Yet more SSE insns.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@1696 a5019735-40e9-0310-863c-91ae7b9d1cf9
diff --git a/addrcheck/ac_main.c b/addrcheck/ac_main.c
index 92c80d2..407d001 100644
--- a/addrcheck/ac_main.c
+++ b/addrcheck/ac_main.c
@@ -961,6 +961,20 @@
VG_(copy_UInstr)(cb, u_in);
break;
+ case SSE3ag_MemRd_RegWr:
+ sk_assert(u_in->size == 4 || u_in->size == 8);
+ goto do_Access_ARG1;
+ do_Access_ARG1:
+ sk_assert(u_in->tag1 == TempReg);
+ t_addr = u_in->val1;
+ t_size = newTemp(cb);
+ uInstr2(cb, MOV, 4, Literal, 0, TempReg, t_size);
+ uLiteral(cb, u_in->size);
+ uInstr2(cb, CCALL, 0, TempReg, t_addr, TempReg, t_size);
+ uCCall(cb, (Addr) & ac_fpu_ACCESS_check, 2, 2, False );
+ VG_(copy_UInstr)(cb, u_in);
+ break;
+
case MMX2_MemRd:
case MMX2_MemWr:
sk_assert(u_in->size == 4 || u_in->size == 8);
@@ -999,15 +1013,14 @@
// case SSE2a1_MemRd:
// case SSE2a1_MemWr:
- case SSE3g1_RegWr:
case SSE3g1_RegRd:
- // case SSE3ag_MemRd_RegWr:
// case SSE3a1_MemRd:
// case SSE3a1_MemWr:
VG_(pp_UInstr)(0,u_in);
VG_(skin_panic)("AddrCheck: unhandled SSE uinstr");
break;
+ case SSE3g1_RegWr:
case SSE5:
case SSE3g_RegWr:
case SSE3g_RegRd:
diff --git a/coregrind/vg_from_ucode.c b/coregrind/vg_from_ucode.c
index af63fa2..8b699e6 100644
--- a/coregrind/vg_from_ucode.c
+++ b/coregrind/vg_from_ucode.c
@@ -1563,11 +1563,11 @@
}
static void emit_SSE4 ( FlagSet uses_sflags,
- FlagSet sets_sflags,
- UChar first_byte,
- UChar second_byte,
- UChar third_byte,
- UChar fourth_byte )
+ FlagSet sets_sflags,
+ UChar first_byte,
+ UChar second_byte,
+ UChar third_byte,
+ UChar fourth_byte )
{
VG_(new_emit)(True, uses_sflags, sets_sflags);
VG_(emitB) ( first_byte );
@@ -1581,12 +1581,12 @@
}
static void emit_SSE5 ( FlagSet uses_sflags,
- FlagSet sets_sflags,
- UChar first_byte,
- UChar second_byte,
- UChar third_byte,
- UChar fourth_byte,
- UChar fifth_byte )
+ FlagSet sets_sflags,
+ UChar first_byte,
+ UChar second_byte,
+ UChar third_byte,
+ UChar fourth_byte,
+ UChar fifth_byte )
{
VG_(new_emit)(True, uses_sflags, sets_sflags);
VG_(emitB) ( first_byte );
@@ -1602,10 +1602,10 @@
}
static void emit_SSE3 ( FlagSet uses_sflags,
- FlagSet sets_sflags,
- UChar first_byte,
- UChar second_byte,
- UChar third_byte )
+ FlagSet sets_sflags,
+ UChar first_byte,
+ UChar second_byte,
+ UChar third_byte )
{
VG_(new_emit)(True, uses_sflags, sets_sflags);
VG_(emitB) ( first_byte );
@@ -1617,6 +1617,28 @@
(UInt)third_byte );
}
+static void emit_SSE3ag_MemRd_RegWr ( FlagSet uses_sflags,
+ FlagSet sets_sflags,
+ UChar first_byte,
+ UChar second_byte,
+ UChar third_byte,
+ Int addr_reg,
+ Int dest_reg )
+{
+ VG_(new_emit)(True, uses_sflags, sets_sflags);
+ VG_(emitB) ( first_byte );
+ VG_(emitB) ( second_byte );
+ VG_(emitB) ( third_byte );
+ /* 4th byte can be completely synthesised from addr_reg and
+ dest_reg. */
+ emit_amode_regmem_reg ( addr_reg, dest_reg );
+ if (dis)
+ VG_(printf)("\n\t\tsse-0x%x:0x%x:0x%x(addr=%s, dest=%s)\n",
+ (UInt)first_byte, (UInt)second_byte,
+ (UInt)third_byte, nameIReg(4, addr_reg),
+ nameIReg(4, dest_reg));
+}
+
static void emit_MMX2_reg_to_mmxreg ( FlagSet uses_sflags,
FlagSet sets_sflags,
UChar first_byte,
@@ -3944,6 +3966,23 @@
u->val2 & 0xFF );
break;
+ case SSE3ag_MemRd_RegWr:
+ vg_assert(u->size == 4 || u->size == 8);
+ vg_assert(u->tag1 == RealReg);
+ vg_assert(u->tag2 == RealReg);
+ vg_assert(u->tag3 == NoValue);
+ vg_assert(!anyFlagUse(u));
