From 7d08a54ad9412d92757c5c4193b8f4c24e212d95 Mon Sep 17 00:00:00 2001
From: Stenzek <stenzek@gmail.com>
Date: Tue, 31 Jan 2023 20:50:45 +1000
Subject: [PATCH] GS/HW: Optimize TC source size based on CLAMP
---
bin/resources/shaders/dx11/tfx.fx | 39 +-
.../shaders/opengl/common_header.glsl | 8 +-
bin/resources/shaders/opengl/tfx_fs.glsl | 33 +-
bin/resources/shaders/vulkan/tfx.glsl | 40 +-
pcsx2/GS/GSDrawingContext.cpp | 49 +--
pcsx2/GS/GSDrawingContext.h | 9 -
pcsx2/GS/GSRegs.h | 1 +
pcsx2/GS/GSState.cpp | 8 +-
pcsx2/GS/Renderers/Common/GSDevice.h | 5 +-
pcsx2/GS/Renderers/DX11/GSTextureFX11.cpp | 3 +-
pcsx2/GS/Renderers/DX12/GSDevice12.cpp | 3 +-
pcsx2/GS/Renderers/HW/GSRendererHW.cpp | 81 ++--
pcsx2/GS/Renderers/HW/GSTextureCache.cpp | 351 ++++++++++++------
pcsx2/GS/Renderers/HW/GSTextureCache.h | 58 ++-
.../GS/Renderers/HW/GSTextureReplacements.cpp | 130 +++++--
pcsx2/GS/Renderers/HW/GSTextureReplacements.h | 3 +-
pcsx2/GS/Renderers/Metal/GSDeviceMTL.mm | 5 +-
pcsx2/GS/Renderers/Metal/GSMTLSharedHeader.h | 11 +-
pcsx2/GS/Renderers/Metal/tfx.metal | 28 +-
pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp | 3 +-
pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp | 3 +-
pcsx2/ShaderCacheVersion.h | 2 +-
22 files changed, 554 insertions(+), 319 deletions(-)
diff --git a/bin/resources/shaders/dx11/tfx.fx b/bin/resources/shaders/dx11/tfx.fx
index 518185aade..ab15f78c95 100644
--- a/bin/resources/shaders/dx11/tfx.fx
+++ b/bin/resources/shaders/dx11/tfx.fx
@@ -21,6 +21,8 @@
#define PS_FST 0
#define PS_WMS 0
#define PS_WMT 0
+#define PS_ADJS 0
+#define PS_ADJT 0
#define PS_AEM_FMT FMT_32
#define PS_AEM 0
#define PS_TFX 0
@@ -42,7 +44,6 @@
#define PS_CHANNEL_FETCH 0
#define PS_TALES_OF_ABYSS_HLE 0
#define PS_URBAN_CHAOS_HLE 0
-#define PS_INVALID_TEX0 0
#define PS_SCALE_FACTOR 1.0
#define PS_HDR 0
#define PS_COLCLIP 0
@@ -158,10 +159,10 @@ cbuffer cb1
float2 TA;
float MaxDepthPS;
float Af;
- uint4 MskFix;
uint4 FbMask;
float4 HalfTexel;
float4 MinMax;
+ float4 STRange;
int4 ChannelShuffle;
float2 TC_OffsetHack;
float2 STScale;
@@ -183,7 +184,20 @@ float4 sample_c(float2 uv, float uv_w)
// As of 2018 this issue is still present.
uv = (trunc(uv * WH.zw) + float2(0.5, 0.5)) / WH.zw;
}
+#if !PS_ADJS && !PS_ADJT
uv *= STScale;
+#else
+ #if PS_ADJS
+ uv.x = (uv.x - STRange.x) * STRange.z;
+ #else
+ uv.x = uv.x * STScale.x;
+ #endif
+ #if PS_ADJT
+ uv.y = (uv.y - STRange.y) * STRange.w;
+ #else
+ uv.y = uv.y * STScale.y;
+ #endif
+#endif
#if PS_AUTOMATIC_LOD == 1
return Texture.Sample(TextureSampler, uv);
@@ -218,12 +232,7 @@ float4 sample_p_norm(float u)
float4 clamp_wrap_uv(float4 uv)
{
- float4 tex_size;
-
- if (PS_INVALID_TEX0 == 1)
- tex_size = WH.zwzw;
- else
- tex_size = WH.xyxy;
+ float4 tex_size = WH.xyxy;
if(PS_WMS == PS_WMT)
{
@@ -238,7 +247,7 @@ float4 clamp_wrap_uv(float4 uv)
// textures. Fixes Xenosaga's hair issue.
uv = frac(uv);
#endif
- uv = (float4)(((uint4)(uv * tex_size) & MskFix.xyxy) | MskFix.zwzw) / tex_size;
+ uv = (float4)(((uint4)(uv * tex_size) & asuint(MinMax.xyxy)) | asuint(MinMax.zwzw)) / tex_size;
}
}
else
@@ -252,7 +261,7 @@ float4 clamp_wrap_uv(float4 uv)
#if PS_FST == 0
uv.xz = frac(uv.xz);
#endif
- uv.xz = (float2)(((uint2)(uv.xz * tex_size.xx) & MskFix.xx) | MskFix.zz) / tex_size.xx;
+ uv.xz = (float2)(((uint2)(uv.xz * tex_size.xx) & asuint(MinMax.xx)) | asuint(MinMax.zz)) / tex_size.xx;
}
if(PS_WMT == 2)
{
@@ -263,7 +272,7 @@ float4 clamp_wrap_uv(float4 uv)
#if PS_FST == 0
uv.yw = frac(uv.yw);
#endif
- uv.yw = (float2)(((uint2)(uv.yw * tex_size.yy) & MskFix.yy) | MskFix.ww) / tex_size.yy;
+ uv.yw = (float2)(((uint2)(uv.yw * tex_size.yy) & asuint(MinMax.yy)) | asuint(MinMax.ww)) / tex_size.yy;
}
}
@@ -353,7 +362,7 @@ float4 fetch_c(int2 uv)
int2 clamp_wrap_uv_depth(int2 uv)
{
- int4 mask = (int4)MskFix << 4;
+ int4 mask = asint(MinMax) << 4;
if (PS_WMS == PS_WMT)
{
if (PS_WMS == 2)
@@ -676,11 +685,7 @@ float4 fog(float4 c, float f)
float4 ps_color(PS_INPUT input)
{
-#if PS_FST == 0 && PS_INVALID_TEX0 == 1
- // Re-normalize coordinate from invalid GS to corrected texture size
- float2 st = (input.t.xy * WH.xy) / (input.t.w * WH.zw);
- float2 st_int = (input.ti.zw * WH.xy) / (input.t.w * WH.zw);
-#elif PS_FST == 0
+#if PS_FST == 0
float2 st = input.t.xy / input.t.w;
float2 st_int = input.ti.zw / input.t.w;
#else
diff --git a/bin/resources/shaders/opengl/common_header.glsl b/bin/resources/shaders/opengl/common_header.glsl
index ac08be64d6..9a7ce8589c 100644
--- a/bin/resources/shaders/opengl/common_header.glsl
+++ b/bin/resources/shaders/opengl/common_header.glsl
@@ -75,13 +75,12 @@ layout(std140, binding = 0) uniform cb21
float MaxDepthPS;
float Af;
- uvec4 MskFix;
-
uvec4 FbMask;
vec4 HalfTexel;
vec4 MinMax;
+ vec4 STRange;
ivec4 ChannelShuffle;
@@ -92,11 +91,6 @@ layout(std140, binding = 0) uniform cb21
};
#endif
-//layout(std140, binding = 22) uniform cb22
-//{
-// vec4 rt_size;
-//};
-
//////////////////////////////////////////////////////////////////////
// Default Sampler
//////////////////////////////////////////////////////////////////////
diff --git a/bin/resources/shaders/opengl/tfx_fs.glsl b/bin/resources/shaders/opengl/tfx_fs.glsl
index 669cd34448..2f0a82aef2 100644
--- a/bin/resources/shaders/opengl/tfx_fs.glsl
+++ b/bin/resources/shaders/opengl/tfx_fs.glsl
@@ -109,7 +109,20 @@ vec4 sample_c(vec2 uv)
// As of 2018 this issue is still present.
uv = (trunc(uv * WH.zw) + vec2(0.5, 0.5)) / WH.zw;
#endif
- uv *= STScale;
+#if !PS_ADJS && !PS_ADJT
+ uv *= STScale;
+#else
+ #if PS_ADJS
+ uv.x = (uv.x - STRange.x) * STRange.z;
+ #else
+ uv.x = uv.x * STScale.x;
+ #endif
+ #if PS_ADJT
+ uv.y = (uv.y - STRange.y) * STRange.w;
+ #else
+ uv.y = uv.y * STScale.y;
+ #endif
+#endif
#if PS_AUTOMATIC_LOD == 1
return texture(TextureSampler, uv);
@@ -146,11 +159,7 @@ vec4 sample_p_norm(float u)
vec4 clamp_wrap_uv(vec4 uv)
{
vec4 uv_out = uv;
-#if PS_INVALID_TEX0 == 1
- vec4 tex_size = WH.zwzw;
-#else
vec4 tex_size = WH.xyxy;
-#endif
#if PS_WMS == PS_WMT
@@ -162,7 +171,7 @@ vec4 clamp_wrap_uv(vec4 uv)
// textures. Fixes Xenosaga's hair issue.
