[PATCH v3 13/16] x86emul: support AVX10.2 saturating convert insns

[Jan Beulich <jbeulich@xxxxxxxx>]

While the to-byte ones are somewhat different from what has been there
(yet then nicely regular from an operands perspective), the others are
pretty similar to various existing insns.

Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
---
Spec rev 002 says VCVTTNEBF162I{,U}BS, yet that's going to change to
VCVTTBF162I{,U}BS.
---
SDE: ???
---
v3: New.

--- a/tools/tests/x86_emulator/evex-disp8.c
+++ b/tools/tests/x86_emulator/evex-disp8.c
@@ -719,6 +719,30 @@ static const struct test avx10_2_all[] =
     INSN(comxsd,           f3,   0f, 2f,    el,    q, el),
     INSN(comxsh,           f2, map5, 2f,    el, fp16, el),
     INSN(comxss,           f2,   0f, 2f,    el,    d, el),
+    INSN(cvtnebf162ibs,    f2, map5, 69,    vl, bf16, vl),
+    INSN(cvtnebf162iubs,   f2, map5, 6b,    vl, bf16, vl),
+    INSN(cvtph2ibs,          , map5, 69,    vl, fp16, vl),
+    INSN(cvtph2iubs,         , map5, 6b,    vl, fp16, vl),
+    INSN(cvtps2ibs,        66, map5, 69,    vl,    d, vl),
+    INSN(cvtps2iubs,       66, map5, 6b,    vl,    d, vl),
+    INSN(cvttbf162ibs,     f2, map5, 68,    vl, bf16, vl),
+    INSN(cvttbf162iubs,    f2, map5, 6a,    vl, bf16, vl),
+    INSN(cvttpd2dqs,         , map5, 6d,    vl,    q, vl),
+    INSN(cvttpd2qqs,       66, map5, 6d,    vl,    q, vl),
+    INSN(cvttpd2udqs,        , map5, 6c,    vl,    q, vl),
+    INSN(cvttpd2uqqs,      66, map5, 6c,    vl,    q, vl),
+    INSN(cvttph2ibs,         , map5, 68,    vl, fp16, vl),
+    INSN(cvttph2iubs,        , map5, 6a,    vl, fp16, vl),
+    INSN(cvttps2dqs,         , map5, 6d,    vl,    d, vl),
+    INSN(cvttps2ibs,       66, map5, 68,    vl,    d, vl),
+    INSN(cvttps2iubs,      66, map5, 6a,    vl,    d, vl),
+    INSN(cvttps2qqs,       66, map5, 6d,  vl_2,    d, vl),
+    INSN(cvttps2udqs,        , map5, 6c,    vl,    d, vl),
+    INSN(cvttps2uqqs,      66, map5, 6c,  vl_2,    d, vl),
+    INSN(cvttsd2sis,       f2, map5, 6d,    el,    q, el),
+    INSN(cvttsd2usis,      f2, map5, 6c,    el,    q, el),
+    INSN(cvttss2sis,       f3, map5, 6d,    el,    d, el),
+    INSN(cvttss2usis,      f3, map5, 6c,    el,    d, el),
     INSN(divnepbf16,       66, map5, 5e,    vl, bf16, vl),
     INSN(dpphps,             , 0f38, 52,    vl,    d, vl),
     INSN(fmadd132nepbf16,    , map6, 98,    vl, bf16, vl),
--- a/tools/tests/x86_emulator/predicates.c
+++ b/tools/tests/x86_emulator/predicates.c
@@ -2162,6 +2162,26 @@ static const struct evex {
     { { 0x5f }, 2, T, R, pfx_no, W0, Ln }, /* vmaxph */
     { { 0x5f }, 2, T, R, pfx_66, W0, Ln }, /* vmaxpbf16 */
     { { 0x5f }, 2, T, R, pfx_f3, W0, LIG }, /* vmaxsh */
+    { { 0x68 }, 2, T, R, pfx_no, W0, Ln }, /* vcvttph2ibs */
+    { { 0x68 }, 2, T, R, pfx_66, W0, Ln }, /* vcvttps2ibs */
+    { { 0x68 }, 2, T, R, pfx_f2, W0, Ln }, /* vcvttbf162ibs */
+    { { 0x69 }, 2, T, R, pfx_no, W0, Ln }, /* vcvtph2ibs */
+    { { 0x69 }, 2, T, R, pfx_66, W0, Ln }, /* vcvtps2ibs */
+    { { 0x69 }, 2, T, R, pfx_f2, W0, Ln }, /* vcvtnebf162ibs */
+    { { 0x6a }, 2, T, R, pfx_no, W0, Ln }, /* vcvttph2iubs */
+    { { 0x6a }, 2, T, R, pfx_66, W0, Ln }, /* vcvttps2iubs */
+    { { 0x6a }, 2, T, R, pfx_f2, W0, Ln }, /* vcvttbf162iubs */
+    { { 0x6b }, 2, T, R, pfx_no, W0, Ln }, /* vcvtph2iubs */
+    { { 0x6b }, 2, T, R, pfx_66, W0, Ln }, /* vcvtps2iubs */
+    { { 0x6b }, 2, T, R, pfx_f2, W0, Ln }, /* vcvtnebf162iubs */
+    { { 0x6c }, 2, T, R, pfx_no, Wn, Ln }, /* vcvttp{s,d}2udqs */
+    { { 0x6c }, 2, T, R, pfx_66, Wn, Ln }, /* vcvttp{s,d}2uqqs */
+    { { 0x6c }, 2, T, R, pfx_f3, Wn, LIG }, /* vcvttss2usis */
+    { { 0x6c }, 2, T, R, pfx_f2, Wn, LIG }, /* vcvttsd2usis */
+    { { 0x6d }, 2, T, R, pfx_no, Wn, Ln }, /* vcvttp{s,d}2dqs */
+    { { 0x6d }, 2, T, R, pfx_66, Wn, Ln }, /* vcvttp{s,d}2qqs */
+    { { 0x6d }, 2, T, R, pfx_f3, Wn, LIG }, /* vcvttss2sis */
+    { { 0x6d }, 2, T, R, pfx_f2, Wn, LIG }, /* vcvttsd2sis */
     { { 0x6e }, 2, T, R, pfx_66, WIG, L0 }, /* vmovw */
     { { 0x6e }, 2, T, R, pfx_f3, W0, L0 }, /* vmovw */
     { { 0x78 }, 2, T, R, pfx_no, W0, Ln }, /* vcvttph2udq */
--- a/xen/arch/x86/x86_emulate/decode.c
+++ b/xen/arch/x86/x86_emulate/decode.c
@@ -1547,6 +1547,19 @@ int x86emul_decode(struct x86_emulate_st
                     s->fp16 = true;
                 break;
 
+            case 0x68: /* vcvtt{ph,ps,bf16}2ibs */
+            case 0x69: /* vcvt{ph,ps,nebf16}2ibs */
+            case 0x6a: /* vcvtt{ph,ps,bf16}2iubs */
+            case 0x6b: /* vcvt{ph,ps,nebf16}2iubs */
+                if ( !s->evex.pfx || s->evex.pfx == vex_f2 )
+                    s->fp16 = true;
+                /* fall through */
+            case 0x6c: /* vcvttp{s,d}2u{d,q}qs, vcvtts{s,d}2usis */
+            case 0x6d: /* vcvttp{s,d}2{d,q}qs, vcvtts{s,d}2sis */
+                d |= TwoOp;
+                s->simd_size = simd_other;
+                break;
+
             case 0x6e: /* vmovw r/x/m16, xmm */
                 d = (d & ~SrcMask) | SrcMem16;
                 /* fall through */
@@ -1612,6 +1625,14 @@ int x86emul_decode(struct x86_emulate_st
                     --disp8scale;
                 break;
 
+            case 0x6c: /* vcvttps2uqqs and vcvts{s,d}2usi need special casing 
*/
+            case 0x6d: /* vcvttps2qqs and vcvts{s,d}2si need special casing */
+                if ( s->evex.pfx == vex_66 && !s->evex.w && !s->evex.brs )
+                    --disp8scale;
+                else if ( s->evex.pfx & VEX_PREFIX_SCALAR_MASK )
+                    disp8scale = s->evex.pfx & VEX_PREFIX_DOUBLE_MASK ? 3 : 2;
+                break;
+
             case 0x7a: case 0x7b: /* vcvt{,t}ph2qq need special casing */
                 if ( s->evex.pfx == vex_66 && !s->evex.brs )
                     disp8scale = s->evex.brs ? 1 : 2 + s->evex.lr;
--- a/xen/arch/x86/x86_emulate/x86_emulate.c
+++ b/xen/arch/x86/x86_emulate/x86_emulate.c
@@ -8025,6 +8025,55 @@ x86_emulate(
         op_bytes = 8 << evex.lr;
         goto simd_zmm;
 
+    case X86EMUL_OPC_EVEX_F2(5, 0x68): /* vcvttbf162ibs [xyz]mm/mem,[xyz]mm{k} 
*/
+    case X86EMUL_OPC_EVEX_F2(5, 0x69): /* vcvtnebf162ibs 
[xyz]mm/mem,[xyz]mm{k} */
+    case X86EMUL_OPC_EVEX_F2(5, 0x6a): /* vcvttbf162iubs 
[xyz]mm/mem,[xyz]mm{k} */
+    case X86EMUL_OPC_EVEX_F2(5, 0x6b): /* vcvtnebf162iubs 
[xyz]mm/mem,[xyz]mm{k} */
+        generate_exception_if(ea.type != OP_MEM && evex.brs, X86_EXC_UD);
+        /* fall through */
+    case X86EMUL_OPC_EVEX   (5, 0x68): /* vcvttph2ibs [xyz]mm/mem,[xyz]mm{k} */
+    case X86EMUL_OPC_EVEX_66(5, 0x68): /* vcvttps2ibs [xyz]mm/mem,[xyz]mm{k} */
+    case X86EMUL_OPC_EVEX   (5, 0x69): /* vcvtph2ibs [xyz]mm/mem,[xyz]mm{k} */
+    case X86EMUL_OPC_EVEX_66(5, 0x69): /* vcvtps2ibs [xyz]mm/mem,[xyz]mm{k} */
+    case X86EMUL_OPC_EVEX   (5, 0x6a): /* vcvttph2iubs [xyz]mm/mem,[xyz]mm{k} 
*/
+    case X86EMUL_OPC_EVEX_66(5, 0x6a): /* vcvttps2iubs [xyz]mm/mem,[xyz]mm{k} 
*/
+    case X86EMUL_OPC_EVEX   (5, 0x6b): /* vcvtph2iubs [xyz]mm/mem,[xyz]mm{k} */
+    case X86EMUL_OPC_EVEX_66(5, 0x6b): /* vcvtps2iubs [xyz]mm/mem,[xyz]mm{k} */
+        generate_exception_if(evex.w, X86_EXC_UD);
+        vcpu_must_have(avx10, 2);
+        if ( ea.type != OP_REG || !evex.brs )
+            avx512_vlen_check(false);
+        op_bytes = 16 << evex.lr;
+        goto simd_zmm;
+
+    case X86EMUL_OPC_EVEX   (5, 0x6c): /* vcvttps2udqs [xyz]mm/mem,[xyz]mm{k} 
*/
+                                       /* vcvttpd2udqs [xyz]mm/mem,{x,y}mm{k} 
*/
+    case X86EMUL_OPC_EVEX_66(5, 0x6c): /* vcvttps2uqqs {x,y}mm/mem,[xyz]mm{k} 
*/
+                                       /* vcvttpd2uqqs [xyz]mm/mem,[xyz]mm{k} 
*/
+    case X86EMUL_OPC_EVEX   (5, 0x6d): /* vcvttps2dqs [xyz]mm/mem,[xyz]mm{k} */
+                                       /* vcvttpd2dqs [xyz]mm/mem,{x,y}mm{k} */
+    case X86EMUL_OPC_EVEX_66(5, 0x6d): /* vcvttps2qqs {x,y}mm/mem,[xyz]mm{k} */
+                                       /* vcvttpd2qqs [xyz]mm/mem,[xyz]mm{k} */
+        vcpu_must_have(avx10, 2);
+        if ( ea.type != OP_REG || !evex.brs )
+            avx512_vlen_check(false);
+        op_bytes = 8 << ((evex.w || !evex.pfx) + evex.lr);
+        goto simd_zmm;
+
+    CASE_SIMD_SCALAR_FP(_EVEX, 5, 0x6c): /* vcvtts{s,d}2usis xmm/mem,reg */
+    CASE_SIMD_SCALAR_FP(_EVEX, 5, 0x6d): /* vcvtts{s,d}2sis xmm/mem,reg */
+        generate_exception_if((evex.reg != 0xf || !evex.RX || !evex.R ||
+                               evex.opmsk),
+                              X86_EXC_UD);
+        vcpu_must_have(avx10, 2);
+        if ( !evex.brs )
+            avx512_vlen_check(true);
+        else
+            generate_exception_if(ea.type != OP_REG || !evex.u, X86_EXC_UD);
+        get_fpu(X86EMUL_FPU_zmm);
+        opc = init_evex(stub);
+        goto cvts_2si;
+
     case X86EMUL_OPC_EVEX_66(5, 0x78): /* vcvttph2uqq xmm/mem,[xyz]mm{k} */
     case X86EMUL_OPC_EVEX_66(5, 0x79): /* vcvtph2uqq xmm/mem,[xyz]mm{k} */
     case X86EMUL_OPC_EVEX_66(5, 0x7a): /* vcvttph2qq xmm/mem,[xyz]mm{k} */