PPC: Support v[u]pkd3d128 VPACK_NORMSHORT4

src/xenia/cpu/backend/x64/x64_emitter.cc

@@ -639,12 +639,16 @@ static const vec128_t xmm_consts[] = {
                                          0x01000302u),
     /* XMMUnpackFLOAT16_4     */ vec128i(0x09080B0Au, 0x0D0C0F0Eu, 0xFFFFFFFFu,
                                          0xFFFFFFFFu),
-    /* XMMPackSHORT_2Min      */ vec128i(0x403F8001u),
+    /* XMMPackSHORT_Min       */ vec128i(0x403F8001u),
-    /* XMMPackSHORT_2Max      */ vec128i(0x40407FFFu),
+    /* XMMPackSHORT_Max       */ vec128i(0x40407FFFu),
     /* XMMPackSHORT_2         */ vec128i(0xFFFFFFFFu, 0xFFFFFFFFu, 0xFFFFFFFFu,
                                          0x01000504u),
+    /* XMMPackSHORT_4         */ vec128i(0xFFFFFFFFu, 0xFFFFFFFFu, 0x01000504u,
+                                         0x09080D0Cu),
     /* XMMUnpackSHORT_2       */ vec128i(0xFFFF0F0Eu, 0xFFFF0D0Cu, 0xFFFFFFFFu,
                                          0xFFFFFFFFu),
+    /* XMMUnpackSHORT_4       */ vec128i(0xFFFF0B0Au, 0xFFFF0908u, 0xFFFF0F0Eu,
+                                         0xFFFF0D0Cu),
     /* XMMOneOver255          */ vec128f(1.0f / 255.0f),
     /* XMMMaskEvenPI16        */ vec128i(0x0000FFFFu, 0x0000FFFFu, 0x0000FFFFu,
                                          0x0000FFFFu),

src/xenia/cpu/backend/x64/x64_emitter.h

@@ -70,10 +70,12 @@ enum XmmConst {
   XMMUnpackFLOAT16_2,
   XMMPackFLOAT16_4,
   XMMUnpackFLOAT16_4,
-  XMMPackSHORT_2Min,
+  XMMPackSHORT_Min,
-  XMMPackSHORT_2Max,
+  XMMPackSHORT_Max,
   XMMPackSHORT_2,
+  XMMPackSHORT_4,
   XMMUnpackSHORT_2,
+  XMMUnpackSHORT_4,
   XMMOneOver255,
   XMMMaskEvenPI16,
   XMMShiftMaskEvenPI16,