uv = fract(uv);
#endif
- uv_out = vec4((uvec4(uv * tex_size) & MskFix.xyxy) | MskFix.zwzw) / tex_size;
+ uv_out = vec4((uvec4(uv * tex_size) & floatBitsToUint(MinMax.xyxy)) | floatBitsToUint(MinMax.zwzw)) / tex_size;
#endif
#else // PS_WMS != PS_WMT
@@ -174,7 +183,7 @@ vec4 clamp_wrap_uv(vec4 uv)
#if PS_FST == 0
uv.xz = fract(uv.xz);
#endif
- uv_out.xz = vec2((uvec2(uv.xz * tex_size.xx) & MskFix.xx) | MskFix.zz) / tex_size.xx;
+ uv_out.xz = vec2((uvec2(uv.xz * tex_size.xx) & floatBitsToUint(MinMax.xx)) | floatBitsToUint(MinMax.zz)) / tex_size.xx;
#endif
@@ -185,7 +194,7 @@ vec4 clamp_wrap_uv(vec4 uv)
#if PS_FST == 0
uv.yw = fract(uv.yw);
#endif
- uv_out.yw = vec2((uvec2(uv.yw * tex_size.yy) & MskFix.yy) | MskFix.ww) / tex_size.yy;
+ uv_out.yw = vec2((uvec2(uv.yw * tex_size.yy) & floatBitsToUint(MinMax.yy)) | floatBitsToUint(MinMax.ww)) / tex_size.yy;
#endif
#endif
@@ -288,7 +297,7 @@ ivec2 clamp_wrap_uv_depth(ivec2 uv)
// Keep the full precision
// It allow to multiply the ScalingFactor before the 1/16 coeff
- ivec4 mask = ivec4(MskFix) << 4;
+ ivec4 mask = floatBitsToInt(MinMax) << 4;
#if PS_WMS == PS_WMT
@@ -591,11 +600,7 @@ void fog(inout vec4 C, float f)
vec4 ps_color()
{
//FIXME: maybe we can set gl_Position.w = q in VS
-#if (PS_FST == 0) && (PS_INVALID_TEX0 == 1)
- // Re-normalize coordinate from invalid GS to corrected texture size
- vec2 st = (PSin.t_float.xy * WH.xy) / (vec2(PSin.t_float.w) * WH.zw);
- vec2 st_int = (PSin.t_int.zw * WH.xy) / (vec2(PSin.t_float.w) * WH.zw);
-#elif (PS_FST == 0)
+#if (PS_FST == 0)
vec2 st = PSin.t_float.xy / vec2(PSin.t_float.w);
vec2 st_int = PSin.t_int.zw / vec2(PSin.t_float.w);
#else
diff --git a/bin/resources/shaders/vulkan/tfx.glsl b/bin/resources/shaders/vulkan/tfx.glsl
index 1f588a804e..bdb336f0eb 100644
--- a/bin/resources/shaders/vulkan/tfx.glsl
+++ b/bin/resources/shaders/vulkan/tfx.glsl
@@ -312,6 +312,8 @@ void main()
#define PS_FST 0
#define PS_WMS 0
#define PS_WMT 0
+#define PS_ADJS 0
+#define PS_ADJT 0
#define PS_FMT FMT_32
#define PS_AEM 0
#define PS_TFX 0
@@ -332,7 +334,6 @@ void main()
#define PS_CHANNEL_FETCH 0
#define PS_TALES_OF_ABYSS_HLE 0
#define PS_URBAN_CHAOS_HLE 0
-#define PS_INVALID_TEX0 0
#define PS_SCALE_FACTOR 1.0
#define PS_HDR 0
#define PS_COLCLIP 0
@@ -361,10 +362,10 @@ layout(std140, set = 0, binding = 1) uniform cb1
vec2 TA;
float MaxDepthPS;
float Af;
- uvec4 MskFix;
uvec4 FbMask;
vec4 HalfTexel;
vec4 MinMax;
+ vec4 STRange;
ivec4 ChannelShuffle;
vec2 TC_OffsetHack;
vec2 STScale;
@@ -420,7 +421,20 @@ vec4 sample_c(vec2 uv)
// As of 2018 this issue is still present.
uv = (trunc(uv * WH.zw) + vec2(0.5, 0.5)) / WH.zw;
#endif
+#if !PS_ADJS && !PS_ADJT
uv *= STScale;
+#else
+ #if PS_ADJS
+ uv.x = (uv.x - STRange.x) * STRange.z;
+ #else
+ uv.x = uv.x * STScale.x;
+ #endif
+ #if PS_ADJT
+ uv.y = (uv.y - STRange.y) * STRange.w;
+ #else
+ uv.y = uv.y * STScale.y;
+ #endif
+#endif
#if PS_AUTOMATIC_LOD == 1
return texture(Texture, uv);
@@ -455,13 +469,7 @@ vec4 sample_p_norm(float u)
vec4 clamp_wrap_uv(vec4 uv)
{
- vec4 tex_size;
-
- #if PS_INVALID_TEX0
- tex_size = WH.zwzw;
- #else
- tex_size = WH.xyxy;
- #endif
+ vec4 tex_size = WH.xyxy;
#if PS_WMS == PS_WMT
{
@@ -476,7 +484,7 @@ vec4 clamp_wrap_uv(vec4 uv)
// textures. Fixes Xenosaga's hair issue.
uv = fract(uv);
#endif
- uv = vec4((uvec4(uv * tex_size) & MskFix.xyxy) | MskFix.zwzw) / tex_size;
+ uv = vec4((uvec4(uv * tex_size) & floatBitsToUint(MinMax.xyxy)) | floatBitsToUint(MinMax.zwzw)) / tex_size;
}
#endif
}
@@ -491,7 +499,7 @@ vec4 clamp_wrap_uv(vec4 uv)
#if PS_FST == 0
uv.xz = fract(uv.xz);
#endif
- uv.xz = vec2((uvec2(uv.xz * tex_size.xx) & MskFix.xx) | MskFix.zz) / tex_size.xx;
+ uv.xz = vec2((uvec2(uv.xz * tex_size.xx) & floatBitsToUint(MinMax.xx)) | floatBitsToUint(MinMax.zz)) / tex_size.xx;
}
#endif
#if PS_WMT == 2
@@ -503,7 +511,7 @@ vec4 clamp_wrap_uv(vec4 uv)
#if PS_FST == 0
uv.yw = fract(uv.yw);
#endif
- uv.yw = vec2((uvec2(uv.yw * tex_size.yy) & MskFix.yy) | MskFix.ww) / tex_size.yy;
+ uv.yw = vec2((uvec2(uv.yw * tex_size.yy) & floatBitsToUint(MinMax.yy)) | floatBitsToUint(MinMax.ww)) / tex_size.yy;
}
#endif
}
@@ -590,7 +598,7 @@ vec4 fetch_c(ivec2 uv)
ivec2 clamp_wrap_uv_depth(ivec2 uv)
{
- ivec4 mask = ivec4(MskFix << 4);
+ ivec4 mask = floatBitsToInt(MinMax) << 4;
#if (PS_WMS == PS_WMT)
{
#if (PS_WMS == 2)
@@ -907,11 +915,7 @@ vec4 fog(vec4 c, float f)
vec4 ps_color()
{
-#if PS_FST == 0 && PS_INVALID_TEX0 == 1
- // Re-normalize coordinate from invalid GS to corrected texture size
- vec2 st = (vsIn.t.xy * WH.xy) / (vsIn.t.w * WH.zw);
- vec2 st_int = (vsIn.ti.zw * WH.xy) / (vsIn.t.w * WH.zw);
-#elif PS_FST == 0
+#if PS_FST == 0
vec2 st = vsIn.t.xy / vsIn.t.w;
vec2 st_int = vsIn.ti.zw / vsIn.t.w;
#else
diff --git a/pcsx2/GS/GSDrawingContext.cpp b/pcsx2/GS/GSDrawingContext.cpp
index 522185836a..011a664781 100644
--- a/pcsx2/GS/GSDrawingContext.cpp
+++ b/pcsx2/GS/GSDrawingContext.cpp
@@ -130,7 +130,7 @@ GIFRegTEX0 GSDrawingContext::GetSizeFixedTEX0(const GSVector4& st, bool linear,
res.TW = tw > 10 ? 0 : tw;
res.TH = th > 10 ? 0 : th;
- if (GSConfig.Renderer == GSRendererType::SW && (TEX0.TW != res.TW || TEX0.TH != res.TH))
+ if (TEX0.TW != res.TW || TEX0.TH != res.TH)
{
GL_DBG("FixedTEX0 %05x %d %d tw %d=>%d th %d=>%d st (%.0f,%.0f,%.0f,%.0f) uvmax %d,%d wm %d,%d (%d,%d,%d,%d)",
(int)TEX0.TBP0, (int)TEX0.TBW, (int)TEX0.PSM,
@@ -142,50 +142,3 @@ GIFRegTEX0 GSDrawingContext::GetSizeFixedTEX0(const GSVector4& st, bool linear,
return res;
}
-
-void GSDrawingContext::ComputeFixedTEX0(const GSVector4& st)
-{
- // It is quite complex to handle rescaling so this function is less stricter than GetSizeFixedTEX0,
- // therefore we remove the reduce optimization and we don't handle bilinear filtering which might create wrong interpolation at the border.
