[CPU] vupkd3d: NaN on negative overflow and tests

4 files changed, 41 insertions, 0 deletions

diff --git a/src/xenia/cpu/backend/x64/x64_emitter.cc b/src/xenia/cpu/backend/x64/x64_emitter.cc
index 77361aa54..5642e11d7 100644
--- a/src/xenia/cpu/backend/x64/x64_emitter.cc
+++ b/src/xenia/cpu/backend/x64/x64_emitter.cc
@@ -650,6 +650,7 @@ static const vec128_t xmm_consts[] = {
     vec128i(0xFFFF0F0Eu, 0xFFFF0D0Cu, 0xFFFFFFFFu, 0xFFFFFFFFu),
     /* XMMUnpackSHORT_4       */
     vec128i(0xFFFF0B0Au, 0xFFFF0908u, 0xFFFF0F0Eu, 0xFFFF0D0Cu),
+    /* XMMUnpackSHORT_Overflow */ vec128i(0x403F8000u),
     /* XMMPackUINT_2101010_MinUnpacked */
     vec128i(0x403FFE01u, 0x403FFE01u, 0x403FFE01u, 0x40400000u),
     /* XMMPackUINT_2101010_MaxUnpacked */
@@ -659,6 +660,8 @@ static const vec128_t xmm_consts[] = {
     /* XMMPackUINT_2101010_MaskPacked */
     vec128i(0x3FFu, 0x3FFu << 10, 0x3FFu << 20, 0x3u << 30),
     /* XMMPackUINT_2101010_Shift */ vec128i(0, 10, 20, 30),
+    /* XMMUnpackUINT_2101010_Overflow */ vec128i(0x403FFE00u),
+    /* XMMUnpackOverflowNaN   */ vec128i(0x7FC00000u),
     /* XMMOneOver255          */ vec128f(1.0f / 255.0f),
     /* XMMMaskEvenPI16        */
     vec128i(0x0000FFFFu, 0x0000FFFFu, 0x0000FFFFu, 0x0000FFFFu),
diff --git a/src/xenia/cpu/backend/x64/x64_emitter.h b/src/xenia/cpu/backend/x64/x64_emitter.h
index 479fcc865..33ce2c0a2 100644
--- a/src/xenia/cpu/backend/x64/x64_emitter.h
+++ b/src/xenia/cpu/backend/x64/x64_emitter.h
@@ -78,11 +78,14 @@ enum XmmConst {
   XMMPackSHORT_4,
   XMMUnpackSHORT_2,
   XMMUnpackSHORT_4,
+  XMMUnpackSHORT_Overflow,
   XMMPackUINT_2101010_MinUnpacked,
   XMMPackUINT_2101010_MaxUnpacked,
   XMMPackUINT_2101010_MaskUnpacked,
   XMMPackUINT_2101010_MaskPacked,
   XMMPackUINT_2101010_Shift,
+  XMMUnpackUINT_2101010_Overflow,
+  XMMUnpackOverflowNaN,
   XMMOneOver255,
   XMMMaskEvenPI16,
   XMMShiftMaskEvenPI16,
diff --git a/src/xenia/cpu/backend/x64/x64_sequences.cc b/src/xenia/cpu/backend/x64/x64_sequences.cc
index d0eb88e11..d9cb1cc55 100644
--- a/src/xenia/cpu/backend/x64/x64_sequences.cc
+++ b/src/xenia/cpu/backend/x64/x64_sequences.cc
@@ -7426,6 +7426,9 @@ struct UNPACK : Sequence<UNPACK, I<OPCODE_UNPACK, V128Op, V128Op>> {
     e.vpsrad(i.dest, 16);
     // Add 3,3,0,1.
     e.vpaddd(i.dest, e.GetXmmConstPtr(XMM3301));
+    // Return quiet NaNs in case of negative overflow.
+    e.vcmpeqps(e.xmm0, i.dest, e.GetXmmConstPtr(XMMUnpackSHORT_Overflow));
+    e.vblendvps(i.dest, i.dest, e.GetXmmConstPtr(XMMUnpackOverflowNaN), e.xmm0);
   }
   static void EmitSHORT_4(X64Emitter& e, const EmitArgType& i) {
     // (VD.x) = 3.0 + (VB.x>>16)*2^-22
@@ -7452,6 +7455,9 @@ struct UNPACK : Sequence<UNPACK, I<OPCODE_UNPACK, V128Op, V128Op>> {
     e.vpsrad(i.dest, 16);
     // Add 3,3,3,3.
     e.vpaddd(i.dest, e.GetXmmConstPtr(XMM3333));
+    // Return quiet NaNs in case of negative overflow.
+    e.vcmpeqps(e.xmm0, i.dest, e.GetXmmConstPtr(XMMUnpackSHORT_Overflow));
+    e.vblendvps(i.dest, i.dest, e.GetXmmConstPtr(XMMUnpackOverflowNaN), e.xmm0);
   }
   static void EmitUINT_2101010(X64Emitter& e, const EmitArgType& i) {
     Xmm src;
@@ -7489,6 +7495,10 @@ struct UNPACK : Sequence<UNPACK, I<OPCODE_UNPACK, V128Op, V128Op>> {
     e.vpsrad(i.dest, 22);
     // Add 3,3,3,1.
     e.vpaddd(i.dest, e.GetXmmConstPtr(XMM3331));
+    // Return quiet NaNs in case of negative overflow.
+    e.vcmpeqps(e.xmm0, i.dest,
+               e.GetXmmConstPtr(XMMUnpackUINT_2101010_Overflow));
+    e.vblendvps(i.dest, i.dest, e.GetXmmConstPtr(XMMUnpackOverflowNaN), e.xmm0);
     // To convert XYZ to -1 to 1, games multiply by 0x46004020 & sub 0x46C06030.
     // For W to 0 to 1, they multiply by and subtract 0x4A2AAAAB.
   }
diff --git a/src/xenia/cpu/ppc/testing/instr_vupkd3d128.s b/src/xenia/cpu/ppc/testing/instr_vupkd3d128.s
index cca99605c..b16b299d4 100644
--- a/src/xenia/cpu/ppc/testing/instr_vupkd3d128.s
+++ b/src/xenia/cpu/ppc/testing/instr_vupkd3d128.s
@@ -32,6 +32,12 @@ test_vupkd3d128_short2_2:
   .long 0x18641FF0
   blr
   #_ REGISTER_OUT v3 [40407FFF, 403FF333, 00000000, 3f800000]
+test_vupkd3d128_short2_3:
+  #_ REGISTER_IN v3 [CDCDCDCD, CDCDCDCD, CDCDCDCD, 00008000]
+  # vupkd3d128 v3, v3, 1
+  .long 0x18641FF0
+  blr
+  #_ REGISTER_OUT v3 [40400000, 7FC00000, 00000000, 3f800000]
 
