3DNow! instruction set emulation

(Michael Tross)


git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@4180 c046a42c-6fe2-441c-8c8c-71466251a162

4 files changed, 246 insertions, 10 deletions

diff --git a/target-i386/cpu.h b/target-i386/cpu.h
index 819db70..1c49b36 100644
--- a/target-i386/cpu.h
+++ b/target-i386/cpu.h
@@ -428,8 +428,9 @@ typedef union {
 
 typedef union {
     uint8_t _b[8];
-    uint16_t _w[2];
-    uint32_t _l[1];
+    uint16_t _w[4];
+    uint32_t _l[2];
+    float32 _s[2];
     uint64_t q;
 } MMXReg;
 
@@ -444,6 +445,7 @@ typedef union {
 #define MMX_B(n) _b[7 - (n)]
 #define MMX_W(n) _w[3 - (n)]
 #define MMX_L(n) _l[1 - (n)]
+#define MMX_S(n) _s[1 - (n)]
 #else
 #define XMM_B(n) _b[n]
 #define XMM_W(n) _w[n]
@@ -455,6 +457,7 @@ typedef union {
 #define MMX_B(n) _b[n]
 #define MMX_W(n) _w[n]
 #define MMX_L(n) _l[n]
+#define MMX_S(n) _s[n]
 #endif
 #define MMX_Q(n) q
 
@@ -520,6 +523,7 @@ typedef struct CPUX86State {
         int64_t i64;
     } fp_convert;
 
+    float_status mmx_status; /* for 3DNow! float ops */
     float_status sse_status;
     uint32_t mxcsr;
     XMMReg xmm_regs[CPU_NB_REGS];
diff --git a/target-i386/helper2.c b/target-i386/helper2.c
index 551a0d8..87b79ae 100644
--- a/target-i386/helper2.c
+++ b/target-i386/helper2.c
@@ -150,7 +150,8 @@ static x86_def_t x86_defs[] = {
             CPUID_PSE36,
         .ext_features = CPUID_EXT_SSE3,
         .ext2_features = (PPRO_FEATURES & 0x0183F3FF) | 
-            CPUID_EXT2_LM | CPUID_EXT2_SYSCALL | CPUID_EXT2_NX,
+            CPUID_EXT2_LM | CPUID_EXT2_SYSCALL | CPUID_EXT2_NX |
+            CPUID_EXT2_3DNOW | CPUID_EXT2_3DNOWEXT,
         .ext3_features = CPUID_EXT3_SVM,
         .xlevel = 0x8000000A,
     },
@@ -201,6 +202,19 @@ static x86_def_t x86_defs[] = {
         .features = 0x0383F9FF,
         .xlevel = 0,
     },
+    {
+        .name = "athlon",
+        .level = 2,
+        .vendor1 = 0x68747541, /* "Auth" */
+        .vendor2 = 0x69746e65, /* "enti" */
+        .vendor3 = 0x444d4163, /* "cAMD" */
+        .family = 6,
+        .model = 2,
+        .stepping = 3,
+        .features = PPRO_FEATURES | PPRO_FEATURES | CPUID_PSE36 | CPUID_VME | CPUID_MTRR | CPUID_MCA,
+        .ext2_features = (PPRO_FEATURES & 0x0183F3FF) | CPUID_EXT2_MMXEXT | CPUID_EXT2_3DNOW | CPUID_EXT2_3DNOWEXT,
+        .xlevel = 0x80000008,
+    },
 };
 
 static int cpu_x86_find_by_name(x86_def_t *x86_cpu_def, const char *cpu_model)
diff --git a/target-i386/ops_sse.h b/target-i386/ops_sse.h
index 82d1ec0..6610e75 100644
--- a/target-i386/ops_sse.h
+++ b/target-i386/ops_sse.h
@@ -1,5 +1,5 @@
 /*
- *  MMX/SSE/SSE2/PNI support
+ *  MMX/3DNow!/SSE/SSE2/SSE3/PNI support
  *
  *  Copyright (c) 2005 Fabrice Bellard
  *
@@ -409,6 +409,7 @@ static inline int satsw(int x)
 #define FCMPEQ(a, b) (a) == (b) ? -1 : 0
 