- int tw = TEX0.TW;
- int th = TEX0.TH;
-
- int wms = (int)CLAMP.WMS;
- int wmt = (int)CLAMP.WMT;
-
- int minu = (int)CLAMP.MINU;
- int minv = (int)CLAMP.MINV;
- int maxu = (int)CLAMP.MAXU;
- int maxv = (int)CLAMP.MAXV;
-
- if (wms != CLAMP_REGION_CLAMP)
- tw = tw > 10 ? 0 : tw;
-
- if (wmt != CLAMP_REGION_CLAMP)
- th = th > 10 ? 0 : th;
-
- GSVector4i uv = GSVector4i(st.floor().xyzw(st.ceil()));
-
- uv.x = findmax(uv.x, uv.z, (1 << tw) - 1, wms, minu, maxu);
- uv.y = findmax(uv.y, uv.w, (1 << th) - 1, wmt, minv, maxv);
-
- if (wms == CLAMP_REGION_CLAMP || wms == CLAMP_REGION_REPEAT)
- tw = extend(uv.x, tw);
-
- if (wmt == CLAMP_REGION_CLAMP || wmt == CLAMP_REGION_REPEAT)
- th = extend(uv.y, th);
-
- tw = std::clamp<int>(tw, 0, 10);
- th = std::clamp<int>(th, 0, 10);
-
- if ((tw != (int)TEX0.TW) || (th != (int)TEX0.TH))
- {
- m_fixed_tex0 = true;
- TEX0.TW = tw;
- TEX0.TH = th;
-
- GL_DBG("FixedTEX0 TW %d=>%d, TH %d=>%d wm %d,%d",
- (int)stack.TEX0.TW, (int)TEX0.TW, (int)stack.TEX0.TH, (int)TEX0.TH,
- (int)CLAMP.WMS, (int)CLAMP.WMT);
- }
-}
diff --git a/pcsx2/GS/GSDrawingContext.h b/pcsx2/GS/GSDrawingContext.h
index dedba2910c..2122880383 100644
--- a/pcsx2/GS/GSDrawingContext.h
+++ b/pcsx2/GS/GSDrawingContext.h
@@ -69,12 +69,8 @@ public:
GIFRegZBUF ZBUF;
} stack;
- bool m_fixed_tex0;
-
GSDrawingContext()
{
- m_fixed_tex0 = false;
-
memset(&offset, 0, sizeof(offset));
Reset();
@@ -140,8 +136,6 @@ public:
}
GIFRegTEX0 GetSizeFixedTEX0(const GSVector4& st, bool linear, bool mipmap = false) const;
- void ComputeFixedTEX0(const GSVector4& st);
- bool HasFixedTEX0() const { return m_fixed_tex0; }
// Save & Restore before/after draw allow to correct/optimize current register for current draw
// Note: we could avoid the restore part if all renderer code is updated to use a local copy instead
@@ -159,9 +153,6 @@ public:
stack.FBA = FBA;
stack.FRAME = FRAME;
stack.ZBUF = ZBUF;
-
- // This function is called before the draw so take opportunity to reset m_fixed_tex0
- m_fixed_tex0 = false;
}
void RestoreReg()
diff --git a/pcsx2/GS/GSRegs.h b/pcsx2/GS/GSRegs.h
index 1ab1aa12cd..4076141964 100644
--- a/pcsx2/GS/GSRegs.h
+++ b/pcsx2/GS/GSRegs.h
@@ -823,6 +823,7 @@ union
REG_END2
__forceinline bool IsRepeating() const
{
+ // This is actually "does the texture span more than one page".
if (TBW < 2)
{
if (PSM == PSM_PSMT8)
diff --git a/pcsx2/GS/GSState.cpp b/pcsx2/GS/GSState.cpp
index 5077d94f3a..0fda71972e 100644
--- a/pcsx2/GS/GSState.cpp
+++ b/pcsx2/GS/GSState.cpp
@@ -1699,7 +1699,6 @@ inline void GSState::CopyEnv(GSDrawingEnvironment* dest, GSDrawingEnvironment* s
{
memcpy(dest, src, 88);
memcpy(&dest->CTXT[ctx], &src->CTXT[ctx], 96);
- dest->CTXT[ctx].m_fixed_tex0 = src->CTXT[ctx].m_fixed_tex0;
}
void GSState::Flush(GSFlushReason reason)
@@ -3583,8 +3582,11 @@ GSState::TextureMinMaxResult GSState::GetTextureMinMax(const GIFRegTEX0& TEX0, c
const int minu = (int)CLAMP.MINU;
const int minv = (int)CLAMP.MINV;
- const int maxu = (int)CLAMP.MAXU;
- const int maxv = (int)CLAMP.MAXV;
+
+ // For the FixedTEX0 case, in hardware, we handle this in the texture cache. Don't OR the bits in here, otherwise
+ // we'll end up with an invalid rectangle, we want the passed-in rectangle to be relative to the normalized size.
+ const int maxu = (wms != CLAMP_REGION_REPEAT || (int)CLAMP.MAXU < w) ? (int)CLAMP.MAXU : 0;
+ const int maxv = (wmt != CLAMP_REGION_REPEAT || (int)CLAMP.MAXV < h) ? (int)CLAMP.MAXV : 0;
GSVector4i vr = tr;
diff --git a/pcsx2/GS/Renderers/Common/GSDevice.h b/pcsx2/GS/Renderers/Common/GSDevice.h
index 74357290c4..5b536207e1 100644
--- a/pcsx2/GS/Renderers/Common/GSDevice.h
+++ b/pcsx2/GS/Renderers/Common/GSDevice.h
@@ -309,6 +309,8 @@ struct alignas(16) GSHWDrawConfig
u32 tcc : 1;
u32 wms : 2;
u32 wmt : 2;
+ u32 adjs : 1;
+ u32 adjt : 1;
u32 ltf : 1;
// Shuffle and fbmask effect
u32 shuffle : 1;
@@ -352,7 +354,6 @@ struct alignas(16) GSHWDrawConfig
u32 automatic_lod : 1;
u32 manual_lod : 1;
u32 point_sampler : 1;
- u32 invalid_tex0 : 1; // Lupin the 3rd
// Scan mask
u32 scanmsk : 2;
@@ -554,11 +555,11 @@ struct alignas(16) GSHWDrawConfig
GSVector4 FogColor_AREF;
GSVector4 WH;
GSVector4 TA_MaxDepth_Af;
- GSVector4i MskFix;
GSVector4i FbMask;
GSVector4 HalfTexel;
GSVector4 MinMax;
+ GSVector4 STRange;
GSVector4i ChannelShuffle;
GSVector2 TCOffsetHack;
GSVector2 STScale;
diff --git a/pcsx2/GS/Renderers/DX11/GSTextureFX11.cpp b/pcsx2/GS/Renderers/DX11/GSTextureFX11.cpp
index 60c3621568..fafe075d88 100644
--- a/pcsx2/GS/Renderers/DX11/GSTextureFX11.cpp
+++ b/pcsx2/GS/Renderers/DX11/GSTextureFX11.cpp
@@ -142,6 +142,8 @@ void GSDevice11::SetupPS(const PSSelector& sel, const GSHWDrawConfig::PSConstant
sm.AddMacro("PS_FST", sel.fst);
sm.AddMacro("PS_WMS", sel.wms);
sm.AddMacro("PS_WMT", sel.wmt);
+ sm.AddMacro("PS_ADJS", sel.adjs);
+ sm.AddMacro("PS_ADJT", sel.adjt);
sm.AddMacro("PS_AEM_FMT", sel.aem_fmt);
sm.AddMacro("PS_AEM", sel.aem);
sm.AddMacro("PS_TFX", sel.tfx);
@@ -164,7 +166,6 @@ void GSDevice11::SetupPS(const PSSelector& sel, const GSHWDrawConfig::PSConstant
sm.AddMacro("PS_DFMT", sel.dfmt);
sm.AddMacro("PS_DEPTH_FMT", sel.depth_fmt);
sm.AddMacro("PS_PAL_FMT", sel.pal_fmt);
- sm.AddMacro("PS_INVALID_TEX0", sel.invalid_tex0);
sm.AddMacro("PS_HDR", sel.hdr);
sm.AddMacro("PS_COLCLIP", sel.colclip);
sm.AddMacro("PS_BLEND_A", sel.blend_a);
diff --git a/pcsx2/GS/Renderers/DX12/GSDevice12.cpp b/pcsx2/GS/Renderers/DX12/GSDevice12.cpp
index 0b24c4c735..e67fd9bde5 100644
--- a/pcsx2/GS/Renderers/DX12/GSDevice12.cpp
+++ b/pcsx2/GS/Renderers/DX12/GSDevice12.cpp
@@ -1483,6 +1483,8 @@ const ID3DBlob* GSDevice12::GetTFXPixelShader(const GSHWDrawConfig::PSSelector&
sm.AddMacro("PS_FST", sel.fst);
sm.AddMacro("PS_WMS", sel.wms);
sm.AddMacro("PS_WMT", sel.wmt);
+ sm.AddMacro("PS_ADJS", sel.adjs);
+ sm.AddMacro("PS_ADJT", sel.adjt);
sm.AddMacro("PS_AEM_FMT", sel.aem_fmt);
sm.AddMacro("PS_AEM", sel.aem);
sm.AddMacro("PS_TFX", sel.tfx);
@@ -1505,7 +1507,6 @@ const ID3DBlob* GSDevice12::GetTFXPixelShader(const GSHWDrawConfig::PSSelector&
sm.AddMacro("PS_DFMT", sel.dfmt);
sm.AddMacro("PS_DEPTH_FMT", sel.depth_fmt);
sm.AddMacro("PS_PAL_FMT", sel.pal_fmt);
- sm.AddMacro("PS_INVALID_TEX0", sel.invalid_tex0);
sm.AddMacro("PS_HDR", sel.hdr);
sm.AddMacro("PS_COLCLIP", sel.colclip);
sm.AddMacro("PS_BLEND_A", sel.blend_a);
diff --git a/pcsx2/GS/Renderers/HW/GSRendererHW.cpp b/pcsx2/GS/Renderers/HW/GSRendererHW.cpp
index e5ae60f811..bb94c21ab9 100644
--- a/pcsx2/GS/Renderers/HW/GSRendererHW.cpp
+++ b/pcsx2/GS/Renderers/HW/GSRendererHW.cpp
@@ -1279,10 +1279,6 @@ void GSRendererHW::Draw()
return;
}
- // Fix TEX0 size
- if (PRIM->TME && !IsMipMapActive())
- m_context->ComputeFixedTEX0(m_vt.m_min.t.xyxy(m_vt.m_max.t));
-
// skip alpha test if possible
// Note: do it first so we know if frame/depth writes are masked
@@ -1528,8 +1524,8 @@ void GSRendererHW::Draw()
TextureMinMaxResult tmm = GetTextureMinMax(TEX0, MIP_CLAMP, m_vt.IsLinear());
- m_src = tex_psm.depth ? m_tc->LookupDepthSource(TEX0, env.TEXA, tmm.coverage) :
- m_tc->LookupSource(TEX0, env.TEXA, tmm.coverage, (GSConfig.HWMipmap >= HWMipmapLevel::Basic ||
+ m_src = tex_psm.depth ? m_tc->LookupDepthSource(TEX0, env.TEXA, MIP_CLAMP, tmm.coverage) :
+ m_tc->LookupSource(TEX0, env.TEXA, MIP_CLAMP, tmm.coverage, (GSConfig.HWMipmap >= HWMipmapLevel::Basic ||
GSConfig.TriFilter == TriFiltering::Forced) ? &hash_lod_range : nullptr);
// Hypothesis: texture shuffle is used as a postprocessing effect so texture will be an old target.
