mirror of
https://github.com/hrydgard/ppsspp.git
synced 2024-11-23 13:30:02 +00:00
Unify range culling detection
This commit is contained in:
parent
1ae7c0132c
commit
09bcf3ec13
@ -646,6 +646,63 @@ void D3D9Context::InvalidateCachedState() {
|
||||
curPipeline_ = nullptr;
|
||||
}
|
||||
|
||||
// TODO: Move this detection elsewhere when it's needed elsewhere, not before. It's ugly.
|
||||
// Source: https://envytools.readthedocs.io/en/latest/hw/pciid.html#gf100
|
||||
enum NVIDIAGeneration {
|
||||
NV_PRE_KEPLER,
|
||||
NV_KEPLER,
|
||||
NV_MAXWELL,
|
||||
NV_PASCAL,
|
||||
NV_VOLTA,
|
||||
NV_TURING, // or later
|
||||
};
|
||||
|
||||
static NVIDIAGeneration NVIDIAGetDeviceGeneration(int deviceID) {
|
||||
if (deviceID >= 0x1180 && deviceID <= 0x11bf)
|
||||
return NV_KEPLER; // GK104
|
||||
if (deviceID >= 0x11c0 && deviceID <= 0x11fa)
|
||||
return NV_KEPLER; // GK106
|
||||
if (deviceID >= 0x0fc0 && deviceID <= 0x0fff)
|
||||
return NV_KEPLER; // GK107
|
||||
if (deviceID >= 0x1003 && deviceID <= 0x1028)
|
||||
return NV_KEPLER; // GK110(B)
|
||||
if (deviceID >= 0x1280 && deviceID <= 0x12ba)
|
||||
return NV_KEPLER; // GK208
|
||||
if (deviceID >= 0x1381 && deviceID <= 0x13b0)
|
||||
return NV_MAXWELL; // GM107
|
||||
if (deviceID >= 0x1340 && deviceID <= 0x134d)
|
||||
return NV_MAXWELL; // GM108
|
||||
if (deviceID >= 0x13c0 && deviceID <= 0x13d9)
|
||||
return NV_MAXWELL; // GM204
|
||||
if (deviceID >= 0x1401 && deviceID <= 0x1427)
|
||||
return NV_MAXWELL; // GM206
|
||||
if (deviceID >= 0x15f7 && deviceID <= 0x15f9)
|
||||
return NV_PASCAL; // GP100
|
||||
if (deviceID >= 0x15f7 && deviceID <= 0x15f9)
|
||||
return NV_PASCAL; // GP100
|
||||
if (deviceID >= 0x1b00 && deviceID <= 0x1b38)
|
||||
return NV_PASCAL; // GP102
|
||||
if (deviceID >= 0x1b80 && deviceID <= 0x1be1)
|
||||
return NV_PASCAL; // GP104
|
||||
if (deviceID >= 0x1c02 && deviceID <= 0x1c62)
|
||||
return NV_PASCAL; // GP106
|
||||
if (deviceID >= 0x1c81 && deviceID <= 0x1c92)
|
||||
return NV_PASCAL; // GP107
|
||||
if (deviceID >= 0x1d01 && deviceID <= 0x1d12)
|
||||
return NV_PASCAL; // GP108
|
||||
if (deviceID >= 0x1d81 && deviceID <= 0x1dba)
|
||||
return NV_VOLTA; // GV100
|
||||
if (deviceID >= 0x1e02 && deviceID <= 0x1e3c)
|
||||
return NV_TURING; // TU102
|
||||
if (deviceID >= 0x1e82 && deviceID <= 0x1ed0)
|
||||
return NV_TURING; // TU104
|
||||
if (deviceID >= 0x1f02 && deviceID <= 0x1f51)
|
||||
return NV_TURING; // TU104
|
||||
if (deviceID >= 0x1e02)
|
||||
return NV_TURING; // More TU models or later, probably.
|
||||
return NV_PRE_KEPLER;
|
||||
}
|
||||
|
||||
#define FB_DIV 1
|
||||
#define FOURCC_INTZ ((D3DFORMAT)(MAKEFOURCC('I', 'N', 'T', 'Z')))
|
||||
|
||||
@ -704,6 +761,21 @@ D3D9Context::D3D9Context(IDirect3D9 *d3d, IDirect3D9Ex *d3dEx, int adapterId, ID
|
||||
caps_.textureNPOTFullySupported = true;
|
||||
}
|
||||
|
||||
// VS range culling (killing triangles in the vertex shader using NaN) causes problems on Intel.