 #define FMULLW(a, b) (a) * (b)
+#define FMULHRW(a, b) ((int16_t)(a) * (int16_t)(b) + 0x8000) >> 16
 #define FMULHUW(a, b) (a) * (b) >> 16
 #define FMULHW(a, b) (int16_t)(a) * (int16_t)(b) >> 16
 
@@ -455,6 +456,9 @@ SSE_OP_W(op_pcmpeqw, FCMPEQ)
 SSE_OP_L(op_pcmpeql, FCMPEQ)
 
 SSE_OP_W(op_pmullw, FMULLW)
+#if SHIFT == 0
+SSE_OP_W(op_pmulhrw, FMULHRW)
+#endif
 SSE_OP_W(op_pmulhuw, FMULHUW)
 SSE_OP_W(op_pmulhw, FMULHW)
 
@@ -1383,6 +1387,175 @@ void OPPROTO glue(op_punpck ## base_name ## qdq, SUFFIX) (void)  \
 UNPCK_OP(l, 0)
 UNPCK_OP(h, 1)
 
+/* 3DNow! float ops */
+#if SHIFT == 0
+void OPPROTO op_pi2fd(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_S(0) = int32_to_float32(s->MMX_L(0), &env->mmx_status);
+    d->MMX_S(1) = int32_to_float32(s->MMX_L(1), &env->mmx_status);
+}
+
+void OPPROTO op_pi2fw(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_S(0) = int32_to_float32((int16_t)s->MMX_W(0), &env->mmx_status);
+    d->MMX_S(1) = int32_to_float32((int16_t)s->MMX_W(2), &env->mmx_status);
+}
+
+void OPPROTO op_pf2id(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_L(0) = float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status);
+    d->MMX_L(1) = float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status);
+}
+
+void OPPROTO op_pf2iw(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_L(0) = satsw(float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status));
+    d->MMX_L(1) = satsw(float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status));
+}
+
+void OPPROTO op_pfacc(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    MMXReg r;
+    r.MMX_S(0) = float32_add(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
+    r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
+    *d = r;
+}
+
+void OPPROTO op_pfadd(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_S(0) = float32_add(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
+    d->MMX_S(1) = float32_add(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
+}
+
+void OPPROTO op_pfcmpeq(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_L(0) = float32_eq(d->MMX_S(0), s->MMX_S(0), &env->mmx_status) ? -1 : 0;
+    d->MMX_L(1) = float32_eq(d->MMX_S(1), s->MMX_S(1), &env->mmx_status) ? -1 : 0;
+}
+
+void OPPROTO op_pfcmpge(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_L(0) = float32_le(s->MMX_S(0), d->MMX_S(0), &env->mmx_status) ? -1 : 0;
+    d->MMX_L(1) = float32_le(s->MMX_S(1), d->MMX_S(1), &env->mmx_status) ? -1 : 0;
+}
+
+void OPPROTO op_pfcmpgt(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_L(0) = float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status) ? -1 : 0;
+    d->MMX_L(1) = float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status) ? -1 : 0;
+}
+
+void OPPROTO op_pfmax(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    if (float32_lt(d->MMX_S(0), s->MMX_S(0), &env->mmx_status))
+        d->MMX_S(0) = s->MMX_S(0);
+    if (float32_lt(d->MMX_S(1), s->MMX_S(1), &env->mmx_status))
+        d->MMX_S(1) = s->MMX_S(1);
+}
+
+void OPPROTO op_pfmin(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    if (float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status))
+        d->MMX_S(0) = s->MMX_S(0);
+    if (float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status))
+        d->MMX_S(1) = s->MMX_S(1);
+}
+
+void OPPROTO op_pfmul(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_S(0) = float32_mul(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
+    d->MMX_S(1) = float32_mul(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
+}
+
+void OPPROTO op_pfnacc(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    MMXReg r;
+    r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
+    r.MMX_S(1) = float32_sub(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
+    *d = r;
+}
+
+void OPPROTO op_pfpnacc(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    MMXReg r;
+    r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
+    r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
+    *d = r;
+}
+
+void OPPROTO op_pfrcp(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_S(0) = approx_rcp(s->MMX_S(0));
+    d->MMX_S(1) = d->MMX_S(0);
+}
+
+void OPPROTO op_pfrsqrt(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_L(1) = s->MMX_L(0) & 0x7fffffff;
+    d->MMX_S(1) = approx_rsqrt(d->MMX_S(1));
+    d->MMX_L(1) |= s->MMX_L(0) & 0x80000000;
+    d->MMX_L(0) = d->MMX_L(1);
+}
+
+void OPPROTO op_pfsub(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_S(0) = float32_sub(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
+    d->MMX_S(1) = float32_sub(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
+}
+
+void OPPROTO op_pfsubr(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    d->MMX_S(0) = float32_sub(s->MMX_S(0), d->MMX_S(0), &env->mmx_status);
+    d->MMX_S(1) = float32_sub(s->MMX_S(1), d->MMX_S(1), &env->mmx_status);
+}
+
+void OPPROTO op_pswapd(void)
+{
+    MMXReg *d = (MMXReg *)((char *)env + PARAM1);
+    MMXReg *s = (MMXReg *)((char *)env + PARAM2);
+    MMXReg r;
+    r.MMX_L(0) = s->MMX_L(1);
+    r.MMX_L(1) = s->MMX_L(0);
+    *d = r;
+}
+#endif
+
 #undef SHIFT
 #undef XMM_ONLY
 #undef Reg
diff --git a/target-i386/translate.c b/target-i386/translate.c
index 356ceff..74ad2dc 100644
--- a/target-i386/translate.c
+++ b/target-i386/translate.c
@@ -2408,12 +2408,16 @@ static GenOpFunc1 *gen_sto_env_A0[3] = {
 };
 