@@ -1642,7 +1638,7 @@ void GSRendererHW::Draw()
for (int layer = m_lod.x + 1; layer <= m_lod.y; layer++)
{
- const GIFRegTEX0& MIP_TEX0 = GetTex0Layer(layer);
+ const GIFRegTEX0 MIP_TEX0(GetTex0Layer(layer));
m_context->offset.tex = m_mem.GetOffset(MIP_TEX0.TBP0, MIP_TEX0.TBW, MIP_TEX0.PSM);
@@ -3105,6 +3101,26 @@ void GSRendererHW::EmulateBlending(bool& DATE_PRIMID, bool& DATE_BARRIER, bool&
}
}
+__ri static constexpr bool IsRedundantClamp(u8 clamp, u32 clamp_min, u32 clamp_max, u32 tsize)
+{
+ // Don't shader sample when the clamp/repeat is configured to the texture size.
+ // That way trilinear etc still works.
+ const u32 textent = (1u << tsize) - 1u;
+ if (clamp == CLAMP_REGION_CLAMP)
+ return (clamp_min == 0 && clamp_max == textent);
+ else if (clamp == CLAMP_REGION_REPEAT)
+ return (clamp_max == 0 && clamp_min == textent);
+ else
+ return false;
+}
+
+__ri static constexpr u8 EffectiveClamp(u8 clamp, bool has_region)
+{
+ // When we have extracted the region in the texture, we can use the hardware sampler for repeat/clamp.
+ // (weird flip here because clamp/repeat is inverted for region vs non-region).
+ return (clamp >= CLAMP_REGION_CLAMP && has_region) ? (clamp ^ 3) : clamp;
+}
+
void GSRendererHW::EmulateTextureSampler(const GSTextureCache::Source* tex)
{
// Warning fetch the texture PSM format rather than the context format. The latter could have been corrected in the texture cache for depth.
@@ -3112,9 +3128,16 @@ void GSRendererHW::EmulateTextureSampler(const GSTextureCache::Source* tex)
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[tex->m_TEX0.PSM];
const GSLocalMemory::psm_t& cpsm = psm.pal > 0 ? GSLocalMemory::m_psm[m_context->TEX0.CPSM] : psm;
- const u8 wms = m_context->CLAMP.WMS;
- const u8 wmt = m_context->CLAMP.WMT;
+ static constexpr const char* clamp_modes[] = { "REPEAT", "CLAMP", "REGION_CLAMP", "REGION_REPEAT" };
+ const bool redundant_wms = IsRedundantClamp(m_context->CLAMP.WMS, m_context->CLAMP.MINU, m_context->CLAMP.MAXU, tex->m_TEX0.TW);
+ const bool redundant_wmt = IsRedundantClamp(m_context->CLAMP.WMT, m_context->CLAMP.MINV, m_context->CLAMP.MAXV, tex->m_TEX0.TH);
+ const u8 wms = EffectiveClamp(m_context->CLAMP.WMS, tex->m_region.HasX());
+ const u8 wmt = EffectiveClamp(m_context->CLAMP.WMT, tex->m_region.HasY());
const bool complex_wms_wmt = !!((wms | wmt) & 2);
+ GL_CACHE("WMS: %s [%s%s] WMT: %s [%s%s] Complex: %d MINU: %d MINV: %d MINV: %d MAXV: %d",
+ clamp_modes[m_context->CLAMP.WMS], redundant_wms ? "redundant," : "", clamp_modes[wms],
+ clamp_modes[m_context->CLAMP.WMT], redundant_wmt ? "redundant," : "", clamp_modes[wmt],
+ complex_wms_wmt, m_context->CLAMP.MINU, m_context->CLAMP.MINV, m_context->CLAMP.MAXU, m_context->CLAMP.MAXV);
const bool need_mipmap = IsMipMapDraw();
const bool shader_emulated_sampler = tex->m_palette || cpsm.fmt != 0 || complex_wms_wmt || psm.depth;
@@ -3290,14 +3313,38 @@ void GSRendererHW::EmulateTextureSampler(const GSTextureCache::Source* tex)
const GSVector4 st_scale = WH.zwzw() / GSVector4(w, h).xyxy();
m_conf.cb_ps.STScale = GSVector2(st_scale.x, st_scale.y);
+ if (tex->m_region.HasX())
+ {
+ m_conf.cb_ps.STRange.x = static_cast<float>(tex->m_region.GetMinX()) / static_cast<float>(miptw);
+ m_conf.cb_ps.STRange.z = static_cast<float>(miptw) / static_cast<float>(tex->m_region.GetWidth());
+ m_conf.ps.adjs = 1;
+ }
+ if (tex->m_region.HasY())
+ {
+ m_conf.cb_ps.STRange.y = static_cast<float>(tex->m_region.GetMinY()) / static_cast<float>(mipth);
+ m_conf.cb_ps.STRange.w = static_cast<float>(mipth) / static_cast<float>(tex->m_region.GetHeight());
+ m_conf.ps.adjt = 1;
+ }
+
m_conf.ps.fst = !!PRIM->FST;
m_conf.cb_ps.WH = WH;
m_conf.cb_ps.HalfTexel = GSVector4(-0.5f, 0.5f).xxyy() / WH.zwzw();
if (complex_wms_wmt)
{
- m_conf.cb_ps.MskFix = GSVector4i(m_context->CLAMP.MINU, m_context->CLAMP.MINV, m_context->CLAMP.MAXU, m_context->CLAMP.MAXV);;
- m_conf.cb_ps.MinMax = GSVector4(m_conf.cb_ps.MskFix) / WH.xyxy();
+ const GSVector4i clamp(m_context->CLAMP.MINU, m_context->CLAMP.MINV, m_context->CLAMP.MAXU, m_context->CLAMP.MAXV);
+ const GSVector4 region_repeat(GSVector4::cast(clamp));
+ const GSVector4 region_clamp(GSVector4(clamp) / WH.xyxy());
+ if (wms >= CLAMP_REGION_CLAMP)
+ {
+ m_conf.cb_ps.MinMax.x = (wms == CLAMP_REGION_CLAMP && !m_conf.ps.depth_fmt) ? region_clamp.x : region_repeat.x;
+ m_conf.cb_ps.MinMax.z = (wms == CLAMP_REGION_CLAMP && !m_conf.ps.depth_fmt) ? region_clamp.z : region_repeat.z;
+ }
+ if (wmt >= CLAMP_REGION_CLAMP)
+ {
+ m_conf.cb_ps.MinMax.y = (wmt == CLAMP_REGION_CLAMP && !m_conf.ps.depth_fmt) ? region_clamp.y : region_repeat.y;
+ m_conf.cb_ps.MinMax.w = (wmt == CLAMP_REGION_CLAMP && !m_conf.ps.depth_fmt) ? region_clamp.w : region_repeat.w;
+ }
}
else if (trilinear_manual)
{
@@ -3318,18 +3365,6 @@ void GSRendererHW::EmulateTextureSampler(const GSTextureCache::Source* tex)
m_conf.cb_ps.TCOffsetHack = GSVector2(tc_oh_ts.z, tc_oh_ts.w);
m_conf.cb_vs.texture_scale = GSVector2(tc_oh_ts.x, tc_oh_ts.y);
- // Must be done after all coordinates math
- if (m_context->HasFixedTEX0() && !PRIM->FST)
- {
- m_conf.ps.invalid_tex0 = 1;
- // Use invalid size to denormalize ST coordinate
- m_conf.cb_ps.WH.x = static_cast<float>(1 << m_context->stack.TEX0.TW);
- m_conf.cb_ps.WH.y = static_cast<float>(1 << m_context->stack.TEX0.TH);
-
- // We can't handle m_target with invalid_tex0 atm due to upscaling
- ASSERT(!tex->m_target);
- }
-
// Only enable clamping in CLAMP mode. REGION_CLAMP will be done manually in the shader
m_conf.sampler.tau = (wms != CLAMP_CLAMP);
m_conf.sampler.tav = (wmt != CLAMP_CLAMP);
diff --git a/pcsx2/GS/Renderers/HW/GSTextureCache.cpp b/pcsx2/GS/Renderers/HW/GSTextureCache.cpp
index a6c4c33040..56d2b564f0 100644
--- a/pcsx2/GS/Renderers/HW/GSTextureCache.cpp
+++ b/pcsx2/GS/Renderers/HW/GSTextureCache.cpp
@@ -119,7 +119,7 @@ void GSTextureCache::AddDirtyRectTarget(Target* target, GSVector4i rect, u32 psm
target->m_dirty.push_back(GSDirtyRect(rect, psm, bw));
}
-GSTextureCache::Source* GSTextureCache::LookupDepthSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GSVector4i& r, bool palette)
+GSTextureCache::Source* GSTextureCache::LookupDepthSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, bool palette)
{
if (GSConfig.UserHacks_DisableDepthSupport)
{
@@ -177,7 +177,7 @@ GSTextureCache::Source* GSTextureCache::LookupDepthSource(const GIFRegTEX0& TEX0
TEX0.TBP0, psm_str(psm));
// Create a shared texture source
- src = new Source(TEX0, TEXA, true);
+ src = new Source(TEX0, TEXA);
src->m_texture = dst->m_texture;
src->m_shared_texture = true;
src->m_target = true; // So renderer can check if a conversion is required
@@ -201,7 +201,7 @@ GSTextureCache::Source* GSTextureCache::LookupDepthSource(const GIFRegTEX0& TEX0
else if (g_gs_renderer->m_game.title == CRC::SVCChaos || g_gs_renderer->m_game.title == CRC::KOF2002)
{
// SVCChaos black screen & KOF2002 blue screen on main menu, regardless of depth enabled or disabled.
- return LookupSource(TEX0, TEXA, r, nullptr);
+ return LookupSource(TEX0, TEXA, CLAMP, r, nullptr);
}
else
{
@@ -227,24 +227,13 @@ GSTextureCache::Source* GSTextureCache::LookupDepthSource(const GIFRegTEX0& TEX0
return src;
}
-GSTextureCache::Source* GSTextureCache::LookupSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GSVector4i& r, const GSVector2i* lod)
+__ri static GSTextureCache::Source* FindSourceInMap(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA,
+ const GSLocalMemory::psm_t& psm_s, const u32* clut, const GSTexture* gpu_clut, const GSVector2i& compare_lod,
+ const GSTextureCache::SourceRegion& region, FastList<GSTextureCache::Source*>& map)
{
- GL_CACHE("TC: Lookup Source <%d,%d => %d,%d> (0x%x, %s, BW: %u, CBP: 0x%x)", r.x, r.y, r.z, r.w, TEX0.TBP0, psm_str(TEX0.PSM), TEX0.TBW, TEX0.CBP);
-
- const GSLocalMemory::psm_t& psm_s = GSLocalMemory::m_psm[TEX0.PSM];
- //const GSLocalMemory::psm_t& cpsm = psm.pal > 0 ? GSLocalMemory::m_psm[TEX0.CPSM] : psm;
-
- const u32* const clut = g_gs_renderer->m_mem.m_clut;
- GSTexture* const gpu_clut = (psm_s.pal > 0) ? g_gs_renderer->m_mem.m_clut.GetGPUTexture() : nullptr;
-
- Source* src = NULL;
-
- auto& m = m_src.m_map[TEX0.TBP0 >> 5];
-
- const GSVector2i compare_lod(lod ? *lod : GSVector2i(0, 0));
- for (auto i = m.begin(); i != m.end(); ++i)
+ for (auto i = map.begin(); i != map.end(); ++i)
{
- Source* s = *i;
+ GSTextureCache::Source* s = *i;
if (((TEX0.U32[0] ^ s->m_TEX0.U32[0]) | ((TEX0.U32[1] ^ s->m_TEX0.U32[1]) & 3)) != 0) // TBP0 TBW PSM TW TH
continue;
@@ -272,20 +261,92 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const GIFRegTEX0& TEX0, con
continue;
}
+ if (s->m_region.bits != 0 && s->m_region.bits != region.bits)
+ continue;
+
// Same base mip texture, but we need to check that MXL was the same as well.
// When mipmapping is off, this will be 0,0 vs 0,0.
if (s->m_lod != compare_lod)
continue;
}
- m.MoveFront(i.Index());
-
- src = s;
-
- break;
+ map.MoveFront(i.Index());
+ return s;
}
- Target* dst = NULL;
+ return nullptr;
+}
+
+GSTextureCache::Source* GSTextureCache::LookupSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, const GSVector2i* lod)
+{
+ GL_CACHE("TC: Lookup Source <%d,%d => %d,%d> (0x%x, %s, BW: %u, CBP: 0x%x, TW: %d, TH: %d)", r.x, r.y, r.z, r.w, TEX0.TBP0, psm_str(TEX0.PSM), TEX0.TBW, TEX0.CBP, 1 << TEX0.TW, 1 << TEX0.TH);
+
+ const GSLocalMemory::psm_t& psm_s = GSLocalMemory::m_psm[TEX0.PSM];
+ //const GSLocalMemory::psm_t& cpsm = psm.pal > 0 ? GSLocalMemory::m_psm[TEX0.CPSM] : psm;
+
+ const u32* const clut = g_gs_renderer->m_mem.m_clut;
+ GSTexture* const gpu_clut = (psm_s.pal > 0) ? g_gs_renderer->m_mem.m_clut.GetGPUTexture() : nullptr;
+
+ SourceRegion region = {};
+ if (CLAMP.WMS == CLAMP_REGION_CLAMP && CLAMP.MAXU >= CLAMP.MINU)
+ {
+ // Another Lupin case here, it uses region clamp with UV (not ST), puts a clamp region further
+ // into the texture, but a smaller TW/TH. Catch this by looking for a clamp range above TW.
+ const u32 rw = CLAMP.MAXU - CLAMP.MAXU + 1;
+ if (rw < (1u << TEX0.TW) || CLAMP.MAXU >= (1u << TEX0.TW))
+ {
+ region.SetX(CLAMP.MINU, CLAMP.MAXU + 1);
+ GL_CACHE("TC: Region clamp optimization: %d width -> %d", 1 << TEX0.TW, region.GetWidth());
+ }
+ }
+ else if (CLAMP.WMS == CLAMP_REGION_REPEAT && CLAMP.MINU != 0)
+ {
+ // Lupin the 3rd is really evil, it sets TW/TH to the texture size, but then uses region repeat
+ // to offset the actual texture data to elsewhere. So, we'll just force any cases like this down
+ // the region texture path.
+ const u32 rw = ((CLAMP.MINU | CLAMP.MAXU) - CLAMP.MAXU) + 1;
+ if (rw < (1u << TEX0.TW) || CLAMP.MAXU != 0)
+ {
+ region.SetX(CLAMP.MAXU, (CLAMP.MINU | CLAMP.MAXU) + 1);
+ GL_CACHE("TC: Region repeat optimization: %d width -> %d", 1 << TEX0.TW, region.GetWidth());
+ }
+ }
+ if (CLAMP.WMT == CLAMP_REGION_CLAMP && CLAMP.MAXV >= CLAMP.MINV)
+ {
+ const u32 rh = CLAMP.MAXV - CLAMP.MINV + 1;
+ if (rh < (1u << TEX0.TH) || CLAMP.MAXV >= (1u << TEX0.TH))
+ {
+ region.SetY(CLAMP.MINV, CLAMP.MAXV + 1);
+ GL_CACHE("TC: Region clamp optimization: %d height -> %d", 1 << TEX0.TW, region.GetHeight());
+ }
+ }
+ else if (CLAMP.WMT == CLAMP_REGION_REPEAT && CLAMP.MINV != 0)
+ {
+ const u32 rh = ((CLAMP.MINV | CLAMP.MAXV) - CLAMP.MAXV) + 1;
+ if (rh < (1u << TEX0.TH) || CLAMP.MAXV != 0)
+ {
+ region.SetY(CLAMP.MAXV, (CLAMP.MINV | CLAMP.MAXV) + 1);
+ GL_CACHE("TC: Region repeat optimization: %d height -> %d", 1 << TEX0.TW, region.GetHeight());
+ }
+ }
+
+ const GSVector2i compare_lod(lod ? *lod : GSVector2i(0, 0));
+ Source* src = nullptr;
+
+ // Region textures might be placed in a different page, so check that first.
+ const u32 lookup_page = TEX0.TBP0 >> 5;
+ if (region.GetMinX() != 0 || region.GetMinY() != 0)
+ {
+ const GSOffset offset(psm_s.info, TEX0.TBP0, TEX0.TBW, TEX0.PSM);
+ const u32 region_page = offset.bn(region.GetMinX(), region.GetMinY()) >> 5;
+ if (lookup_page != region_page)
+ src = FindSourceInMap(TEX0, TEXA, psm_s, clut, gpu_clut, compare_lod, region, m_src.m_map[region_page]);
+ }
+ if (!src)
+ src = FindSourceInMap(TEX0, TEXA, psm_s, clut, gpu_clut, compare_lod, region, m_src.m_map[lookup_page]);
+
+
+ Target* dst = nullptr;
bool half_right = false;
int x_offset = 0;
int y_offset = 0;
@@ -293,7 +354,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const GIFRegTEX0& TEX0, con
#ifdef DISABLE_HW_TEXTURE_CACHE
if (0)
#else
- if (src == NULL)
+ if (!src)
#endif
{
const u32 bp = TEX0.TBP0;
@@ -466,11 +527,11 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const GIFRegTEX0& TEX0, con
GIFRegTEX0 depth_TEX0;
depth_TEX0.U32[0] = TEX0.U32[0] | (0x30u << 20u);
depth_TEX0.U32[1] = TEX0.U32[1];
- return LookupDepthSource(depth_TEX0, TEXA, r);
+ return LookupDepthSource(depth_TEX0, TEXA, CLAMP, r);
}
else
{
- return LookupDepthSource(TEX0, TEXA, r, true);
+ return LookupDepthSource(TEX0, TEXA, CLAMP, r, true);
}
}
}
@@ -496,7 +557,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const GIFRegTEX0& TEX0, con
GL_CACHE("TC: src miss (0x%x, 0x%x, %s)", TEX0.TBP0, psm_s.pal > 0 ? TEX0.CBP : 0, psm_str(TEX0.PSM));
}
#endif
- src = CreateSource(TEX0, TEXA, dst, half_right, x_offset, y_offset, lod, &r, gpu_clut);
+ src = CreateSource(TEX0, TEXA, dst, half_right, x_offset, y_offset, lod, &r, gpu_clut, region);
}
else
{
@@ -1893,13 +1954,13 @@ void GSTextureCache::IncAge()
}
//Fixme: Several issues in here. Not handling depth stencil, pitch conversion doesnt work.
-GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Target* dst, bool half_right, int x_offset, int y_offset, const GSVector2i* lod, const GSVector4i* src_range, GSTexture* gpu_clut)
+GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Target* dst, bool half_right, int x_offset, int y_offset, const GSVector2i* lod, const GSVector4i* src_range, GSTexture* gpu_clut, SourceRegion region)
{
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[TEX0.PSM];
- Source* src = new Source(TEX0, TEXA, false);
+ Source* src = new Source(TEX0, TEXA);
- const int tw = 1 << TEX0.TW;
- const int th = 1 << TEX0.TH;
+ int tw = 1 << TEX0.TW;
+ int th = 1 << TEX0.TH;
//int tp = TEX0.TBW << 6;
int tlevels = 1;
if (lod)
@@ -2211,8 +2272,13 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
bool paltex = (GSConfig.GPUPaletteConversion && psm.pal > 0) || gpu_clut;
const u32* clut = (psm.pal > 0) ? static_cast<const u32*>(g_gs_renderer->m_mem.m_clut) : nullptr;
+ // adjust texture size to fit
+ src->m_region = region;
+ tw = region.HasX() ? region.GetWidth() : tw;
+ th = region.HasY() ? region.GetHeight() : th;
+
// try the hash cache
- if ((src->m_from_hash_cache = LookupHashCache(TEX0, TEXA, paltex, clut, lod)) != nullptr)
+ if ((src->m_from_hash_cache = LookupHashCache(TEX0, TEXA, paltex, clut, lod, region)) != nullptr)
{
src->m_texture = src->m_from_hash_cache->texture;
if (gpu_clut)
@@ -2245,6 +2311,8 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
ASSERT(src->m_from_target == (dst ? &dst->m_texture : nullptr));
ASSERT(src->m_texture->GetScale() == ((!dst || TEX0.PSM == PSM_PSMT8) ? GSVector2(1, 1) : dst->m_texture->GetScale()));
+ src->SetPages();
+
m_src.Add(src, TEX0, g_gs_renderer->m_context->offset.tex);
return src;
@@ -2253,7 +2321,7 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
// This really needs a better home...
extern bool FMVstarted;
-GSTextureCache::HashCacheEntry* GSTextureCache::LookupHashCache(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, bool& paltex, const u32* clut, const GSVector2i* lod)
+GSTextureCache::HashCacheEntry* GSTextureCache::LookupHashCache(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, bool& paltex, const u32* clut, const GSVector2i* lod, SourceRegion region)
{
// don't bother hashing if we're not dumping or replacing.
const bool dump = GSConfig.DumpReplaceableTextures && (!FMVstarted || GSConfig.DumpTexturesWithFMVActive) &&
@@ -2265,13 +2333,13 @@ GSTextureCache::HashCacheEntry* GSTextureCache::LookupHashCache(const GIFRegTEX0
// need the hash either for replacing, dumping or caching.
// if dumping/replacing is on, we compute the clut hash regardless, since replacements aren't indexed
- HashCacheKey key{HashCacheKey::Create(TEX0, TEXA, (dump || replace || !paltex) ? clut : nullptr, lod)};
+ HashCacheKey key{HashCacheKey::Create(TEX0, TEXA, (dump || replace || !paltex) ? clut : nullptr, lod, region)};
// handle dumping first, this is mostly isolated.
if (dump)
{
// dump base level
- GSTextureReplacements::DumpTexture(key, TEX0, TEXA, g_gs_renderer->m_mem, 0);
+ GSTextureReplacements::DumpTexture(key, TEX0, TEXA, region, g_gs_renderer->m_mem, 0);
// and the mips
if (lod && GSConfig.DumpReplaceableMipmaps)
@@ -2281,7 +2349,7 @@ GSTextureCache::HashCacheEntry* GSTextureCache::LookupHashCache(const GIFRegTEX0
for (int mip = 1; mip < nmips; mip++)
{
const GIFRegTEX0 MIP_TEX0{g_gs_renderer->GetTex0Layer(basemip + mip)};
- GSTextureReplacements::DumpTexture(key, MIP_TEX0, TEXA, g_gs_renderer->m_mem, mip);
+ GSTextureReplacements::DumpTexture(key, MIP_TEX0, TEXA, region, g_gs_renderer->m_mem, mip);
}
}
}
@@ -2355,8 +2423,8 @@ GSTextureCache::HashCacheEntry* GSTextureCache::LookupHashCache(const GIFRegTEX0
return nullptr;
// expand/upload texture
- const int tw = 1 << TEX0.TW;
- const int th = 1 << TEX0.TH;
+ const int tw = region.HasX() ? region.GetWidth() : (1 << TEX0.TW);
+ const int th = region.HasY() ? region.GetHeight() : (1 << TEX0.TH);
const int tlevels = lod ? ((GSConfig.HWMipmap != HWMipmapLevel::Full) ? -1 : (lod->y - lod->x + 1)) : 1;
GSTexture* tex = g_gs_device->CreateTexture(tw, th, tlevels, paltex ? GSTexture::Format::UNorm8 : GSTexture::Format::Color);
if (!tex)
@@ -2366,7 +2434,7 @@ GSTextureCache::HashCacheEntry* GSTextureCache::LookupHashCache(const GIFRegTEX0
}
// upload base level
- PreloadTexture(TEX0, TEXA, g_gs_renderer->m_mem, paltex, tex, 0);
+ PreloadTexture(TEX0, TEXA, region, g_gs_renderer->m_mem, paltex, tex, 0);
// upload mips if present
if (lod)
@@ -2376,7 +2444,7 @@ GSTextureCache::HashCacheEntry* GSTextureCache::LookupHashCache(const GIFRegTEX0
for (int mip = 1; mip < nmips; mip++)
{
const GIFRegTEX0 MIP_TEX0{g_gs_renderer->GetTex0Layer(basemip + mip)};
- PreloadTexture(MIP_TEX0, TEXA, g_gs_renderer->m_mem, paltex, tex, mip);
+ PreloadTexture(MIP_TEX0, TEXA, region.AdjustForMipmap(mip), g_gs_renderer->m_mem, paltex, tex, mip);
}
}
@@ -2649,12 +2717,13 @@ bool GSTextureCache::Surface::Overlaps(u32 bp, u32 bw, u32 psm, const GSVector4i
// GSTextureCache::Source
-GSTextureCache::Source::Source(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, bool dummy_container)
+GSTextureCache::Source::Source(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA)
: m_palette_obj(nullptr)
, m_palette(nullptr)
, m_valid_rect(0, 0)
, m_lod(0, 0)
, m_target(false)
+ , m_repeating(false)
, m_p2t(NULL)
, m_from_target(NULL)
, m_from_target_TEX0(TEX0)
@@ -2662,32 +2731,8 @@ GSTextureCache::Source::Source(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, b
m_TEX0 = TEX0;
m_TEXA = TEXA;
- if (dummy_container)
- {
- // Dummy container only contain a m_texture that is a pointer to another source.
-
- m_write.rect = NULL;
- m_write.count = 0;
-
- m_repeating = false;
- }
- else
- {
- memset(m_layer_TEX0, 0, sizeof(m_layer_TEX0));
- memset(m_layer_hash, 0, sizeof(m_layer_hash));
-
- m_write.rect = (GSVector4i*)_aligned_malloc(3 * sizeof(GSVector4i), 32);
- m_write.count = 0;
-
- m_repeating = m_TEX0.IsRepeating();
-
- if (m_repeating && !CanPreload())
- {
- m_p2t = g_gs_renderer->m_mem.GetPage2TileMap(m_TEX0);
- }
-
- m_pages = g_gs_renderer->m_context->offset.tex.pageLooperForRect(GSVector4i(0, 0, 1 << TEX0.TW, 1 << TEX0.TH));
- }
+ memset(m_layer_TEX0, 0, sizeof(m_layer_TEX0));
+ memset(m_layer_hash, 0, sizeof(m_layer_hash));
}
GSTextureCache::Source::~Source()
@@ -2703,6 +2748,23 @@ GSTextureCache::Source::~Source()
}
}
+void GSTextureCache::Source::SetPages()
+{
+ const int tw = 1 << m_TEX0.TW;
+ const int th = 1 << m_TEX0.TH;
+
+ m_repeating = !m_from_hash_cache && m_TEX0.IsRepeating() && !m_region.IsFixedTEX0(tw, th);
+
+ if (m_repeating && !CanPreload())
+ {
+ // TODO: wrong for lupin/invalid tex0
+ m_p2t = g_gs_renderer->m_mem.GetPage2TileMap(m_TEX0);
+ }
+
+ const GSVector4i rect(m_region.GetRect(tw, th));
+ m_pages = g_gs_renderer->m_context->offset.tex.pageLooperForRect(rect);
+}
+
void GSTextureCache::Source::Update(const GSVector4i& rect, int level)
{
Surface::UpdateAge();
@@ -2719,9 +2781,17 @@ void GSTextureCache::Source::Update(const GSVector4i& rect, int level)
const GSVector2i& bs = GSLocalMemory::m_psm[m_TEX0.PSM].bs;
const int tw = 1 << m_TEX0.TW;
const int th = 1 << m_TEX0.TH;
- const GSVector4i r = rect.ralign<Align_Outside>(bs);
- if (r.eq(GSVector4i(0, 0, tw, th)))
+ GSVector4i r(rect);
+ const GSVector4i region_rect(m_region.GetRect(tw, th));
+
+ // Offset the pages we use by the clamp region.
+ if (m_region.HasEither())
+ r = (r + m_region.GetOffset(tw, th)).rintersect(region_rect);
+
+ r = r.ralign<Align_Outside>(bs);
+
+ if (region_rect.eq(m_region.HasEither() ? r.rintersect(region_rect) : r))
m_complete_layers |= (1u << level);
const GSOffset& off = g_gs_renderer->m_context->offset.tex;
@@ -2818,6 +2888,9 @@ void GSTextureCache::Source::UpdateLayer(const GIFRegTEX0& TEX0, const GSVector4
void GSTextureCache::Source::Write(const GSVector4i& r, int layer)
{
+ if (!m_write.rect)
+ m_write.rect = static_cast<GSVector4i*>(_aligned_malloc(3 * sizeof(GSVector4i), 32));
+
m_write.rect[m_write.count++] = r;
while (m_write.count >= 2)
@@ -2857,11 +2930,12 @@ void GSTextureCache::Source::Flush(u32 count, int layer)
// However the function is never called for these cases. This is just for information
// should someone wish to use this function for these cases later.
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[m_TEX0.PSM];
+ const SourceRegion region((layer == 0) ? m_region : m_region.AdjustForMipmap(layer));
- const int tw = 1 << m_TEX0.TW;
- const int th = 1 << m_TEX0.TH;
-
- const GSVector4i tr(0, 0, tw, th);
+ // For the invalid tex0 case, the region might be larger than TEX0.TW/TH.
+ const int tw = std::max(region.GetWidth(), 1u << m_TEX0.TW);
+ const int th = std::max(region.GetHeight(), 1u << m_TEX0.TH);
+ const GSVector4i tex_r(region.GetRect(tw, th));
int pitch = std::max(tw, psm.bs.x) * sizeof(u32);
@@ -2877,35 +2951,33 @@ void GSTextureCache::Source::Flush(u32 count, int layer)
rtx = psm.rtxP;
}
- u8* buff = s_unswizzle_buffer;
-
for (u32 i = 0; i < count; i++)
{
- const GSVector4i r = m_write.rect[i];
+ const GSVector4i r(m_write.rect[i]);
- if ((r > tr).mask() & 0xff00)
- {
- rtx(mem, off, r, buff, pitch, m_TEXA);
-
- m_texture->Update(r.rintersect(tr), buff, pitch, layer);
- }
- else
+ // if update rect lies to the left/above of the region rectangle, or extends past the texture bounds, we can't use a direct map
+ if (((r > tex_r).mask() & 0xff00) == 0 && ((tex_r > r).mask() & 0x00ff) == 0)
{
GSTexture::GSMap m;
-
- if (m_texture->Map(m, &r, layer))
+ const GSVector4i map_r(r - tex_r.xyxy());
+ if (m_texture->Map(m, &map_r, layer))
{
rtx(mem, off, r, m.bits, m.pitch, m_TEXA);
-
m_texture->Unmap();
- }
- else
- {
- rtx(mem, off, r, buff, pitch, m_TEXA);
-
- m_texture->Update(r, buff, pitch, layer);
+ continue;
}
}
+
+ const GSVector4i rint(r.rintersect(tex_r));
+ if (rint.width() == 0 || rint.height() == 0)
+ continue;
+
+ rtx(mem, off, r, s_unswizzle_buffer, pitch, m_TEXA);
+
+ // need to offset if we're a region texture
+ const u8* src = s_unswizzle_buffer + (pitch * static_cast<u32>(std::max(tex_r.top - r.top, 0))) +
+ (static_cast<u32>(std::max(tex_r.left - r.left, 0)) << (m_palette ? 0 : 2));
+ m_texture->Update(rint - tex_r.xyxy(), src, pitch, layer);
}
if (count < m_write.count)
@@ -2920,7 +2992,7 @@ void GSTextureCache::Source::Flush(u32 count, int layer)
void GSTextureCache::Source::PreloadLevel(int level)
{
// m_TEX0 is adjusted for mips (messy, should be changed).
- const HashType hash = HashTexture(m_TEX0, m_TEXA);
+ const HashType hash = HashTexture(m_TEX0, m_TEXA, m_region);
// Layer is complete again, regardless of whether the hash matches or not (and we reupload).
const u8 layer_bit = static_cast<u8>(1) << level;
@@ -2934,7 +3006,7 @@ void GSTextureCache::Source::PreloadLevel(int level)
m_layer_hash[level] = hash;
// And upload the texture.
- PreloadTexture(m_TEX0, m_TEXA, g_gs_renderer->m_mem, m_palette != nullptr, m_texture, level);
+ PreloadTexture(m_TEX0, m_TEXA, m_region.AdjustForMipmap(level), g_gs_renderer->m_mem, m_palette != nullptr, m_texture, level);
}
bool GSTextureCache::Source::ClutMatch(const PaletteKey& palette_key)
@@ -3674,6 +3746,47 @@ bool GSTextureCache::SurfaceOffsetKeyEqual::operator()(const GSTextureCache::Sur
return true;
}
+bool GSTextureCache::SourceRegion::IsFixedTEX0(int tw, int th) const
+{
+ return (GetMinX() >= static_cast<u32>(tw) || GetMinY() >= static_cast<u32>(th));
+}
+
+GSVector4i GSTextureCache::SourceRegion::GetRect(int tw, int th) const
+{
+ return GSVector4i(HasX() ? GetMinX() : 0, HasY() ? GetMinY() : 0, HasX() ? GetMaxX() : tw, HasY() ? GetMaxY() : th);
+}
+
+GSVector4i GSTextureCache::SourceRegion::GetOffset(int tw, int th) const
+{
+ const int xoffs = (GetMaxX() > static_cast<u32>(tw)) ? static_cast<int>(GetMinX()) : 0;
+ const int yoffs = (GetMaxY() > static_cast<u32>(th)) ? static_cast<int>(GetMinY()) : 0;
+ return GSVector4i(xoffs, yoffs, xoffs, yoffs);
+}
+
+GSTextureCache::SourceRegion GSTextureCache::SourceRegion::AdjustForMipmap(u32 level) const
+{
+ SourceRegion ret = {};
+ if (HasX())
+ {
+ const u32 new_minx = GetMinX() >> level;
+ const u32 new_maxx = ((GetMaxX() - 1) >> level) + 1;
+ ret.SetX(new_minx, new_maxx);
+ }
+ if (HasY())
+ {
+ const u32 new_miny = GetMinY() >> level;
+ const u32 new_maxy = ((GetMaxY() - 1) >> level) + 1;
+ ret.SetY(new_miny, new_maxy);
+ }
+ return ret;
+}
+
+void GSTextureCache::SourceRegion::AdjustTEX0(GIFRegTEX0* TEX0) const
+{
+ const GSOffset offset(GSLocalMemory::m_psm[TEX0->PSM].info, TEX0->TBP0, TEX0->TBW, TEX0->PSM);
+ TEX0->TBP0 += offset.bn(GetMinX(), GetMinY());
+}
+
using BlockHashState = XXH3_state_t;
__fi static void BlockHashReset(BlockHashState& st)
@@ -3696,16 +3809,16 @@ __fi static GSTextureCache::HashType FinishBlockHash(BlockHashState& st)
return GSXXH3_64bits_digest(&st);
}
-static void HashTextureLevel(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, BlockHashState& hash_st, u8* temp)
+static void HashTextureLevel(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, GSTextureCache::SourceRegion region, BlockHashState& hash_st, u8* temp)
{
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[TEX0.PSM];
const GSVector2i& bs = psm.bs;
- const int tw = 1 << TEX0.TW;
- const int th = 1 << TEX0.TH;
+ const int tw = region.HasX() ? region.GetWidth() : (1 << TEX0.TW);
+ const int th = region.HasY() ? region.GetHeight() : (1 << TEX0.TH);
// From GSLocalMemory foreachBlock(), used for reading textures.
// We want to hash the exact same blocks here.
- const GSVector4i rect(0, 0, tw, th);
+ const GSVector4i rect(region.GetRect(tw, th));
const GSVector4i block_rect(rect.ralign<Align_Outside>(bs));
GSLocalMemory& mem = g_gs_renderer->m_mem;
const GSOffset off = mem.GetOffset(TEX0.TBP0, TEX0.TBW, TEX0.PSM);
@@ -3717,7 +3830,7 @@ static void HashTextureLevel(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Blo
// the texture data with other textures/framebuffers/etc (which is common).
// Even though you might think this would be slower than just hashing for the hash
// cache, it actually ends up faster (unswizzling is faster than hashing).
- if (tw < bs.x || th < bs.y || psm.fmsk != 0xFFFFFFFFu)
+ if (tw < bs.x || th < bs.y || psm.fmsk != 0xFFFFFFFFu || region.GetMaxX() > 0 || region.GetMinY() > 0)
{
// Expand texture indices. Align to 32 bytes for AVX2.
const u32 pitch = Common::AlignUpPow2(static_cast<u32>(block_rect.z), 32);
@@ -3728,7 +3841,8 @@ static void HashTextureLevel(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Blo
rtx(mem, off, block_rect, temp, pitch, TEXA);
// Hash the expanded texture.