src/xenia/cpu/backend/x64/x64_sequences.cc

@@ -6881,6 +6881,9 @@ struct PACK : Sequence<PACK, I<OPCODE_PACK, V128Op, V128Op, V128Op>> {
       case PACK_TYPE_SHORT_2:
         EmitSHORT_2(e, i);
         break;
+      case PACK_TYPE_SHORT_4:
+        EmitSHORT_4(e, i);
+        break;
       case PACK_TYPE_UINT_2101010:
         EmitUINT_2101010(e, i);
         break;
@@ -6970,11 +6973,19 @@ struct PACK : Sequence<PACK, I<OPCODE_PACK, V128Op, V128Op, V128Op>> {
   static void EmitSHORT_2(X64Emitter& e, const EmitArgType& i) {
     assert_true(i.src2.value->IsConstantZero());
     // Saturate.
-    e.vmaxps(i.dest, i.src1, e.GetXmmConstPtr(XMMPackSHORT_2Min));
+    e.vmaxps(i.dest, i.src1, e.GetXmmConstPtr(XMMPackSHORT_Min));
-    e.vminps(i.dest, i.dest, e.GetXmmConstPtr(XMMPackSHORT_2Max));
+    e.vminps(i.dest, i.dest, e.GetXmmConstPtr(XMMPackSHORT_Max));
     // Pack.
     e.vpshufb(i.dest, i.dest, e.GetXmmConstPtr(XMMPackSHORT_2));
   }
+  static void EmitSHORT_4(X64Emitter& e, const EmitArgType& i) {
+    assert_true(i.src2.value->IsConstantZero());
+    // Saturate.
+    e.vmaxps(i.dest, i.src1, e.GetXmmConstPtr(XMMPackSHORT_Min));
+    e.vminps(i.dest, i.dest, e.GetXmmConstPtr(XMMPackSHORT_Max));
+    // Pack.
+    e.vpshufb(i.dest, i.dest, e.GetXmmConstPtr(XMMPackSHORT_4));
+  }
   static __m128i EmulatePackUINT_2101010(void*, __m128i src1) {
     // https://www.opengl.org/registry/specs/ARB/vertex_type_2_10_10_10_rev.txt
     union {
@@ -7229,15 +7240,15 @@ struct UNPACK : Sequence<UNPACK, I<OPCODE_UNPACK, V128Op, V128Op>> {
       case PACK_TYPE_FLOAT16_2:
         EmitFLOAT16_2(e, i);
         break;
-      case PACK_TYPE_FLOAT16_3:
-        EmitFLOAT16_3(e, i);
-        break;
       case PACK_TYPE_FLOAT16_4:
         EmitFLOAT16_4(e, i);
         break;
       case PACK_TYPE_SHORT_2:
         EmitSHORT_2(e, i);
         break;
+      case PACK_TYPE_SHORT_4:
+        EmitSHORT_4(e, i);
+        break;
       case PACK_TYPE_UINT_2101010:
         EmitUINT_2101010(e, i);
         break;
@@ -7323,27 +7334,6 @@ struct UNPACK : Sequence<UNPACK, I<OPCODE_UNPACK, V128Op, V128Op>> {
       e.vmovaps(i.dest, e.xmm0);
     }
   }
-  // FIXME: This has not been verified on a real 360, but from context the
-  // return values are used in floating point math.
-  static __m128 EmulateFLOAT16_3(void*, __m128i src1) {
-    alignas(16) uint16_t a[8];
-    alignas(16) float b[4];
-    _mm_store_si128(reinterpret_cast<__m128i*>(a), src1);
-
-    for (int i = 0; i < 3; i++) {
-      b[i] = half_float::detail::half2float(a[VEC128_W(5 + i)]);
-    }
-
-    // FIXME: Correct?
-    b[3] = 1.0f;
-
-    return _mm_load_ps(b);
-  }
-  static void EmitFLOAT16_3(X64Emitter& e, const EmitArgType& i) {
-    e.lea(e.r8, e.StashXmm(0, i.src1));
-    e.CallNativeSafe(reinterpret_cast<void*>(EmulateFLOAT16_3));
-    e.vmovaps(i.dest, e.xmm0);
-  }
   static __m128 EmulateFLOAT16_4(void*, __m128i src1) {
     alignas(16) uint16_t a[8];
     alignas(16) float b[4];
@@ -7398,6 +7388,36 @@ struct UNPACK : Sequence<UNPACK, I<OPCODE_UNPACK, V128Op, V128Op>> {
     // Add 3,3,0,1.
     e.vpaddd(i.dest, e.GetXmmConstPtr(XMM3301));
   }
+  static void EmitSHORT_4(X64Emitter& e, const EmitArgType& i) {
+    // (VD.x) = 3.0 + (VB.x>>16)*2^-22
+    // (VD.y) = 3.0 + (VB.x)*2^-22
+    // (VD.z) = 3.0 + (VB.y>>16)*2^-22
+    // (VD.w) = 3.0 + (VB.y)*2^-22
+
+    // XMLoadShortN4 plus 3,3,3,3 (for some reason)
+    // src is (xx,xx,VALUE,VALUE)
+    // (VALUE,VALUE,VALUE,VALUE)
+    Xmm src;
+    if (i.src1.is_constant) {
+      if (i.src1.value->IsConstantZero()) {
+        e.vmovdqa(i.dest, e.GetXmmConstPtr(XMM3333));
+        return;
+      } else {
+        // TODO(benvanik): check other common constants/perform shuffle/or here.
+        src = e.xmm0;
+        e.LoadConstantXmm(src, i.src1.constant());
+      }
+    } else {
+      src = i.src1;
+    }
+    // Shuffle bytes.
+    e.vpshufb(i.dest, src, e.GetXmmConstPtr(XMMUnpackSHORT_4));
+    // Sign extend words.
+    e.vpslld(i.dest, 16);
+    e.vpsrad(i.dest, 16);
+    // Add 3,3,3,3.
+    e.vpaddd(i.dest, e.GetXmmConstPtr(XMM3333));
+  }
   static void EmitUINT_2101010(X64Emitter& e, const EmitArgType& i) {
     assert_always("not implemented");
   }

src/xenia/cpu/hir/opcodes.h

@@ -77,7 +77,7 @@ enum PackType : uint16_t {
   // Special types:
   PACK_TYPE_D3DCOLOR = 0,
   PACK_TYPE_FLOAT16_2 = 1,
-  PACK_TYPE_FLOAT16_3 = 2,  // FIXME: Not verified, but looks correct.
+  PACK_TYPE_SHORT_4 = 2,
   PACK_TYPE_FLOAT16_4 = 3,
   PACK_TYPE_SHORT_2 = 4,
   PACK_TYPE_UINT_2101010 = 5,