 #define SSE_SPECIAL ((GenOpFunc2 *)1)
+#define SSE_DUMMY ((GenOpFunc2 *)2)
 
 #define MMX_OP2(x) { gen_op_ ## x ## _mmx, gen_op_ ## x ## _xmm }
 #define SSE_FOP(x) { gen_op_ ## x ## ps, gen_op_ ## x ## pd, \
                      gen_op_ ## x ## ss, gen_op_ ## x ## sd, }
 
 static GenOpFunc2 *sse_op_table1[256][4] = {
+    /* 3DNow! extensions */
+    [0x0e] = { SSE_DUMMY }, /* femms */
+    [0x0f] = { SSE_DUMMY }, /* pf... */
     /* pure SSE operations */
     [0x10] = { SSE_SPECIAL, SSE_SPECIAL, SSE_SPECIAL, SSE_SPECIAL }, /* movups, movupd, movss, movsd */
     [0x11] = { SSE_SPECIAL, SSE_SPECIAL, SSE_SPECIAL, SSE_SPECIAL }, /* movups, movupd, movss, movsd */
@@ -2480,7 +2484,7 @@ static GenOpFunc2 *sse_op_table1[256][4] = {
     [0x74] = MMX_OP2(pcmpeqb),
     [0x75] = MMX_OP2(pcmpeqw),
     [0x76] = MMX_OP2(pcmpeql),
-    [0x77] = { SSE_SPECIAL }, /* emms */
+    [0x77] = { SSE_DUMMY }, /* emms */
     [0x7c] = { NULL, gen_op_haddpd, NULL, gen_op_haddps },
     [0x7d] = { NULL, gen_op_hsubpd, NULL, gen_op_hsubps },
     [0x7e] = { SSE_SPECIAL, SSE_SPECIAL, SSE_SPECIAL }, /* movd, movd, , movq */
@@ -2577,6 +2581,33 @@ static GenOpFunc2 *sse_op_table4[8][4] = {
     SSE_FOP(cmpord),
 };
 