|
||||
// Also causes problems on old NVIDIA.
|
||||
switch (caps_.vendor) {
|
||||
case Draw::GPUVendor::VENDOR_INTEL:
|
||||
bugs_.Infest(Bugs::BROKEN_NAN_IN_CONDITIONAL);
|
||||
break;
|
||||
case Draw::GPUVendor::VENDOR_NVIDIA:
|
||||
// Older NVIDIAs don't seem to like NaNs in their DX9 vertex shaders.
|
||||
// No idea if KEPLER is the right cutoff, but let's go with it.
|
||||
if (NVIDIAGetDeviceGeneration(caps_.deviceID) < NV_KEPLER) {
|
||||
bugs_.Infest(Bugs::BROKEN_NAN_IN_CONDITIONAL);
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
if (d3d) {
|
||||
D3DDISPLAYMODE displayMode;
|
||||
d3d->GetAdapterDisplayMode(D3DADAPTER_DEFAULT, &displayMode);
|
||||
|
@ -111,19 +111,6 @@ u32 GPU_D3D11::CheckGPUFeatures() const {
|
||||
features |= GPU_SUPPORTS_DEPTH_TEXTURE;
|
||||
if (draw_->GetDeviceCaps().depthClampSupported)
|
||||
features |= GPU_SUPPORTS_DEPTH_CLAMP;
|
||||
if (draw_->GetDeviceCaps().clipDistanceSupported)
|
||||
features |= GPU_SUPPORTS_CLIP_DISTANCE;
|
||||
if (draw_->GetDeviceCaps().cullDistanceSupported)
|
||||
features |= GPU_SUPPORTS_CULL_DISTANCE;
|
||||
if (!draw_->GetBugs().Has(Draw::Bugs::BROKEN_NAN_IN_CONDITIONAL)) {
|
||||
// Ignore the compat setting if clip and cull are both enabled.
|
||||
// When supported, we can do the depth side of range culling more correctly.
|
||||
const bool supported = draw_->GetDeviceCaps().clipDistanceSupported && draw_->GetDeviceCaps().cullDistanceSupported;
|
||||
const bool disabled = PSP_CoreParameter().compat.flags().DisableRangeCulling;
|
||||
if (supported || !disabled) {
|
||||
features |= GPU_SUPPORTS_VS_RANGE_CULLING;
|
||||
}
|
||||
}
|
||||
|
||||
features |= GPU_SUPPORTS_TEXTURE_FLOAT;
|
||||
features |= GPU_SUPPORTS_INSTANCE_RENDERING;
|
||||
|
@ -98,63 +98,6 @@ GPU_DX9::GPU_DX9(GraphicsContext *gfxCtx, Draw::DrawContext *draw)
|
||||
}
|
||||
}
|
||||
|
||||
// TODO: Move this detection elsewhere when it's needed elsewhere, not before. It's ugly.
|
||||
// Source: https://envytools.readthedocs.io/en/latest/hw/pciid.html#gf100
|
||||
enum NVIDIAGeneration {
|
||||
NV_PRE_KEPLER,
|
||||
NV_KEPLER,
|
||||
NV_MAXWELL,
|
||||
NV_PASCAL,
|
||||
NV_VOLTA,
|
||||
NV_TURING, // or later
|
||||
};
|
||||
|
||||
static NVIDIAGeneration NVIDIAGetDeviceGeneration(int deviceID) {
|
||||
if (deviceID >= 0x1180 && deviceID <= 0x11bf)
|
||||
return NV_KEPLER; // GK104
|
||||
if (deviceID >= 0x11c0 && deviceID <= 0x11fa)
|
||||
return NV_KEPLER; // GK106
|
||||
if (deviceID >= 0x0fc0 && deviceID <= 0x0fff)
|
||||
return NV_KEPLER; // GK107
|
||||
if (deviceID >= 0x1003 && deviceID <= 0x1028)
|
||||
return NV_KEPLER; // GK110(B)
|
||||
if (deviceID >= 0x1280 && deviceID <= 0x12ba)
|
||||
return NV_KEPLER; // GK208
|
||||
if (deviceID >= 0x1381 && deviceID <= 0x13b0)
|
||||
return NV_MAXWELL; // GM107
|
||||
if (deviceID >= 0x1340 && deviceID <= 0x134d)
|
||||
return NV_MAXWELL; // GM108
|
||||
if (deviceID >= 0x13c0 && deviceID <= 0x13d9)
|
||||
return NV_MAXWELL; // GM204
|
||||
if (deviceID >= 0x1401 && deviceID <= 0x1427)
|
||||
return NV_MAXWELL; // GM206
|
||||
if (deviceID >= 0x15f7 && deviceID <= 0x15f9)
|
||||
return NV_PASCAL; // GP100
|
||||
if (deviceID >= 0x15f7 && deviceID <= 0x15f9)
|
||||
return NV_PASCAL; // GP100