src/xenia/cpu/ppc/ppc_emit_altivec.cc

@@ -2058,6 +2058,9 @@ int InstrEmit_vpkd3d128(PPCHIRBuilder& f, const InstrData& i) {
     case 3:  // VPACK_... 2 FLOAT16s DXGI_FORMAT_R16G16_FLOAT
       v = f.Pack(v, PACK_TYPE_FLOAT16_2);
       break;
+    case 4:  // VPACK_NORMSHORT4
+      v = f.Pack(v, PACK_TYPE_SHORT_4);
+      break;
     case 5:  // VPACK_... 4 FLOAT16s DXGI_FORMAT_R16G16B16A16_FLOAT
       v = f.Pack(v, PACK_TYPE_FLOAT16_4);
       break;
@@ -2158,8 +2161,8 @@ int InstrEmit_vupkd3d128(PPCHIRBuilder& f, const InstrData& i) {
     case 3:  // VPACK_... 2 FLOAT16s DXGI_FORMAT_R16G16_FLOAT
       v = f.Unpack(v, PACK_TYPE_FLOAT16_2);
       break;
-    case 4:
+    case 4:  // VPACK_NORMSHORT4
-      v = f.Unpack(v, PACK_TYPE_FLOAT16_3);
+      v = f.Unpack(v, PACK_TYPE_SHORT_4);
       break;
     case 5:  // VPACK_... 4 FLOAT16s DXGI_FORMAT_R16G16B16A16_FLOAT
       v = f.Unpack(v, PACK_TYPE_FLOAT16_4);

src/xenia/cpu/ppc/testing/instr_vpkd3d128.s

@@ -2,7 +2,9 @@
 # type:
 #   0 = PACK_TYPE_D3DCOLOR
 #   1 = PACK_TYPE_SHORT_2
+#   2 = PACK_TYPE_2_10_10_10
 #   3 = PACK_TYPE_FLOAT16_2
+#   4 = ?
 #   5 = PACK_TYPE_FLOAT16_4
 # mask:
 #   must not be zero
@@ -177,6 +179,15 @@ test_vpkd3d128_short2_2:
   #_ REGISTER_OUT v3 [4040FFFE, 403FF333, 42A23EC8, 403DB757]
   #_ REGISTER_OUT v4 [CDCDCDCD, CDCDCDCD, CDCDCDCD, 7FFFF333]
 
+test_vpkd3d128_short4_0:
+  # v3 = [1.5, -1.5, 1.5, -1.5]
+  #_ REGISTER_IN v3 [403F8001, 403FFFF8, 4040007F, 40400000]
+  #_ REGISTER_IN v4 [CDCDCDCD, CDCDCDCD, CDCDCDCD, CDCDCDCD]
+  # vpkd3d128 v4, v3, 4, 2, 0
+  .long 0x18921E10
+  blr
+  #_ REGISTER_OUT v3 [403F8001, 403FFFF8, 4040007F, 40400000]
+  #_ REGISTER_OUT v4 [CDCDCDCD, CDCDCDCD, 8001FFF8, 007F0000]
 
 test_vpkd3d128_uint_2101010_0:
   #_ REGISTER_IN v3 [B8FF8000, B8FF8000, C04001FF, 4E9A5A5A]
@@ -221,7 +232,6 @@ test_vpkd3d128_float16_2_0:
   #_ REGISTER_OUT v3 [3F000000, BF000000, 00000000, 00000000]
   #_ REGISTER_OUT v4 [CDCDCDCD, CDCDCDCD, CDCDCDCD, 3800B800]
 
-
 test_vpkd3d128_float16_4_invalid_0:
   #_ REGISTER_IN v3 [3FC00000, BFC00000, 3FC00000, BFC00000]
   #_ REGISTER_IN v4 [CDCDCDCD, CDCDCDCD, CDCDCDCD, CDCDCDCD]

src/xenia/cpu/ppc/testing/instr_vupkd3d128.s

@@ -33,6 +33,13 @@ test_vupkd3d128_short2_2:
   blr
   #_ REGISTER_OUT v3 [40407FFF, 403FF333, 00000000, 3f800000]
 
+test_vupkd3d128_short4_0:
+  #_ REGISTER_IN v3 [CDCDCDCD, CDCDCDCD, 7FFFFFFF, 007FFFF8]
+  # vupkd3d128 v3, v3, 4
+  .long 0x18701FF0
+  blr
+  #_ REGISTER_OUT v3 [40407FFF, 403FFFFF, 4040007F, 403FFFF8]
+
 test_vupkd3d128_float16_2_0:
   #_ REGISTER_IN v3 [CDCDCDCD, CDCDCDCD, CDCDCDCD, 3800B800]
   # vupkd3d128 v3, v3, 3