+static GenOpFunc2 *sse_op_table5[256] = {
+    [0x0c] = gen_op_pi2fw,
+    [0x0d] = gen_op_pi2fd,
+    [0x1c] = gen_op_pf2iw,
+    [0x1d] = gen_op_pf2id,
+    [0x8a] = gen_op_pfnacc,
+    [0x8e] = gen_op_pfpnacc,
+    [0x90] = gen_op_pfcmpge,
+    [0x94] = gen_op_pfmin,
+    [0x96] = gen_op_pfrcp,
+    [0x97] = gen_op_pfrsqrt,
+    [0x9a] = gen_op_pfsub,
+    [0x9e] = gen_op_pfadd,
+    [0xa0] = gen_op_pfcmpgt,
+    [0xa4] = gen_op_pfmax,
+    [0xa6] = gen_op_movq, /* pfrcpit1; no need to actually increase precision */
+    [0xa7] = gen_op_movq, /* pfrsqit1 */
+    [0xaa] = gen_op_pfsubr,
+    [0xae] = gen_op_pfacc,
+    [0xb0] = gen_op_pfcmpeq,
+    [0xb4] = gen_op_pfmul,
+    [0xb6] = gen_op_movq, /* pfrcpit2 */
+    [0xb7] = gen_op_pmulhrw_mmx,
+    [0xbb] = gen_op_pswapd,
+    [0xbf] = gen_op_pavgb_mmx /* pavgusb */
+};
+
 static void gen_sse(DisasContext *s, int b, target_ulong pc_start, int rex_r)
 {
     int b1, op1_offset, op2_offset, is_xmm, val, ot;
@@ -2596,7 +2627,7 @@ static void gen_sse(DisasContext *s, int b, target_ulong pc_start, int rex_r)
     sse_op2 = sse_op_table1[b][b1];
     if (!sse_op2)
         goto illegal_op;
-    if (b <= 0x5f || b == 0xc6 || b == 0xc2) {
+    if ((b <= 0x5f && b >= 0x10) || b == 0xc6 || b == 0xc2) {
         is_xmm = 1;
     } else {
         if (b1 == 0) {
@@ -2618,8 +2649,8 @@ static void gen_sse(DisasContext *s, int b, target_ulong pc_start, int rex_r)
     }
     if (is_xmm && !(s->flags & HF_OSFXSR_MASK))
         goto illegal_op;
-    if (b == 0x77) {
-        /* emms */
+    if (b == 0x77 || b == 0x0e) {
+        /* emms or femms */
         gen_op_emms();
         return;
     }
@@ -3151,6 +3182,13 @@ static void gen_sse(DisasContext *s, int b, target_ulong pc_start, int rex_r)
             }
         }
         switch(b) {
+        case 0x0f: /* 3DNow! data insns */
+            val = ldub_code(s->pc++);
+            sse_op2 = sse_op_table5[val];
+            if (!sse_op2)
+                goto illegal_op;
+            sse_op2(op1_offset, op2_offset);
+            break;
         case 0x70: /* pshufx insn */
         case 0xc6: /* pshufx insn */
             val = ldub_code(s->pc++);
@@ -6148,7 +6186,7 @@ static target_ulong disas_insn(DisasContext *s, target_ulong pc_start)
             gen_eob(s);
         }
         break;
-    /* MMX/SSE/SSE2/PNI support */
+    /* MMX/3DNow!/SSE/SSE2/SSE3 support */
     case 0x1c3: /* MOVNTI reg, mem */
         if (!(s->cpuid_features & CPUID_SSE2))
             goto illegal_op;
@@ -6214,6 +6252,7 @@ static target_ulong disas_insn(DisasContext *s, target_ulong pc_start)
         case 7: /* sfence / clflush */
             if ((modrm & 0xc7) == 0xc0) {
                 /* sfence */
+                /* XXX: also check for cpuid_ext2_features & CPUID_EXT2_EMMX */
                 if (!(s->cpuid_features & CPUID_SSE))
                     goto illegal_op;
             } else {
@@ -6227,8 +6266,11 @@ static target_ulong disas_insn(DisasContext *s, target_ulong pc_start)
             goto illegal_op;
         }
         break;
-    case 0x10d: /* prefetch */
+    case 0x10d: /* 3DNow! prefetch(w) */
         modrm = ldub_code(s->pc++);
+        mod = (modrm >> 6) & 3;
+        if (mod == 3)
+            goto illegal_op;
         gen_lea_modrm(s, modrm, &reg_addr, &offset_addr);
         /* ignore for now */
         break;
@@ -6245,6 +6287,9 @@ static target_ulong disas_insn(DisasContext *s, target_ulong pc_start)
         gen_op_rsm();
         gen_eob(s);
         break;
+    case 0x10e ... 0x10f:
+        /* 3DNow! instructions, ignore prefixes */
+        s->prefix &= ~(PREFIX_REPZ | PREFIX_REPNZ | PREFIX_DATA);
     case 0x110 ... 0x117:
     case 0x128 ... 0x12f:
     case 0x150 ... 0x177: