ShuriZma
/
Cxbx-Reloaded
mirror of https://github.com/Cxbx-Reloaded/Cxbx-Reloaded.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #1683 from LukeUsher/vertex-shader-register-tweaks 

Vertex shader register tweaks
This commit is contained in:
PatrickvL 2019-07-29 11:30:09 +02:00 committed by GitHub
parent ec3f97e4f0 31253815e1
commit f62030a548
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 244 additions and 272 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

64
src/core/hle/D3D8/Direct3D9/Direct3D9.cpp
View File

@ -4168,6 +4168,23 @@ VOID WINAPI XTL::EMUPATCH(D3DDevice_SetVertexData4f)
HRESULT hRet = D3D_OK;
// Get the vertex shader flags (if any is active) :
uint32_t ActiveVertexAttributeFlags = 0;
if (VshHandleIsVertexShader(g_CurrentXboxVertexShaderHandle)) {
LOG_TEST_CASE("D3DDevice_SetVertexData4f with active VertexShader");
X_D3DVertexShader* pXboxVertexShader = VshHandleToXboxVertexShader(g_CurrentXboxVertexShaderHandle);
if (!(pXboxVertexShader->Flags & 0x10/*=X_VERTEXSHADER_PROGRAM*/)) {
ActiveVertexAttributeFlags = pXboxVertexShader->Flags;
}
// If we have an active vertex shader, we also write the input to a vertex shader constant
// This allows us to implement Xbox functionality where SetVertexData4f can be used to specify attributes
// not present in the vertex declaration.
// We use range 193 and up to store these values, as Xbox shaders stop at c192!
FLOAT values[] = {a,b,c,d};
g_pD3DDevice->SetVertexShaderConstantF(X_D3DVS_CONSTREG_VERTEXDATA4F_BASE + Register, values, 1);
}
// Grow g_InlineVertexBuffer_Table to contain at least current, and a potentially next vertex
if (g_InlineVertexBuffer_TableLength <= g_InlineVertexBuffer_TableOffset + 1) {
if (g_InlineVertexBuffer_TableLength == 0) {
@ -4185,16 +4202,6 @@ VOID WINAPI XTL::EMUPATCH(D3DDevice_SetVertexData4f)
// Set first vertex to zero (preventing leaks from prior Begin/End calls)
g_InlineVertexBuffer_Table[0] = {};
// Get the vertex shader flags (if any is active) :
uint32_t ActiveVertexAttributeFlags = 0;
if (VshHandleIsVertexShader(g_CurrentXboxVertexShaderHandle)) {
LOG_TEST_CASE("D3DDevice_SetVertexData4f with active VertexShader");
X_D3DVertexShader *pXboxVertexShader = VshHandleToXboxVertexShader(g_CurrentXboxVertexShaderHandle);
if (!(pXboxVertexShader->Flags & 0x10/*=X_VERTEXSHADER_PROGRAM*/)) {
ActiveVertexAttributeFlags = pXboxVertexShader->Flags;
}
}
// Handle persistent vertex attribute flags, by resetting non-persistent colors
// to their default value (and leaving the persistent colors alone - see the
// "Copy all attributes of the previous vertex" comment below) :
@ -4313,6 +4320,7 @@ VOID WINAPI XTL::EMUPATCH(D3DDevice_SetVertexData4f)
{
g_InlineVertexBuffer_Table[o].Diffuse = D3DCOLOR_COLORVALUE(a, b, c, d);
g_InlineVertexBuffer_FVF |= D3DFVF_DIFFUSE;
HLE_write_NV2A_vertex_attribute_slot(X_D3DVSDE_DIFFUSE, g_InlineVertexBuffer_Table[o].Diffuse);
break;
}
@ -4320,13 +4328,14 @@ VOID WINAPI XTL::EMUPATCH(D3DDevice_SetVertexData4f)
{
g_InlineVertexBuffer_Table[o].Specular = D3DCOLOR_COLORVALUE(a, b, c, d);
g_InlineVertexBuffer_FVF |= D3DFVF_SPECULAR;
HLE_write_NV2A_vertex_attribute_slot(X_D3DVSDE_SPECULAR, g_InlineVertexBuffer_Table[o].Specular);
break;
}
case X_D3DVSDE_FOG: // Xbox extension
{
g_InlineVertexBuffer_Table[o].Fog = a; // TODO : What about the other (b, c and d) arguments?
EmuLog(LOG_LEVEL::WARNING, "Host Direct3D8 doesn''t support FVF FOG");
//EmuLog(LOG_LEVEL::WARNING, "Host Direct3D8 doesn''t support FVF FOG");
break;
}
@ -4335,14 +4344,16 @@ VOID WINAPI XTL::EMUPATCH(D3DDevice_SetVertexData4f)
case X_D3DVSDE_BACKDIFFUSE: // Xbox extension
{
g_InlineVertexBuffer_Table[o].BackDiffuse = D3DCOLOR_COLORVALUE(a, b, c, d);
EmuLog(LOG_LEVEL::WARNING, "Host Direct3D8 doesn''t support FVF BACKDIFFUSE");
//EmuLog(LOG_LEVEL::WARNING, "Host Direct3D8 doesn''t support FVF BACKDIFFUSE");
HLE_write_NV2A_vertex_attribute_slot(X_D3DVSDE_BACKDIFFUSE, g_InlineVertexBuffer_Table[o].BackDiffuse);
break;
}
case X_D3DVSDE_BACKSPECULAR: // Xbox extension
{
g_InlineVertexBuffer_Table[o].BackSpecular = D3DCOLOR_COLORVALUE(a, b, c, d);
EmuLog(LOG_LEVEL::WARNING, "Host Direct3D8 doesn''t support FVF BACKSPECULAR");
//EmuLog(LOG_LEVEL::WARNING, "Host Direct3D8 doesn''t support FVF BACKSPECULAR");
HLE_write_NV2A_vertex_attribute_slot(X_D3DVSDE_BACKSPECULAR, g_InlineVertexBuffer_Table[o].BackSpecular);
break;
}
@ -6487,7 +6498,16 @@ VOID __fastcall XTL::EMUPATCH(D3DDevice_SetRenderState_Simple)
// Just log and return
EmuLog(LOG_LEVEL::DEBUG, "RenderState_Simple(0x%.08X (%s), 0x%.08X) was ignored!", Method, DxbxRenderStateInfo[XboxRenderStateIndex].S, Value);
return;
case X_D3DRS_ALPHATESTENABLE: case X_D3DRS_ALPHABLENDENABLE:
case X_D3DRS_ALPHATESTENABLE:
if (g_LibVersion_D3D8 == 3925) {
// HACK: Many 3925 have missing polygons when this is true
// Until we find out the true underlying cause, and carry on
// Test Cases: Halo, Silent Hill 2.
LOG_TEST_CASE("Applying 3925 alpha test disable hack");
Value = false;
}
break;
case X_D3DRS_ALPHABLENDENABLE:
case X_D3DRS_ALPHAREF: case X_D3DRS_ZWRITEENABLE:
case X_D3DRS_DITHERENABLE: case X_D3DRS_STENCILREF:
case X_D3DRS_STENCILMASK: case X_D3DRS_STENCILWRITEMASK:
@ -7119,6 +7139,15 @@ VOID WINAPI XTL::EMUPATCH(D3DDevice_SetVertexShader)
DEBUG_D3DRESULT(hRet, "g_pD3DDevice->SetVertexDeclaration");
hRet = g_pD3DDevice->SetVertexShader((XTL::IDirect3DVertexShader9*)pVertexShader->Handle);
DEBUG_D3DRESULT(hRet, "g_pD3DDevice->SetVertexShader(VshHandleIsVertexShader)");
// Set default constant values for specular, diffuse, etc
static const float ColorBlack[4] = { 0,0,0,0 };
static const float ColorWhite[4] = { 1,1,1,1 };
g_pD3DDevice->SetVertexShaderConstantF(X_D3DVS_CONSTREG_VERTEXDATA4F_BASE + X_D3DVSDE_DIFFUSE, ColorWhite, 1);
g_pD3DDevice->SetVertexShaderConstantF(X_D3DVS_CONSTREG_VERTEXDATA4F_BASE + X_D3DVSDE_BACKDIFFUSE, ColorWhite, 1);
g_pD3DDevice->SetVertexShaderConstantF(X_D3DVS_CONSTREG_VERTEXDATA4F_BASE + X_D3DVSDE_SPECULAR, ColorBlack, 1);
g_pD3DDevice->SetVertexShaderConstantF(X_D3DVS_CONSTREG_VERTEXDATA4F_BASE + X_D3DVSDE_BACKSPECULAR, ColorBlack, 1);
} else {
hRet = g_pD3DDevice->SetVertexShader(nullptr);
DEBUG_D3DRESULT(hRet, "g_pD3DDevice->SetVertexShader");
@ -7499,7 +7528,12 @@ void XTL::CxbxUpdateNativeD3DResources()
// Some titles set Vertex Shader constants directly via pushbuffers rather than through D3D
// We handle that case by updating any constants that have the dirty flag set on the nv2a.
auto nv2a = g_NV2A->GetDeviceState();
for(int i = 0; i < 192; i++) {
for(int i = 0; i < X_D3DVS_CONSTREG_COUNT; i++) {
// Skip vOffset and vScale constants, we don't want our values to be overwritten by accident
if (i == X_D3DSCM_RESERVED_CONSTANT1_CORRECTED || i == X_D3DSCM_RESERVED_CONSTANT2_CORRECTED) {
continue;
}
if (nv2a->pgraph.vsh_constants_dirty[i]) {
g_pD3DDevice->SetVertexShaderConstantF(i, (float*)&nv2a->pgraph.vsh_constants[i][0], 1);
nv2a->pgraph.vsh_constants_dirty[i] = false;

4
src/core/hle/D3D8/XbD3D8Types.h
View File

@ -1081,6 +1081,10 @@ typedef DWORD X_VERTEXSHADERCONSTANTMODE;
#define X_D3DSCM_CORRECTION 96 // Add 96 to arrive at the range 0..191 (instead of 96..95)
#define X_D3DVS_CONSTREG_COUNT 192
#define X_D3DSCM_RESERVED_CONSTANT1_CORRECTED (X_D3DSCM_RESERVED_CONSTANT1 + X_D3DSCM_CORRECTION)
#define X_D3DSCM_RESERVED_CONSTANT2_CORRECTED (X_D3DSCM_RESERVED_CONSTANT2 + X_D3DSCM_CORRECTION)
#define X_D3DVS_CONSTREG_VERTEXDATA4F_BASE (X_D3DVS_CONSTREG_COUNT + 1)
// Vertex shader types
#define X_VST_NORMAL 1
#define X_VST_READWRITE 2

11
src/core/hle/D3D8/XbState.cpp
View File

@ -103,7 +103,16 @@ void UpdateDeferredRenderStates()
Value = *(DWORD*)&fogValue;
}
} break;
case XTL::X_D3DRS_FOGENABLE:
case XTL::X_D3DRS_FOGENABLE:
if (g_LibVersion_D3D8 == 3925) {
// HACK: Many 3925 games only show a black screen if fog is enabled
// Initially, this was thought to be bad offsets, but it has been verified to be correct
// Until we find out the true underlying cause, disable fog and carry on
// Test Cases: Halo, Silent Hill 2.
