From 8422b1133739343c1b35d0bba30c3209649e43e1 Mon Sep 17 00:00:00 2001 From: bellard Date: Sun, 20 Mar 2005 10:39:24 +0000 Subject: [PATCH] NaN support in FPU comparisons git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@1341 c046a42c-6fe2-441c-8c8c-71466251a162 --- target-i386/exec.h | 6 ++- target-i386/helper.c | 6 --- target-i386/op.c | 94 +++++++++++++++++++++++++++++++++------------- target-i386/ops_sse.h | 102 ++++++++++++++++++++++++++++++++++---------------- 4 files changed, 141 insertions(+), 67 deletions(-) diff --git a/target-i386/exec.h b/target-i386/exec.h index 137774f5a3..e4b6251a29 100644 --- a/target-i386/exec.h +++ b/target-i386/exec.h @@ -325,6 +325,8 @@ static inline void stfl(target_ulong ptr, float v) #define floatx_abs floatx80_abs #define floatx_chs floatx80_chs #define floatx_round_to_int floatx80_round_to_int +#define floatx_compare floatx80_compare +#define floatx_compare_quiet floatx80_compare_quiet #define sin sinl #define cos cosl #define sqrt sqrtl @@ -340,6 +342,8 @@ static inline void stfl(target_ulong ptr, float v) #define floatx_abs float64_abs #define floatx_chs float64_chs #define floatx_round_to_int float64_round_to_int +#define floatx_compare float64_compare +#define floatx_compare_quiet float64_compare_quiet #endif extern CPU86_LDouble sin(CPU86_LDouble x); @@ -547,8 +551,6 @@ void restore_native_fp_state(CPUState *env); void save_native_fp_state(CPUState *env); float approx_rsqrt(float a); float approx_rcp(float a); -double helper_sqrt(double a); -int fpu_isnan(double a); void update_fp_status(void); extern const uint8_t parity_table[256]; diff --git a/target-i386/helper.c b/target-i386/helper.c index 64cb51ab53..a15ec5a6ad 100644 --- a/target-i386/helper.c +++ b/target-i386/helper.c @@ -3223,12 +3223,6 @@ void helper_idivq_EAX_T0(void) #endif -/* XXX: do it */ -int fpu_isnan(double a) -{ - return 0; -} - float approx_rsqrt(float a) { return 1.0 / sqrt(a); diff --git a/target-i386/op.c b/target-i386/op.c index bbc2b19aed..c157b120b0 100644 --- a/target-i386/op.c +++ b/target-i386/op.c @@ -1952,52 +1952,94 @@ void OPPROTO op_fxchg_ST0_STN(void) /* FPU operations */ -/* XXX: handle nans */ void OPPROTO op_fcom_ST0_FT0(void) { - env->fpus &= (~0x4500); /* (C3,C2,C0) <-- 000 */ - if (ST0 < FT0) - env->fpus |= 0x100; /* (C3,C2,C0) <-- 001 */ - else if (ST0 == FT0) - env->fpus |= 0x4000; /* (C3,C2,C0) <-- 100 */ + int cc; + switch(floatx_compare(ST0, FT0, &env->fp_status)) { + case -1: + cc = 0x0100; + break; + case 0: + cc = 0x4000; + break; + case 1: + cc = 0x0000; + break; + case 2: + default: + cc = 0x4500; + break; + } + env->fpus = (env->fpus & ~0x4500) | cc; FORCE_RET(); } -/* XXX: handle nans */ void OPPROTO op_fucom_ST0_FT0(void) { - env->fpus &= (~0x4500); /* (C3,C2,C0) <-- 000 */ - if (ST0 < FT0) - env->fpus |= 0x100; /* (C3,C2,C0) <-- 001 */ - else if (ST0 == FT0) - env->fpus |= 0x4000; /* (C3,C2,C0) <-- 100 */ + int cc; + switch(floatx_compare_quiet(ST0, FT0, &env->fp_status)) { + case -1: + cc = 0x0100; + break; + case 0: + cc = 0x4000; + break; + case 1: + cc = 0x0000; + break; + case 2: + default: + cc = 0x4500; + break; + } + env->fpus = (env->fpus & ~0x4500) | cc; FORCE_RET(); } -/* XXX: handle nans */ void OPPROTO op_fcomi_ST0_FT0(void) { - int eflags; + int eflags, cc; + switch(floatx_compare(ST0, FT0, &env->fp_status)) { + case -1: + cc = CC_C; + break; + case 0: + cc = CC_Z; + break; + case 1: + cc = 0; + break; + case 2: + default: + cc = CC_Z | CC_P | CC_C; + break; + } eflags = cc_table[CC_OP].compute_all(); - eflags &= ~(CC_Z | CC_P | CC_C); - if (ST0 < FT0) - eflags |= CC_C; - else if (ST0 == FT0) - eflags |= CC_Z; + eflags = (eflags & ~(CC_Z | CC_P | CC_C)) | cc; CC_SRC = eflags; FORCE_RET(); } -/* XXX: handle nans */ void OPPROTO op_fucomi_ST0_FT0(void) { - int eflags; + int eflags, cc; + switch(floatx_compare_quiet(ST0, FT0, &env->fp_status)) { + case -1: + cc = CC_C; + break; + case 0: + cc = CC_Z; + break; + case 1: + cc = 0; + break; + case 2: + default: + cc = CC_Z | CC_P | CC_C; + break; + } eflags = cc_table[CC_OP].compute_all(); - eflags &= ~(CC_Z | CC_P | CC_C); - if (ST0 < FT0) - eflags |= CC_C; - else if (ST0 == FT0) - eflags |= CC_Z; + eflags = (eflags & ~(CC_Z | CC_P | CC_C)) | cc; CC_SRC = eflags; FORCE_RET(); } diff --git a/target-i386/ops_sse.h b/target-i386/ops_sse.h index be1d68bfb7..f982594a8d 100644 --- a/target-i386/ops_sse.h +++ b/target-i386/ops_sse.h @@ -704,7 +704,7 @@ SSE_OP_S(sqrt, FPU_SQRT) /* float to float conversions */ void OPPROTO op_cvtps2pd(void) { - float s0, s1; + float32 s0, s1; Reg *d, *s; d = (Reg *)((char *)env + PARAM1); s = (Reg *)((char *)env + PARAM2); @@ -1031,10 +1031,10 @@ void OPPROTO op_ ## name ## ps (void)\ Reg *d, *s;\ d = (Reg *)((char *)env + PARAM1);\ s = (Reg *)((char *)env + PARAM2);\ - d->XMM_L(0) = F(d->XMM_S(0), s->XMM_S(0));\ - d->XMM_L(1) = F(d->XMM_S(1), s->XMM_S(1));\ - d->XMM_L(2) = F(d->XMM_S(2), s->XMM_S(2));\ - d->XMM_L(3) = F(d->XMM_S(3), s->XMM_S(3));\ + d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\ + d->XMM_L(1) = F(32, d->XMM_S(1), s->XMM_S(1));\ + d->XMM_L(2) = F(32, d->XMM_S(2), s->XMM_S(2));\ + d->XMM_L(3) = F(32, d->XMM_S(3), s->XMM_S(3));\ }\ \ void OPPROTO op_ ## name ## ss (void)\ @@ -1042,15 +1042,15 @@ void OPPROTO op_ ## name ## ss (void)\ Reg *d, *s;\ d = (Reg *)((char *)env + PARAM1);\ s = (Reg *)((char *)env + PARAM2);\ - d->XMM_L(0) = F(d->XMM_S(0), s->XMM_S(0));\ + d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\ }\ void OPPROTO op_ ## name ## pd (void)\ {\ Reg *d, *s;\ d = (Reg *)((char *)env + PARAM1);\ s = (Reg *)((char *)env + PARAM2);\ - d->XMM_Q(0) = F(d->XMM_D(0), s->XMM_D(0));\ - d->XMM_Q(1) = F(d->XMM_D(1), s->XMM_D(1));\ + d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\ + d->XMM_Q(1) = F(64, d->XMM_D(1), s->XMM_D(1));\ }\ \ void OPPROTO op_ ## name ## sd (void)\ @@ -1058,17 +1058,17 @@ void OPPROTO op_ ## name ## sd (void)\ Reg *d, *s;\ d = (Reg *)((char *)env + PARAM1);\ s = (Reg *)((char *)env + PARAM2);\ - d->XMM_Q(0) = F(d->XMM_D(0), s->XMM_D(0));\ + d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\ } -#define FPU_CMPEQ(a, b) (a) == (b) ? -1 : 0 -#define FPU_CMPLT(a, b) (a) < (b) ? -1 : 0 -#define FPU_CMPLE(a, b) (a) <= (b) ? -1 : 0 -#define FPU_CMPUNORD(a, b) (fpu_isnan(a) || fpu_isnan(b)) ? - 1 : 0 -#define FPU_CMPNEQ(a, b) (a) == (b) ? 