|
||||
if (deviceID >= 0x1b00 && deviceID <= 0x1b38)
|
||||
return NV_PASCAL; // GP102
|
||||
if (deviceID >= 0x1b80 && deviceID <= 0x1be1)
|
||||
return NV_PASCAL; // GP104
|
||||
if (deviceID >= 0x1c02 && deviceID <= 0x1c62)
|
||||
return NV_PASCAL; // GP106
|
||||
if (deviceID >= 0x1c81 && deviceID <= 0x1c92)
|
||||
return NV_PASCAL; // GP107
|
||||
if (deviceID >= 0x1d01 && deviceID <= 0x1d12)
|
||||
return NV_PASCAL; // GP108
|
||||
if (deviceID >= 0x1d81 && deviceID <= 0x1dba)
|
||||
return NV_VOLTA; // GV100
|
||||
if (deviceID >= 0x1e02 && deviceID <= 0x1e3c)
|
||||
return NV_TURING; // TU102
|
||||
if (deviceID >= 0x1e82 && deviceID <= 0x1ed0)
|
||||
return NV_TURING; // TU104
|
||||
if (deviceID >= 0x1f02 && deviceID <= 0x1f51)
|
||||
return NV_TURING; // TU104
|
||||
if (deviceID >= 0x1e02)
|
||||
return NV_TURING; // More TU models or later, probably.
|
||||
return NV_PRE_KEPLER;
|
||||
}
|
||||
|
||||
u32 GPU_DX9::CheckGPUFeatures() const {
|
||||
u32 features = GPUCommon::CheckGPUFeatures();
|
||||
features |= GPU_SUPPORTS_16BIT_FORMATS;
|
||||
@ -167,24 +110,6 @@ u32 GPU_DX9::CheckGPUFeatures() const {
|
||||
features |= GPU_SUPPORTS_ACCURATE_DEPTH;
|
||||
|
||||
auto vendor = draw_->GetDeviceCaps().vendor;
|
||||
if (!PSP_CoreParameter().compat.flags().DisableRangeCulling) {
|
||||
// VS range culling (killing triangles in the vertex shader using NaN) causes problems on Intel.
|
||||
// Also causes problems on old NVIDIA.
|
||||
switch (vendor) {
|
||||
case Draw::GPUVendor::VENDOR_INTEL:
|
||||
break;
|
||||
case Draw::GPUVendor::VENDOR_NVIDIA:
|
||||
// Older NVIDIAs don't seem to like NaNs in their DX9 vertex shaders.
|
||||
// No idea if KEPLER is the right cutoff, but let's go with it.
|
||||
if (NVIDIAGetDeviceGeneration(draw_->GetDeviceCaps().deviceID) >= NV_KEPLER) {
|
||||
features |= GPU_SUPPORTS_VS_RANGE_CULLING;
|
||||
}
|
||||
break;
|
||||
default:
|
||||
features |= GPU_SUPPORTS_VS_RANGE_CULLING;
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if (!g_Config.bHighQualityDepth) {
|
||||
features |= GPU_SCALE_DEPTH_FROM_24BIT_TO_16BIT;
|
||||
|
@ -166,9 +166,6 @@ u32 GPU_GLES::CheckGPUFeatures() const {
|
||||
if (gl_extensions.GLES3 || !gl_extensions.IsGLES)
|
||||
features |= GPU_SUPPORTS_TEXTURE_LOD_CONTROL;
|
||||
|
||||
if (draw_->GetDeviceCaps().anisoSupported)
|
||||
features |= GPU_SUPPORTS_ANISOTROPY;
|
||||
|
||||
bool canUseInstanceID = gl_extensions.EXT_draw_instanced || gl_extensions.ARB_draw_instanced;
|
||||
bool canDefInstanceID = gl_extensions.IsGLES || gl_extensions.EXT_gpu_shader4 || gl_extensions.VersionGEThan(3, 1);
|
||||
bool instanceRendering = gl_extensions.GLES3 || (canUseInstanceID && canDefInstanceID);
|
||||
@ -189,19 +186,6 @@ u32 GPU_GLES::CheckGPUFeatures() const {
|
||||
}
|
||||
if (draw_->GetDeviceCaps().textureDepthSupported)
|
||||
features |= GPU_SUPPORTS_DEPTH_TEXTURE;
|
||||
if (draw_->GetDeviceCaps().clipDistanceSupported)
|
||||
features |= GPU_SUPPORTS_CLIP_DISTANCE;
|
||||
if (draw_->GetDeviceCaps().cullDistanceSupported)
|
||||
features |= GPU_SUPPORTS_CULL_DISTANCE;
|
||||
if (!draw_->GetBugs().Has(Draw::Bugs::BROKEN_NAN_IN_CONDITIONAL)) {
|
||||
// Ignore the compat setting if clip and cull are both enabled.
