
+        FT0 = float64_div(1.0, FT0, &env->fp_status);
+        FT0 = float64_to_float32(FT0, &env->fp_status);
+#else
         FT0 = float32_div(1.0, FT0, &env->fp_status);
+#endif
     } else {
         p.d = FT0;
         if (p.i == 0x8000000000000000ULL) {
diff --git a/target-ppc/op_helper.h b/target-ppc/op_helper.h
index 3aa6045838..fc3d8af604 100644
--- a/target-ppc/op_helper.h
+++ b/target-ppc/op_helper.h
@@ -93,6 +93,10 @@ void do_fsqrt (void);
 void do_fres (void);
 void do_frsqrte (void);
 void do_fsel (void);
+#if USE_PRECISE_EMULATION
+void do_fmadd (void);
+void do_fmsub (void);
+#endif
 void do_fnmadd (void);
 void do_fnmsub (void);
 void do_fctiw (void);
diff --git a/target-ppc/translate.c b/target-ppc/translate.c
index 90a9fc93c7..fb9b07591a 100644
--- a/target-ppc/translate.c
+++ b/target-ppc/translate.c
@@ -781,6 +781,8 @@ GEN_HANDLER(addic, 0x0C, 0xFF, 0xFF, 0x00000000, PPC_INTEGER)
         else
 #endif
             gen_op_check_addc();
+    } else {
+        gen_op_clear_xer_ca();
     }
     gen_op_store_T0_gpr(rD(ctx->opcode));
 }
@@ -2804,7 +2806,8 @@ GEN_HANDLER(mcrxr, 0x1F, 0x00, 0x10, 0x007FF801, PPC_MISC)
 {
     gen_op_load_xer_cr();
     gen_op_store_T0_crf(crfD(ctx->opcode));
-    gen_op_clear_xer_cr();
+    gen_op_clear_xer_ov();
+    gen_op_clear_xer_ca();
 }
 
 /* mfcr */