+ if (!(*sselive)) {
+ emit_get_sse_state();
+ *sselive = True;
+ }
+ emit_SSE3ag_MemRd_RegWr ( u->flags_r, u->flags_w,
+ (u->lit32 >> 24) & 0xFF,
+ (u->lit32 >> 16) & 0xFF,
+ (u->lit32 >> 8) & 0xFF,
+ u->val1, u->val2 );
+ break;
+
default:
if (VG_(needs).extended_UCode) {
if (*sselive) {
diff --git a/coregrind/vg_to_ucode.c b/coregrind/vg_to_ucode.c
index d7cc5e2..fa8a7da 100644
--- a/coregrind/vg_to_ucode.c
+++ b/coregrind/vg_to_ucode.c
@@ -3659,19 +3659,24 @@
goto decode_success;
}
- /* CVTTSD2SI (0xF2) -- convert a double-precision float value in
- memory or xmm reg to int and put it in an ireg. */
- /* CVTTSS2SI (0xF3) -- convert a single-precision float value in
- memory or xmm reg to int and put it in an ireg. */
+ /* CVTTSD2SI (0xF2,0x0F,0x2C) -- convert a double-precision float value in
+ memory or xmm reg to int and put it in an ireg. Truncate. */
+ /* CVTTSS2SI (0xF3,0x0F,0x2C) -- convert a single-precision float value in
+ memory or xmm reg to int and put it in an ireg. Truncate. */
+ /* CVTSD2SI (0xF2,0x0F,0x2D) -- convert a double-precision float value in
+ memory or xmm reg to int and put it in an ireg. Round as per MXCSR. */
+ /* CVTSS2SI (0xF3,0x0F,0x2D) -- convert a single-precision float value in
+ memory or xmm reg to int and put it in an ireg. Round as per MXCSR. */
if ((insn[0] == 0xF2 || insn[0] == 0xF3)
- && insn[1] == 0x0F && insn[2] == 0x2C) {
+ && insn[1] == 0x0F
+ && (insn[2] == 0x2C || insn[2] == 0x2D)) {
vg_assert(sz == 4);
modrm = insn[3];
if (epartIsReg(modrm)) {
/* We're moving a value in an xmm reg to an ireg. */
eip += 4;
t1 = newTemp(cb);
- /* sz is 4 for both CVTTSD2SI and CVTTSS2SI. */
+ /* sz is 4 for all 4 insns. */
vg_assert(epartIsReg(modrm));
uInstr3(cb, SSE3g_RegWr, 4,
Lit16, (((UShort)insn[0]) << 8) | (UShort)insn[1],
@@ -3679,27 +3684,34 @@
TempReg, t1 );
uInstr2(cb, PUT, 4, TempReg, t1, ArchReg, gregOfRM(modrm));
if (dis)
- VG_(printf)("cvtts{s,d}2si %s, %s\n",
+ VG_(printf)("cvt{t}s{s,d}2si %s, %s\n",
nameXMMReg(eregOfRM(modrm)),
nameIReg(4,gregOfRM(modrm)) );
} else {
-#if 0
/* So, we're reading memory and writing an ireg. This calls
- for the ultra-horrible SSE3ag_MemRd_RegWr uinstr. */
+ for the ultra-horrible SSE3ag_MemRd_RegWr uinstr. We
+ can't do it in a roundabout route because it does some
+ kind of conversion on the way, which we need to have
+ happen too. So our only choice is to re-emit a suitably
+ rehashed version of the instruction. */
+ /* Destination ireg is GREG. Address goes as EREG as
+ usual. */
t1 = newTemp(cb); /* t1 holds value on its way to ireg */
pair = disAMode ( cb, sorb, eip+3, dis?dis_buf:NULL );
t2 = LOW24(pair); /* t2 holds addr */
eip += 3+HI8(pair);
- uInstr2(cb, SSE3ag_MemRd_RegWr, 8,
+ uInstr2(cb, SSE3ag_MemRd_RegWr, insn[0]==0xF2 ? 8 : 4,
TempReg, t2, /* address */
TempReg, t1 /* dest */);
- uLiteral(cb, (((UInt)insn[0]) << 16)
- | (((UInt)insn[1]) << 8)
- | ((UInt)insn[2]) );
+ uLiteral(cb , (((UInt)insn[0]) << 24)
+ | (((UInt)insn[1]) << 16)
+ | (((UInt)insn[2]) << 8)
+ | ((UInt)modrm) );
uInstr2(cb, PUT, 4, TempReg, t1, ArchReg, gregOfRM(modrm));
- /* PRINTING CODE */
-#endif
- VG_(core_panic)("CVTTSD2SI mem");
+ if (dis)
+ VG_(printf)("cvt{t}s{s,d}2si %s, %s\n",
+ dis_buf,
+ nameIReg(4,gregOfRM(modrm)) );
}
goto decode_success;
}
@@ -3839,6 +3851,14 @@
goto decode_success;
}
+ /* XORPD (src)xmmreg-or-mem, (dst)xmmreg */
+ if (sz == 2
+ && insn[0] == 0x0F && insn[1] == 0x57) {