|
||||
// When supported, we can do the depth side of range culling more correctly.
|
||||
const bool supported = draw_->GetDeviceCaps().clipDistanceSupported && draw_->GetDeviceCaps().cullDistanceSupported;
|
||||
const bool disabled = PSP_CoreParameter().compat.flags().DisableRangeCulling;
|
||||
if (supported || !disabled) {
|
||||
features |= GPU_SUPPORTS_VS_RANGE_CULLING;
|
||||
}
|
||||
}
|
||||
|
||||
// If we already have a 16-bit depth buffer, we don't need to round.
|
||||
bool prefer24 = draw_->GetDeviceCaps().preferredDepthBufferFormat == Draw::DataFormat::D24_S8;
|
||||
|
@ -3174,6 +3174,24 @@ u32 GPUCommon::CheckGPUFeatures() const {
|
||||
features |= GPU_SUPPORTS_BLEND_MINMAX;
|
||||
}
|
||||
|
||||
if (draw_->GetDeviceCaps().clipDistanceSupported) {
|
||||
features |= GPU_SUPPORTS_CLIP_DISTANCE;
|
||||
}
|
||||
|
||||
if (draw_->GetDeviceCaps().cullDistanceSupported) {
|
||||
features |= GPU_SUPPORTS_CULL_DISTANCE;
|
||||
}
|
||||
|
||||
if (!draw_->GetBugs().Has(Draw::Bugs::BROKEN_NAN_IN_CONDITIONAL)) {
|
||||
// Ignore the compat setting if clip and cull are both enabled.
|
||||
// When supported, we can do the depth side of range culling more correctly.
|
||||
const bool supported = draw_->GetDeviceCaps().clipDistanceSupported && draw_->GetDeviceCaps().cullDistanceSupported;
|
||||
const bool disabled = PSP_CoreParameter().compat.flags().DisableRangeCulling;
|
||||
if (supported || !disabled) {
|
||||
features |= GPU_SUPPORTS_VS_RANGE_CULLING;
|
||||
}
|
||||
}
|
||||
|
||||
if (PSP_CoreParameter().compat.flags().ClearToRAM) {
|
||||
features |= GPU_USE_CLEAR_RAM_HACK;
|
||||
}
|
||||
|
@ -236,22 +236,6 @@ u32 GPU_Vulkan::CheckGPUFeatures() const {
|
||||
if (enabledFeatures.depthClamp) {
|
||||
features |= GPU_SUPPORTS_DEPTH_CLAMP;
|
||||
}
|
||||
if (enabledFeatures.shaderClipDistance) {
|
||||
features |= GPU_SUPPORTS_CLIP_DISTANCE;
|
||||
}
|
||||
if (enabledFeatures.shaderCullDistance) {
|
||||
// Must support at least 8 if feature supported, so we're fine.
|
||||
features |= GPU_SUPPORTS_CULL_DISTANCE;
|
||||
}
|
||||
if (!draw_->GetBugs().Has(Draw::Bugs::BROKEN_NAN_IN_CONDITIONAL)) {
|
||||
// Ignore the compat setting if clip and cull are both enabled.
|
||||
// When supported, we can do the depth side of range culling more correctly.
|
||||
const bool supported = draw_->GetDeviceCaps().clipDistanceSupported && draw_->GetDeviceCaps().cullDistanceSupported;
|
||||
const bool disabled = PSP_CoreParameter().compat.flags().DisableRangeCulling;
|
||||
if (supported || !disabled) {
|
||||
features |= GPU_SUPPORTS_VS_RANGE_CULLING;
|
||||
}
|
||||
}
|
||||
|
||||
// These are VULKAN_4444_FORMAT and friends.
|
||||
uint32_t fmt4444 = draw_->GetDataFormatSupport(Draw::DataFormat::B4G4R4A4_UNORM_PACK16);
|
||||
|
Loading…
Reference in New Issue
Block a user