0 : -1 -#define FPU_CMPNLT(a, b) (a) < (b) ? 0 : -1 -#define FPU_CMPNLE(a, b) (a) <= (b) ? 0 : -1 -#define FPU_CMPORD(a, b) (!fpu_isnan(a) && !fpu_isnan(b)) ? - 1 : 0 +#define FPU_CMPEQ(size, a, b) float ## size ## _eq(a, b, &env->sse_status) ? -1 : 0 +#define FPU_CMPLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? -1 : 0 +#define FPU_CMPLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? -1 : 0 +#define FPU_CMPUNORD(size, a, b) float ## size ## _unordered(a, b, &env->sse_status) ? - 1 : 0 +#define FPU_CMPNEQ(size, a, b) float ## size ## _eq(a, b, &env->sse_status) ? 0 : -1 +#define FPU_CMPNLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? 0 : -1 +#define FPU_CMPNLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? 0 : -1 +#define FPU_CMPORD(size, a, b) float ## size ## _unordered(a, b, &env->sse_status) ? 0 : -1 SSE_OP_CMP(cmpeq, FPU_CMPEQ) SSE_OP_CMP(cmplt, FPU_CMPLT) @@ -1082,19 +1082,28 @@ SSE_OP_CMP(cmpord, FPU_CMPORD) void OPPROTO op_ucomiss(void) { int eflags; - float s0, s1; + float32 s0, s1; Reg *d, *s; d = (Reg *)((char *)env + PARAM1); s = (Reg *)((char *)env + PARAM2); s0 = d->XMM_S(0); s1 = s->XMM_S(0); - if (s0 < s1) + switch(float32_compare_quiet(s0, s1, &env->sse_status)) { + case -1: eflags = CC_C; - else if (s0 == s1) + break; + case 0: eflags = CC_Z; - else + break; + case 1: eflags = 0; + break; + case 2: + default: + eflags = CC_Z | CC_P | CC_C; + break; + } CC_SRC = eflags; FORCE_RET(); } @@ -1102,19 +1111,28 @@ void OPPROTO op_ucomiss(void) void OPPROTO op_comiss(void) { int eflags; - float s0, s1; + float32 s0, s1; Reg *d, *s; d = (Reg *)((char *)env + PARAM1); s = (Reg *)((char *)env + PARAM2); s0 = d->XMM_S(0); s1 = s->XMM_S(0); - if (s0 < s1) + switch(float32_compare(s0, s1, &env->sse_status)) { + case -1: eflags = CC_C; - else if (s0 == s1) + break; + case 0: eflags = CC_Z; - else + break; + case 1: eflags = 0; + break; + case 2: + default: + eflags = CC_Z | CC_P | CC_C; + break; + } CC_SRC = eflags; FORCE_RET(); } @@ -1122,19 +1140,28 @@ void OPPROTO op_comiss(void) void OPPROTO op_ucomisd(void) { int eflags; - double d0, d1; + float64 d0, d1; Reg *d, *s; d = (Reg *)((char *)env + PARAM1); s = (Reg *)((char *)env + PARAM2); d0 = d->XMM_D(0); d1 = s->XMM_D(0); - if (d0 < d1) + switch(float64_compare_quiet(d0, d1, &env->sse_status)) { + case -1: eflags = CC_C; - else if (d0 == d1) + break; + case 0: eflags = CC_Z; - else + break; + case 1: eflags = 0; + break; + case 2: + default: + eflags = CC_Z | CC_P | CC_C; + break; + } CC_SRC = eflags; FORCE_RET(); } @@ -1142,19 +1169,28 @@ void OPPROTO op_ucomisd(void) void OPPROTO op_comisd(void) { int eflags; - double d0, d1; + float64 d0, d1; Reg *d, *s; d = (Reg *)((char *)env + PARAM1); s = (Reg *)((char *)env + PARAM2); d0 = d->XMM_D(0); d1 = s->XMM_D(0); - if (d0 < d1) + switch(float64_compare(d0, d1, &env->sse_status)) { + case -1: eflags = CC_C; - else if (d0 == d1) + break; + case 0: eflags = CC_Z; - else + break; + case 1: eflags = 0; + break; + case 2: + default: + eflags = CC_Z | CC_P | CC_C; + break; + } CC_SRC = eflags; FORCE_RET(); } -- 2.11.0