+ eip = dis_SSE3_reg_or_mem ( cb, sorb, eip+2, 16, "xorpd",
+ 0x66, insn[0], insn[1] );
+ goto decode_success;
+ }
+
/* PXOR (src)xmmreg-or-mem, (dst)xmmreg */
if (sz == 2
&& insn[0] == 0x0F && insn[1] == 0xEF) {
diff --git a/coregrind/vg_translate.c b/coregrind/vg_translate.c
index 48b4f2d..7f9570b 100644
--- a/coregrind/vg_translate.c
+++ b/coregrind/vg_translate.c
@@ -570,6 +570,8 @@
case SSE3: return LIT0 && SZ0 && CC0 && Ls1 && Ls2 && N3 && XOTHER;
case SSE4: return LIT0 && SZ0 && CCf && Ls1 && Ls2 && N3 && XOTHER;
case SSE5: return LIT0 && SZ0 && CC0 && Ls1 && Ls2 && Ls3 && XOTHER;
+ case SSE3ag_MemRd_RegWr:
+ return SZ48 && CC0 && TR1 && TR2 && N3 && XOTHER;
default:
if (VG_(needs).extended_UCode)
return SK_(sane_XUInstr)(beforeRA, beforeLiveness, u);
@@ -893,6 +895,7 @@
case SSE5: return "SSE5";
case SSE3a_MemWr: return "SSE3a_MWr";
case SSE3a_MemRd: return "SSE3a_MRd";
+ case SSE3ag_MemRd_RegWr: return "SSE3ag_MemRd_RegWr";
default:
if (VG_(needs).extended_UCode)
return SK_(name_XUOpcode)(opc);
@@ -1094,6 +1097,14 @@
u->val3 & 0xFF );
break;
+ case SSE3ag_MemRd_RegWr:
+ VG_(printf)("0x%x(addr=", u->lit32 );
+ VG_(pp_UOperand)(u, 1, 4, False);
+ VG_(printf)(", dst=");
+ VG_(pp_UOperand)(u, 2, 4, False);
+ VG_(printf)(")");
+ break;
+
case GET: case PUT: case MOV: case LOAD: case STORE: case CMOV:
case GETSEG: case PUTSEG:
VG_(pp_UOperand)(u, 1, u->size, u->opcode==LOAD);
@@ -1249,6 +1260,8 @@
case SSE3g1_RegWr:
case SSE3g_RegWr: WR(3); break;
+ case SSE3ag_MemRd_RegWr: RD(1); WR(2); break;
+
case MMX2_RegRd: RD(2); break;
case MMX2_RegWr: WR(2); break;
@@ -1408,7 +1421,7 @@
case SSE3a_MemWr: case SSE3a_MemRd:
case SSE3g_RegRd: case SSE3g_RegWr:
case SSE3g1_RegWr: case SSE3g1_RegRd:
- case SSE4: case SSE3: case SSE5:
+ case SSE4: case SSE3: case SSE5: case SSE3ag_MemRd_RegWr:
case WIDEN:
/* GETSEG and USESEG are to do with ArchRegS, not ArchReg */
case GETSEG: case PUTSEG:
@@ -2298,7 +2311,7 @@
UChar* final_code;
UCodeBlock* cb;
Bool notrace_until_done;
- UInt notrace_until_limit = 23000;
+ UInt notrace_until_limit = 53000;
VGP_PUSHCC(VgpTranslate);
debugging_translation
diff --git a/include/vg_skin.h b/include/vg_skin.h
index e0570d6..6e37a98 100644
--- a/include/vg_skin.h
+++ b/include/vg_skin.h
@@ -662,12 +662,11 @@
*/
SSE3g_RegRd,
-#if 0
/* 4 bytes, reads memory, writes an integer register, but is
nevertheless an SSE insn. The insn is of the form
bbbbbbbb:bbbbbbbb:bbbbbbbb:mod ireg rm where mod indicates
memory (ie is not 11b) and ireg is the int reg written. The
- first 3 bytes are held in lit32[23:0] since there is
+ first 4 bytes are held in lit32[31:0] since there is
insufficient space elsewhere. mod and rm are to be replaced
at codegen time by a reference to the Temp/RealReg holding
the address. Arg1 holds this Temp/RealReg. ireg is to be
@@ -675,10 +674,13 @@
RealReg in which the answer is to be written. Arg2 holds
this Temp/RealReg. Transfer to the destination reg is always
at size 4. However the memory read can be at sizes 4 or 8
- and so this is what the sz field holds.
+ and so this is what the sz field holds. Note that the 4th
+ byte of the instruction (the modrm byte) is redundant, but we
+ store it anyway so as to be consistent with all other SSE
+ uinstrs.
*/
SSE3ag_MemRd_RegWr,
-#endif
+
/* 5 bytes, no memrefs, no iregdefs, copy exactly to the
output. Held in val1[15:0], val2[15:0] and val3[7:0]. */
SSE5,