LOG_TEST_CASE("Applying 3925 fog disable hack");
Value = false;
}
break;
case XTL::X_D3DRS_FOGTABLEMODE:
case XTL::X_D3DRS_FOGDENSITY:
case XTL::X_D3DRS_RANGEFOGENABLE:

437
src/core/hle/D3D8/XbVertexShader.cpp
View File

@ -486,8 +486,8 @@ static const char* OReg_Name[] =
"a0.x"
};
std::array<bool, 16> RegVUsage;
std::array<int, 16> RegVDeclUsage;
std::array<bool, 16> RegVIsPresentInDeclaration;
std::array<bool, 16> RegVIsUsedByShader;
/* TODO : map non-FVF Xbox vertex shader handle to CxbxVertexShader (a struct containing a host Xbox vertex shader handle and the original members)
std::unordered_map<DWORD, CxbxVertexShader> g_CxbxVertexShaders;
@ -790,28 +790,87 @@ static void VshWriteParameter(VSH_IMD_PARAMETER *pParameter,
}
// From D3D8to9
static const BYTE DeclAddressUsages[][2] =
{
{ XTL::D3DDECLUSAGE_POSITION, 0 },
{ XTL::D3DDECLUSAGE_BLENDWEIGHT, 0 },
{ XTL::D3DDECLUSAGE_BLENDINDICES, 0 },
{ XTL::D3DDECLUSAGE_NORMAL, 0 },
{ XTL::D3DDECLUSAGE_PSIZE, 0 },
{ XTL::D3DDECLUSAGE_COLOR, 0 },
{ XTL::D3DDECLUSAGE_COLOR, 1 },
{ XTL::D3DDECLUSAGE_TEXCOORD, 0 },
{ XTL::D3DDECLUSAGE_TEXCOORD, 1 },
{ XTL::D3DDECLUSAGE_TEXCOORD, 2 },
{ XTL::D3DDECLUSAGE_TEXCOORD, 3 },
{ XTL::D3DDECLUSAGE_TEXCOORD, 4 },
{ XTL::D3DDECLUSAGE_TEXCOORD, 5 },
{ XTL::D3DDECLUSAGE_TEXCOORD, 6 },
{ XTL::D3DDECLUSAGE_TEXCOORD, 7 },
{ XTL::D3DDECLUSAGE_POSITION, 1 },
{ XTL::D3DDECLUSAGE_NORMAL, 1 }
};
#define D3DDECLUSAGE_UNSUPPORTED ((D3DDECLUSAGE)-1)
XTL::D3DDECLUSAGE Xb2PCRegisterType
(
DWORD VertexRegister,
BYTE& PCUsageIndex
)
{
using namespace XTL;
D3DDECLUSAGE PCRegisterType;
PCUsageIndex = 0;
switch (VertexRegister)
{
case X_D3DVSDE_VERTEX: // -1
DbgVshPrintf("D3DVSDE_VERTEX /* xbox ext. */");
PCRegisterType = D3DDECLUSAGE_UNSUPPORTED;
break;
case X_D3DVSDE_POSITION: // 0
DbgVshPrintf("D3DVSDE_POSITION");
PCRegisterType = D3DDECLUSAGE_POSITION;
break;
case X_D3DVSDE_BLENDWEIGHT: // 1
DbgVshPrintf("D3DVSDE_BLENDWEIGHT");
PCRegisterType = D3DDECLUSAGE_BLENDWEIGHT;
break;
case X_D3DVSDE_NORMAL: // 2
DbgVshPrintf("D3DVSDE_NORMAL");
PCRegisterType = D3DDECLUSAGE_NORMAL;
break;
case X_D3DVSDE_DIFFUSE: // 3
DbgVshPrintf("D3DVSDE_DIFFUSE");
PCRegisterType = D3DDECLUSAGE_COLOR; PCUsageIndex = 0;
break;
case X_D3DVSDE_SPECULAR: // 4
DbgVshPrintf("D3DVSDE_SPECULAR");
PCRegisterType = D3DDECLUSAGE_COLOR; PCUsageIndex = 1;
break;
case X_D3DVSDE_FOG: // 5
DbgVshPrintf("D3DVSDE_FOG");
PCRegisterType = D3DDECLUSAGE_FOG;
break;
case X_D3DVSDE_POINTSIZE: // 6
DbgVshPrintf("D3DVDSE_POINTSIZE");
PCRegisterType = D3DDECLUSAGE_PSIZE;
break;
case X_D3DVSDE_BACKDIFFUSE: // 7
DbgVshPrintf("D3DVSDE_BACKDIFFUSE /* xbox ext. */");
PCRegisterType = D3DDECLUSAGE_COLOR; PCUsageIndex = 2;
break;
case X_D3DVSDE_BACKSPECULAR: // 8
DbgVshPrintf("D3DVSDE_BACKSPECULAR /* xbox ext. */");
PCRegisterType = D3DDECLUSAGE_COLOR; PCUsageIndex = 3;
break;
case X_D3DVSDE_TEXCOORD0: // 9
DbgVshPrintf("D3DVSDE_TEXCOORD0");
PCRegisterType = D3DDECLUSAGE_TEXCOORD; PCUsageIndex = 0;
break;
case X_D3DVSDE_TEXCOORD1: // 10
DbgVshPrintf("D3DVSDE_TEXCOORD1");
PCRegisterType = D3DDECLUSAGE_TEXCOORD; PCUsageIndex = 1;
break;
case X_D3DVSDE_TEXCOORD2: // 11
DbgVshPrintf("D3DVSDE_TEXCOORD2");
PCRegisterType = D3DDECLUSAGE_TEXCOORD; PCUsageIndex = 2;
break;
case X_D3DVSDE_TEXCOORD3: // 12
DbgVshPrintf("D3DVSDE_TEXCOORD3");
PCRegisterType = D3DDECLUSAGE_TEXCOORD; PCUsageIndex = 3;
break;
default:
DbgVshPrintf("%d /* unknown register */", VertexRegister);
PCRegisterType = D3DDECLUSAGE_UNSUPPORTED;
break;