 test_vupkd3d128_short4_0:
   #_ REGISTER_IN v3 [CDCDCDCD, CDCDCDCD, 7FFFFFFF, 007FFFF8]
@@ -53,3 +59,22 @@ test_vupkd3d128_float16_4_0:
   .long 0x18741FF0
   blr
   #_ REGISTER_OUT v3 [3F000000, bf002000, 3f004000, bf006000]
+
+test_vupkd3d128_uint_2101010_0:
+  #_ REGISTER_IN v3 [CDCDCDCD, CDCDCDCD, CDCDCDCD, 400001FF]
+  # vupkd3d128 v3, v3, 2
+  .long 0x18681FF0
+  blr
+  #_ REGISTER_OUT v3 [404001FF, 40400000, 40400000, 3F800001]
+test_vupkd3d128_uint_2101010_1:
+  #_ REGISTER_IN v3 [CDCDCDCD, CDCDCDCD, CDCDCDCD, 40000201]
+  # vupkd3d128 v3, v3, 2
+  .long 0x18681FF0
+  blr
+  #_ REGISTER_OUT v3 [403FFE01, 40400000, 40400000, 3F800001]
+test_vupkd3d128_uint_2101010_2:
+  #_ REGISTER_IN v3 [CDCDCDCD, CDCDCDCD, CDCDCDCD, 40000200]
+  # vupkd3d128 v3, v3, 2
+  .long 0x18681FF0
+  blr
+  #_ REGISTER_OUT v3 [7FC00000, 40400000, 40400000, 3F800001]