}
return PCRegisterType;
}
extern XTL::D3DCAPS g_D3DCaps;
static void VshWriteShader(VSH_XBOX_SHADER *pShader,
std::stringstream& pDisassembly,
@ -841,55 +900,36 @@ static void VshWriteShader(VSH_XBOX_SHADER *pShader,
}
if (Truncate) {
std::stringstream moveConstantsToTemporaries;
pDisassembly << "; Input usage declarations --\n";
unsigned i = 0;
do {
if (RegVUsage[i]) {
DWORD PCUsageIndex = DeclAddressUsages[i][1];
DWORD usage = DeclAddressUsages[i][0];
// If an override exists, use it
if (RegVDeclUsage[i] >= 0) {
usage = RegVDeclUsage[i];
if (RegVIsUsedByShader[i]) {
if (!RegVIsPresentInDeclaration[i]) {
// Log test case and skip
// Any registers hitting this critera were already replaced with constant/temporary reads
// To correctly use the values given in SetVertexData4f.
// We need to move these constant values to temporaries so they can be used as input alongside other constants!
// We count down from the highest available on the host because Xbox titles don't use values that high, and we read from c192 because Xbox uses 192 constants
static int temporaryRegisterBase = g_D3DCaps.VS20Caps.NumTemps - 13;
moveConstantsToTemporaries << "mov r" << (temporaryRegisterBase + i) << ", c" << (X_D3DVS_CONSTREG_VERTEXDATA4F_BASE + i) << "\n";
LOG_TEST_CASE("Shader uses undeclared Vertex Input Registers");
i++;
continue;
}
std::stringstream dclStream;
switch (usage) {
case XTL::D3DDECLUSAGE_POSITION:
dclStream << "dcl_position" << (int)PCUsageIndex;
break;
case XTL::D3DDECLUSAGE_BLENDWEIGHT:
dclStream << "dcl_blendweight";
break;
case XTL::D3DDECLUSAGE_BLENDINDICES:
dclStream << "dcl_blendindices";
break;
case XTL::D3DDECLUSAGE_NORMAL:
dclStream << "dcl_normal";
break;
case XTL::D3DDECLUSAGE_COLOR:
dclStream << "dcl_color" << (int)PCUsageIndex;
break;
case XTL::D3DDECLUSAGE_FOG:
dclStream << "dcl_fog";
break;
case XTL::D3DDECLUSAGE_TEXCOORD:
dclStream << "dcl_texcoord" << (int)PCUsageIndex;
break;
case XTL::D3DDECLUSAGE_PSIZE:
dclStream << "dcl_psize";
break;
default:
dclStream << "dcl_unknown ("<< (int)PCUsageIndex << ")";
LOG_TEST_CASE("Encountered unknown declaration");
break;
}
pDisassembly << dclStream.str() << " v" << i << "\n";
// dcl_texcoord can be useds for any user-defined data
// We need this because there is no reliable way to detect the real usage
// Xbox has no concept of 'usage types', it only requires a list of attribute register numbers.
// So we treat them all as 'user-defined'
pDisassembly << "dcl_texcoord" << i << " v" << i << "\n";
}
i++;
} while (i < RegVUsage.size());
} while (i < RegVIsUsedByShader.size());
pDisassembly << moveConstantsToTemporaries.str();
}
for (int i = 0; i < pShader->IntermediateCount && (i < VSH_MAX_INSTRUCTION_COUNT || !Truncate); i++)
@ -1388,7 +1428,7 @@ static void VshRemoveScreenSpaceInstructions(VSH_XBOX_SHADER *pShader)
// If we couldn't find the generic screen space transformation we're
// assuming that the shader writes direct screen coordinates that must be
// normalized. This hack will fail if (a) the shader uses custom screen
// space transformation, (b) reads r10 or r11 after we have written to
// space transformation, (b) reads r13 or r12 after we have written to
// them, or (c) doesn't reserve c-38 and c-37 for scale and offset.
if(deleted != 3)
{
@ -1401,17 +1441,17 @@ static void VshRemoveScreenSpaceInstructions(VSH_XBOX_SHADER *pShader)
if( pIntermediate->Output.Type == IMD_OUTPUT_O &&
pIntermediate->Output.Address == OREG_OPOS)
{
// Redirect output to r11.
// Redirect output to r12.
pIntermediate->Output.Type = IMD_OUTPUT_R;
pIntermediate->Output.Address = 11;
pIntermediate->Output.Address = 12;
// Scale r11 to r10. (mul r10.[mask], r11, c58)
// Scale r12 to r13. (mul r13.[mask], r12, c58)
VSH_INTERMEDIATE_FORMAT MulIntermediate;
MulIntermediate.IsCombined = FALSE;
MulIntermediate.InstructionType = IMD_MAC;
MulIntermediate.MAC = MAC_MUL;
MulIntermediate.Output.Type = IMD_OUTPUT_R;
MulIntermediate.Output.Address = 10;
MulIntermediate.Output.Address = 13;
MulIntermediate.Output.Mask[0] = pIntermediate->Output.Mask[0];
MulIntermediate.Output.Mask[1] = pIntermediate->Output.Mask[1];
MulIntermediate.Output.Mask[2] = pIntermediate->Output.Mask[2];
@ -1419,7 +1459,7 @@ static void VshRemoveScreenSpaceInstructions(VSH_XBOX_SHADER *pShader)
MulIntermediate.Parameters[0].Active = TRUE;
MulIntermediate.Parameters[0].IndexesWithA0_X = FALSE;
MulIntermediate.Parameters[0].Parameter.ParameterType = PARAM_R;
MulIntermediate.Parameters[0].Parameter.Address = 11;
MulIntermediate.Parameters[0].Parameter.Address = 12;
MulIntermediate.Parameters[0].Parameter.Neg = FALSE;
VshSetSwizzle(&MulIntermediate.Parameters[0], SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W);
MulIntermediate.Parameters[1].Active = TRUE;
@ -1431,13 +1471,13 @@ static void VshRemoveScreenSpaceInstructions(VSH_XBOX_SHADER *pShader)
MulIntermediate.Parameters[2].Active = FALSE;
VshInsertIntermediate(pShader, &MulIntermediate, ++i);
// Add offset with r10 to oPos (add oPos.[mask], r10, c59)
// Add offset with r13 to oPos (add oPos.[mask], r13, c59)
VSH_INTERMEDIATE_FORMAT AddIntermediate = MulIntermediate;
AddIntermediate.MAC = MAC_ADD;
AddIntermediate.Output.Type = IMD_OUTPUT_O;
AddIntermediate.Output.Address = OREG_OPOS;
AddIntermediate.Parameters[0].Parameter.ParameterType = PARAM_R;
AddIntermediate.Parameters[0].Parameter.Address = 10;
AddIntermediate.Parameters[0].Parameter.Address = 13;
AddIntermediate.Parameters[1].Parameter.Address = ConvertCRegister(59);
VshInsertIntermediate(pShader, &AddIntermediate, ++i);
}
@ -1468,11 +1508,15 @@ static boolean VshConvertShader(VSH_XBOX_SHADER *pShader,
{
using namespace XTL;
extern XTL::D3DCAPS g_D3DCaps;
const DWORD temporaryCount = g_D3DCaps.VS20Caps.NumTemps;
boolean RUsage[VSH_MAX_TEMPORARY_REGISTERS] = { FALSE };
// Set the last 13 register to used (they are used for SetVertexData4f Constants)
for (int i = 1; i <= 13; i++) {
RUsage[VSH_MAX_TEMPORARY_REGISTERS - i] = true;
}
// TODO: What about state shaders and such?
pShader->ShaderHeader.Version = VERSION_VS;
@ -1569,7 +1613,16 @@ static boolean VshConvertShader(VSH_XBOX_SHADER *pShader,
}
if (pIntermediate->Parameters[j].Parameter.ParameterType == PARAM_V) {
RegVUsage[pIntermediate->Parameters[j].Parameter.Address] = TRUE;
RegVIsUsedByShader[pIntermediate->Parameters[j].Parameter.Address] = TRUE;
if (!RegVIsPresentInDeclaration[pIntermediate->Parameters[j].Parameter.Address]) {
// This vertex register was not declared and therefore is not present within the Vertex Data object
// We read from temporary registers instead, that are set based on constants, in-turn, set by SetVertexData4f
// We count down from the highest available on the host because Xbox titles don't use values that high, and we read from c192 because Xbox uses 192 constants
static int temporaryRegisterBase = g_D3DCaps.VS20Caps.NumTemps - 13;
pIntermediate->Parameters[j].Parameter.ParameterType = PARAM_R;
pIntermediate->Parameters[j].Parameter.Address += temporaryRegisterBase;
}
}
}
}
@ -1682,69 +1735,35 @@ static boolean VshConvertShader(VSH_XBOX_SHADER *pShader,
i++;
}
}
int16_t R12Replacement = -1;
if(temporaryCount <= 12 && RUsage[12])
{
// Sigh, they absolutely had to use r12, didn't they?
for (int i = temporaryCount - 1; i >= 0; i--)
{
if(!RUsage[i])
{
R12Replacement = i;
break;
}
}
if(R12Replacement == -1)
{
EmuLog(LOG_LEVEL::WARNING, "Vertex shader uses all r registers, including r12; impossible to convert!");
return FALSE;
}
for (int j = 0; j < pShader->IntermediateCount; j++)
{
VSH_INTERMEDIATE_FORMAT* pIntermediate = &pShader->Intermediate[j];
if(pIntermediate->Output.Type == IMD_OUTPUT_O &&
pIntermediate->Output.Address == OREG_OPOS)
{
// Found instruction writing to oPos
pIntermediate->Output.Type = IMD_OUTPUT_R;
pIntermediate->Output.Address = R12Replacement;
}
for (int k = 0; k < 3; k++)
{
if(pIntermediate->Parameters[k].Active)
{
if(pIntermediate->Parameters[k].Parameter.ParameterType == PARAM_R &&
pIntermediate->Parameters[k].Parameter.Address == 12)
{
// Found a r12 used as a parameter; replace
pIntermediate->Parameters[k].Parameter.Address = R12Replacement;
}
else if(pIntermediate->Parameters[k].Parameter.ParameterType == PARAM_C &&
pIntermediate->Parameters[k].Parameter.Address == 58 &&
!pIntermediate->Parameters[k].IndexesWithA0_X)
{
// Found c-38, replace it with r12.w
pIntermediate->Parameters[k].Parameter.ParameterType = PARAM_R;
pIntermediate->Parameters[k].Parameter.Address = R12Replacement;
VshSetSwizzle(&pIntermediate->Parameters[k], SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W);
}
}
}
// Finally, iterate through the shader, replace all writes to oPos with writes to r12
// This fixes shaders that read from r12 expecting to see oPos
for (int i = 0; i < pShader->IntermediateCount; i++) {
VSH_INTERMEDIATE_FORMAT* pIntermediate = &pShader->Intermediate[i];
if (pIntermediate->Output.Type == IMD_OUTPUT_O && pIntermediate->Output.Address == OREG_OPOS) {
pIntermediate->Output.Type = IMD_OUTPUT_R;
pIntermediate->Output.Address = 12;
}
// Insert mov oPos, r## in the end
VSH_INTERMEDIATE_FORMAT *pOPosWriteBack = VshNewIntermediate(pShader);
pOPosWriteBack->InstructionType = IMD_ILU;
pOPosWriteBack->ILU = ILU_MOV;
pOPosWriteBack->MAC = MAC_NOP;
pOPosWriteBack->Output.Type = IMD_OUTPUT_O;
pOPosWriteBack->Output.Address = OREG_OPOS;
VshSetOutputMask(&pOPosWriteBack->Output, TRUE, TRUE, TRUE, TRUE);
pOPosWriteBack->Parameters[0].Active = TRUE;
pOPosWriteBack->Parameters[0].Parameter.ParameterType = PARAM_R;
pOPosWriteBack->Parameters[0].Parameter.Address = R12Replacement;
VshSetSwizzle(&pOPosWriteBack->Parameters[0], SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W);
}
// We append one additional instruction to mov oPos, r12
VSH_INTERMEDIATE_FORMAT MovIntermediate = {0};
MovIntermediate.MAC = MAC_MOV;
MovIntermediate.Output.Type = IMD_OUTPUT_O;
MovIntermediate.Output.Address = OREG_OPOS;
MovIntermediate.Output.Mask[0] = true;
MovIntermediate.Output.Mask[1] = true;
MovIntermediate.Output.Mask[2] = true;
MovIntermediate.Output.Mask[3] = true;
MovIntermediate.Parameters[0].Active = true;
MovIntermediate.Parameters[0].Parameter.ParameterType = PARAM_R;
MovIntermediate.Parameters[0].Parameter.Address = 12;
MovIntermediate.Parameters[0].Parameter.Swizzle[0] = SWIZZLE_X;
MovIntermediate.Parameters[0].Parameter.Swizzle[1] = SWIZZLE_Y;
MovIntermediate.Parameters[0].Parameter.Swizzle[2] = SWIZZLE_Z;
MovIntermediate.Parameters[0].Parameter.Swizzle[3] = SWIZZLE_W;
VshInsertIntermediate(pShader, &MovIntermediate, pShader->IntermediateCount);
return TRUE;
}
@ -1773,93 +1792,6 @@ static DWORD VshGetDeclarationCount(DWORD *pDeclaration)
return Pos + 1;
}
#define D3DDECLUSAGE_UNSUPPORTED ((D3DDECLUSAGE)-1)
XTL::D3DDECLUSAGE Xb2PCRegisterType
(
DWORD VertexRegister,
boolean IsFixedFunction,
BYTE& PCUsageIndex
)
{
using namespace XTL;
D3DDECLUSAGE PCRegisterType;
PCUsageIndex = 0;
if (!IsFixedFunction) {
DbgVshPrintf("%d", VertexRegister);
PCUsageIndex = DeclAddressUsages[VertexRegister][1];
return (D3DDECLUSAGE)DeclAddressUsages[VertexRegister][0];
}
switch (VertexRegister)
{
case X_D3DVSDE_VERTEX: // -1
DbgVshPrintf("D3DVSDE_VERTEX /* xbox ext. */");
PCRegisterType = D3DDECLUSAGE_UNSUPPORTED;
break;
case X_D3DVSDE_POSITION: // 0
DbgVshPrintf("D3DVSDE_POSITION");
PCRegisterType = D3DDECLUSAGE_POSITION;
break;
case X_D3DVSDE_BLENDWEIGHT: // 1
DbgVshPrintf("D3DVSDE_BLENDWEIGHT");
PCRegisterType = D3DDECLUSAGE_BLENDWEIGHT;
break;
case X_D3DVSDE_NORMAL: // 2
DbgVshPrintf("D3DVSDE_NORMAL");
PCRegisterType = D3DDECLUSAGE_NORMAL;
break;
case X_D3DVSDE_DIFFUSE: // 3
DbgVshPrintf("D3DVSDE_DIFFUSE");
PCRegisterType = D3DDECLUSAGE_COLOR; PCUsageIndex = 0;
break;
case X_D3DVSDE_SPECULAR: // 4
DbgVshPrintf("D3DVSDE_SPECULAR");
PCRegisterType = D3DDECLUSAGE_COLOR; PCUsageIndex = 1;
break;
case X_D3DVSDE_FOG: // 5
DbgVshPrintf("D3DVSDE_FOG");
PCRegisterType = D3DDECLUSAGE_FOG;
break;
case X_D3DVSDE_POINTSIZE: // 6
DbgVshPrintf("D3DVDSE_POINTSIZE");
PCRegisterType = D3DDECLUSAGE_PSIZE;
break;
case X_D3DVSDE_BACKDIFFUSE: // 7
DbgVshPrintf("D3DVSDE_BACKDIFFUSE /* xbox ext. */");
PCRegisterType = D3DDECLUSAGE_COLOR; PCUsageIndex = 2;
break;
case X_D3DVSDE_BACKSPECULAR: // 8
DbgVshPrintf("D3DVSDE_BACKSPECULAR /* xbox ext. */");
PCRegisterType = D3DDECLUSAGE_COLOR; PCUsageIndex = 3;
break;
case X_D3DVSDE_TEXCOORD0: // 9
DbgVshPrintf("D3DVSDE_TEXCOORD0");
PCRegisterType = D3DDECLUSAGE_TEXCOORD; PCUsageIndex = 0;
break;
case X_D3DVSDE_TEXCOORD1: // 10
DbgVshPrintf("D3DVSDE_TEXCOORD1");
PCRegisterType = D3DDECLUSAGE_TEXCOORD; PCUsageIndex = 1;
break;
case X_D3DVSDE_TEXCOORD2: // 11
DbgVshPrintf("D3DVSDE_TEXCOORD2");
PCRegisterType = D3DDECLUSAGE_TEXCOORD; PCUsageIndex = 2;
break;
case X_D3DVSDE_TEXCOORD3: // 12
DbgVshPrintf("D3DVSDE_TEXCOORD3");
PCRegisterType = D3DDECLUSAGE_TEXCOORD; PCUsageIndex = 3;
break;
default:
DbgVshPrintf("%d /* unknown register */", VertexRegister);
PCRegisterType = D3DDECLUSAGE_UNSUPPORTED;
break;
}
return PCRegisterType;
}
static inline DWORD VshGetTokenType(DWORD Token)
{
return (Token & X_D3DVSD_TOKENTYPEMASK) >> X_D3DVSD_TOKENTYPESHIFT;
@ -1921,19 +1853,18 @@ static void VshConvertToken_TESSELATOR(
using namespace XTL;
BYTE Index;
// TODO: Investigate why Xb2PCRegisterType is only used for fixed function vertex shaders
if(*pToken & X_D3DVSD_MASK_TESSUV)
{
XTL::DWORD VertexRegister = VshGetVertexRegister(*pToken);
XTL::DWORD NewVertexRegister = VertexRegister;
DbgVshPrintf("\tD3DVSD_TESSUV(");
NewVertexRegister = Xb2PCRegisterType(VertexRegister, IsFixedFunction, Index);
NewVertexRegister = Xb2PCRegisterType(VertexRegister, Index);
DbgVshPrintf("),\n");
// TODO : Expand on the setting of this TESSUV register element :
pRecompiled->Usage = D3DDECLUSAGE(NewVertexRegister);
pRecompiled->UsageIndex = Xb2PCRegisterType(NewVertexRegister, IsFixedFunction, Index); // TODO : Get Index from Xb2PCRegisterType
pRecompiled->UsageIndex =Index;
}
// D3DVSD_TESSNORMAL
else
@ -1945,18 +1876,18 @@ static void VshConvertToken_TESSELATOR(
XTL::DWORD NewVertexRegisterOut = VertexRegisterOut;
DbgVshPrintf("\tD3DVSD_TESSNORMAL(");
NewVertexRegisterIn = Xb2PCRegisterType(VertexRegisterIn, IsFixedFunction, Index);
NewVertexRegisterIn = Xb2PCRegisterType(VertexRegisterIn, Index);
DbgVshPrintf(", ");
// TODO : Expand on the setting of this TESSNORMAL input register element :
pRecompiled->Usage = D3DDECLUSAGE(NewVertexRegisterIn);
pRecompiled->UsageIndex = 0; // TODO : Get Index from Xb2PCRegisterType
NewVertexRegisterOut = Xb2PCRegisterType(VertexRegisterOut, IsFixedFunction, Index);
NewVertexRegisterOut = Xb2PCRegisterType(VertexRegisterOut, Index);
DbgVshPrintf("),\n");
// TODO : Expand on the setting of this TESSNORMAL output register element :
pRecompiled++;
pRecompiled->Usage = D3DDECLUSAGE(NewVertexRegisterOut);
pRecompiled->UsageIndex = 0; // TODO : Get Index from Xb2PCRegisterType
pRecompiled->UsageIndex = Index; // TODO : Get Index from Xb2PCRegisterType
}
}
@ -2045,18 +1976,30 @@ static void VshConvertToken_STREAMDATA_REG(
{
using namespace XTL;
extern XTL::D3DCAPS g_D3DCaps;
XTL::DWORD VertexRegister = VshGetVertexRegister(*pToken);
XTL::BYTE HostVertexRegister;
XTL::BOOL NeedPatching = FALSE;
XTL::BYTE Index;
BYTE HostVertexRegisterType;
// If this is a fixed-function shader, use Xb2PCRegisterType
DbgVshPrintf("\t\tD3DVSD_REG(");
BYTE XboxVertexRegister = Xb2PCRegisterType(VertexRegister, IsFixedFunction, Index);
HostVertexRegister = XboxVertexRegister;
if (IsFixedFunction) {
HostVertexRegisterType = Xb2PCRegisterType(VertexRegister, Index);
} else {
// D3DDECLUSAGE_TEXCOORD can be useds for any user-defined data
// We need this because there is no reliable way to detect the real usage
// Xbox has no concept of 'usage types', it only requires a list of attribute register numbers.
// So we treat them all as 'user-defined' with an Index of the Vertex Register Index
// this prevents information loss in shaders due to non-matching dcl types!
HostVertexRegisterType = D3DDECLUSAGE_TEXCOORD;
Index = VertexRegister;
DbgVshPrintf("%d", Index);
}
DbgVshPrintf(", ");
// Add this register to the list of declared registers
RegVIsPresentInDeclaration[VertexRegister] = true;
XTL::DWORD XboxVertexElementDataType = (*pToken & X_D3DVSD_DATATYPEMASK) >> X_D3DVSD_DATATYPESHIFT;
XTL::BYTE HostVertexElementDataType = 0;
XTL::WORD HostVertexElementByteSize = 0;
@ -2072,32 +2015,21 @@ static void VshConvertToken_STREAMDATA_REG(
DbgVshPrintf("D3DVSDT_FLOAT2");
HostVertexElementDataType = D3DDECLTYPE_FLOAT2;
HostVertexElementByteSize = 2 * sizeof(FLOAT);
//HostVertexRegister = D3DDECLUSAGE_TEXCOORD;
break;
case X_D3DVSDT_FLOAT3: // 0x32:
DbgVshPrintf("D3DVSDT_FLOAT3");
HostVertexElementDataType = D3DDECLTYPE_FLOAT3;
HostVertexElementByteSize = 3 * sizeof(FLOAT);
/*
if (pPatchData->pCurrentVertexShaderStreamInfo->DeclPosition) {
pPatchData->pCurrentVertexShaderStreamInfo->DeclPosition = true;
HostVertexRegister = D3DDECLUSAGE_POSITION;
} else {
HostVertexRegister = D3DDECLUSAGE_NORMAL;
} */
break;
case X_D3DVSDT_FLOAT4: // 0x42:
DbgVshPrintf("D3DVSDT_FLOAT4");
HostVertexElementDataType = D3DDECLTYPE_FLOAT4;
HostVertexElementByteSize = 4 * sizeof(FLOAT);
//HostVertexRegister = D3DDECLUSAGE_COLOR;
break;
case X_D3DVSDT_D3DCOLOR: // 0x40:
DbgVshPrintf("D3DVSDT_D3DCOLOR");
HostVertexElementDataType = D3DDECLTYPE_D3DCOLOR;
HostVertexElementByteSize = 1 * sizeof(D3DCOLOR);
//HostVertexRegister = D3DDECLUSAGE_COLOR;
break;
case X_D3DVSDT_SHORT2: // 0x25:
DbgVshPrintf("D3DVSDT_SHORT2");
@ -2272,14 +2204,9 @@ static void VshConvertToken_STREAMDATA_REG(
pRecompiled->Offset = pPatchData->pCurrentVertexShaderStreamInfo->HostVertexStride;
pRecompiled->Type = HostVertexElementDataType;
pRecompiled->Method = D3DDECLMETHOD_DEFAULT;
pRecompiled->Usage = HostVertexRegister;
pRecompiled->Usage = HostVertexRegisterType;
pRecompiled->UsageIndex = Index;
// If the xbox and host register number and usage differ, store an override!
if (XboxVertexRegister != HostVertexRegister) {
RegVDeclUsage[XboxVertexRegister] = HostVertexRegister;
}
pRecompiled++;
pPatchData->pCurrentVertexShaderStreamInfo->HostVertexStride += HostVertexElementByteSize;
@ -2363,7 +2290,7 @@ DWORD XTL::EmuRecompileVshDeclaration
CxbxVertexShaderInfo *pVertexShaderInfo
)
{
RegVDeclUsage.fill(-1);
RegVIsPresentInDeclaration.fill(false);
// First of all some info:
// We have to figure out which flags are set and then
@ -2413,8 +2340,6 @@ DWORD XTL::EmuRecompileVshDeclaration
return D3D_OK;
}
extern XTL::D3DCAPS g_D3DCaps;
// recompile xbox vertex shader function
extern HRESULT XTL::EmuRecompileVshFunction
(
@ -2444,9 +2369,9 @@ extern HRESULT XTL::EmuRecompileVshFunction
DWORD* pDeclEnd = (DWORD*)((BYTE*)pDeclToken + DeclarationSize);
do {
DWORD regNum = *pDeclToken & X_D3DVSD_VERTEXREGMASK;
if (regNum >= temporaryCount /*12*/) {
if (regNum >= temporaryCount /*12 for D3D8, D3D9 value depends on host GPU */) {
// Lego Star Wars hits this
LOG_TEST_CASE("RegNum > 12");
LOG_TEST_CASE("RegNum > NumTemps");
pDeclToken++;
continue;
}
@ -2489,7 +2414,7 @@ extern HRESULT XTL::EmuRecompileVshFunction
if(SUCCEEDED(hRet))
{
RegVUsage.fill(false);
RegVIsUsedByShader.fill(false);
for (pToken = (DWORD*)((uint8_t*)pFunction + sizeof(VSH_SHADER_HEADER)); !EOI; pToken += VSH_INSTRUCTION_SIZE)
{