- u8* ptr = temp;
+ u8* ptr = temp + (pitch * static_cast<u32>(rect.top - block_rect.top)) +
+ static_cast<u32>(rect.left - block_rect.left);
if (pitch == row_size)
{
BlockHashAccumulate(hash_st, ptr, pitch * static_cast<u32>(th));
@@ -3741,8 +3855,6 @@ static void HashTextureLevel(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Blo
}
else
{
- BlockHashReset(hash_st);
-
GSOffset::BNHelper bn = off.bnMulti(block_rect.left, block_rect.top);
const int right = block_rect.right >> off.blockShiftX();
const int bottom = block_rect.bottom >> off.blockShiftY();
@@ -3758,27 +3870,27 @@ static void HashTextureLevel(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Blo
}
}
-GSTextureCache::HashType GSTextureCache::HashTexture(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA)
+GSTextureCache::HashType GSTextureCache::HashTexture(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, SourceRegion region)
{
BlockHashState hash_st;
BlockHashReset(hash_st);
- HashTextureLevel(TEX0, TEXA, hash_st, s_unswizzle_buffer);
+ HashTextureLevel(TEX0, TEXA, region, hash_st, s_unswizzle_buffer);
return FinishBlockHash(hash_st);
}
-void GSTextureCache::PreloadTexture(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, GSLocalMemory& mem, bool paltex, GSTexture* tex, u32 level)
+void GSTextureCache::PreloadTexture(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, SourceRegion region, GSLocalMemory& mem, bool paltex, GSTexture* tex, u32 level)
{
// m_TEX0 is adjusted for mips (messy, should be changed).
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[TEX0.PSM];
const GSVector2i& bs = psm.bs;
- const int tw = 1 << TEX0.TW;
- const int th = 1 << TEX0.TH;
+ const int tw = region.HasX() ? region.GetWidth() : (1 << TEX0.TW);
+ const int th = region.HasY() ? region.GetHeight() : (1 << TEX0.TH);
// Expand texture/apply palette.
- const GSVector4i rect(0, 0, tw, th);
+ const GSVector4i rect(region.GetRect(tw, th));
const GSVector4i block_rect(rect.ralign<Align_Outside>(bs));
const GSOffset off(mem.GetOffset(TEX0.TBP0, TEX0.TBW, TEX0.PSM));
- const int read_width = std::max(tw, psm.bs.x);
+ const int read_width = block_rect.width();
u32 pitch = static_cast<u32>(read_width) * sizeof(u32);
GSLocalMemory::readTexture rtx = psm.rtx;
if (paltex)
@@ -3788,8 +3900,9 @@ void GSTextureCache::PreloadTexture(const GIFRegTEX0& TEX0, const GIFRegTEXA& TE
}
// If we can stream it directly to GPU memory, do so, otherwise go through a temp buffer.
+ const GSVector4i unoffset_rect(0, 0, tw, th);
GSTexture::GSMap map;
- if (rect.eq(block_rect) && tex->Map(map, &rect, level))
+ if (rect.eq(block_rect) && tex->Map(map, &unoffset_rect, level))
{
rtx(mem, off, block_rect, map.bits, map.pitch, TEXA);
tex->Unmap();
@@ -3801,7 +3914,10 @@ void GSTextureCache::PreloadTexture(const GIFRegTEX0& TEX0, const GIFRegTEXA& TE
u8* buff = s_unswizzle_buffer;
rtx(mem, off, block_rect, buff, pitch, TEXA);
- tex->Update(rect, buff, pitch, level);
+
+ const u8* ptr = buff + (pitch * static_cast<u32>(rect.top - block_rect.top)) +
+ (static_cast<u32>(rect.left - block_rect.left) << (paltex ? 0 : 2));
+ tex->Update(unoffset_rect, ptr, pitch, level);
}
}
@@ -3813,7 +3929,7 @@ GSTextureCache::HashCacheKey::HashCacheKey()
TEXA.U64 = 0;
}
-GSTextureCache::HashCacheKey GSTextureCache::HashCacheKey::Create(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const u32* clut, const GSVector2i* lod)
+GSTextureCache::HashCacheKey GSTextureCache::HashCacheKey::Create(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const u32* clut, const GSVector2i* lod, SourceRegion region)
{
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[TEX0.PSM];
@@ -3821,12 +3937,13 @@ GSTextureCache::HashCacheKey GSTextureCache::HashCacheKey::Create(const GIFRegTE
ret.TEX0.U64 = TEX0.U64 & 0x00000007FFF00000ULL;
ret.TEXA.U64 = (psm.pal == 0 && psm.fmt > 0) ? (TEXA.U64 & 0x000000FF000080FFULL) : 0;
ret.CLUTHash = clut ? GSTextureCache::PaletteKeyHash{}({clut, psm.pal}) : 0;
+ ret.region = region;
BlockHashState hash_st;
BlockHashReset(hash_st);
// base level is always hashed
- HashTextureLevel(TEX0, TEXA, hash_st, s_unswizzle_buffer);
+ HashTextureLevel(TEX0, TEXA, region, hash_st, s_unswizzle_buffer);
if (lod)
{
@@ -3836,7 +3953,7 @@ GSTextureCache::HashCacheKey GSTextureCache::HashCacheKey::Create(const GIFRegTE
for (int i = 1; i < nmips; i++)
{
const GIFRegTEX0 MIP_TEX0{g_gs_renderer->GetTex0Layer(basemip + i)};
- HashTextureLevel(MIP_TEX0, TEXA, hash_st, s_unswizzle_buffer);
+ HashTextureLevel(MIP_TEX0, TEXA, region.AdjustForMipmap(i), hash_st, s_unswizzle_buffer);
}
}
@@ -3860,6 +3977,6 @@ void GSTextureCache::HashCacheKey::RemoveCLUTHash()
u64 GSTextureCache::HashCacheKeyHash::operator()(const HashCacheKey& key) const
{
std::size_t h = 0;
- HashCombine(h, key.TEX0Hash, key.CLUTHash, key.TEX0.U64, key.TEXA.U64);
+ HashCombine(h, key.TEX0Hash, key.CLUTHash, key.TEX0.U64, key.TEXA.U64, key.region.bits);
return h;
}
diff --git a/pcsx2/GS/Renderers/HW/GSTextureCache.h b/pcsx2/GS/Renderers/HW/GSTextureCache.h
index 61ea6dab88..0bb73ba770 100644
--- a/pcsx2/GS/Renderers/HW/GSTextureCache.h
+++ b/pcsx2/GS/Renderers/HW/GSTextureCache.h
@@ -44,6 +44,42 @@ public:
return valid && overlap;
}
+ struct SourceRegion
+ {
+ u64 bits;
+
+ bool HasX() const { return static_cast<u32>(bits) != 0; }
+ bool HasY() const { return static_cast<u32>(bits >> 32) != 0; }
+ bool HasEither() const { return (bits != 0); }
+
+ void SetX(u32 min, u32 max) { bits |= (min | (max << 16)); }
+ void SetY(u32 min, u32 max) { bits |= ((static_cast<u64>(min) << 32) | (static_cast<u64>(max) << 48)); }
+
+ u32 GetMinX() const { return static_cast<u32>(bits) & 0xFFFFu; }
+ u32 GetMaxX() const { return static_cast<u32>(bits >> 16) & 0xFFFFu; }
+ u32 GetMinY() const { return static_cast<u32>(bits >> 32) & 0xFFFFu; }
+ u32 GetMaxY() const { return static_cast<u32>(bits >> 48); }
+
+ u32 GetWidth() const { return (GetMaxX() - GetMinX()); }
+ u32 GetHeight() const { return (GetMaxY() - GetMinY()); }
+
+ /// Returns true if the area of the region exceeds the TW/TH size (i.e. "fixed tex0").
+ bool IsFixedTEX0(int tw, int th) const;
+
+ /// Returns the rectangle relative to the texture base pointer that the region occupies.
+ GSVector4i GetRect(int tw, int th) const;
+
+ /// When TW/TH is less than the extents covered by the region ("fixed tex0"), returns the offset
+ /// which should be applied to any coordinates to relocate them to the actual region.
+ GSVector4i GetOffset(int tw, int th) const;
+
+ /// Reduces the range of texels relative to the specified mipmap level.
+ SourceRegion AdjustForMipmap(u32 level) const;
+
+ /// Adjusts the texture base pointer and block width relative to the region.
+ void AdjustTEX0(GIFRegTEX0* TEX0) const;
+ };
+
using HashType = u64;
struct HashCacheKey
@@ -51,10 +87,11 @@ public:
HashType TEX0Hash, CLUTHash;
GIFRegTEX0 TEX0;
GIFRegTEXA TEXA;
+ SourceRegion region;
HashCacheKey();
- static HashCacheKey Create(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const u32* clut, const GSVector2i* lod);
+ static HashCacheKey Create(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const u32* clut, const GSVector2i* lod, SourceRegion region);
HashCacheKey WithRemovedCLUTHash() const;
void RemoveCLUTHash();
@@ -148,7 +185,7 @@ public:
{
GSVector4i* rect;
u32 count;
- } m_write;
+ } m_write = {};
void PreloadLevel(int level);
@@ -161,6 +198,7 @@ public:
GSTexture* m_palette;
GSVector4i m_valid_rect;
GSVector2i m_lod;
+ SourceRegion m_region = {};
u8 m_valid_hashes = 0;
u8 m_complete_layers = 0;
bool m_target;
@@ -178,11 +216,13 @@ public:
GSOffset::PageLooper m_pages;
public:
- Source(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, bool dummy_container = false);
+ Source(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA);
virtual ~Source();
__fi bool CanPreload() const { return CanPreloadTextureSize(m_TEX0.TW, m_TEX0.TH); }
+ void SetPages();
+
void Update(const GSVector4i& rect, int layer = 0);
void UpdateLayer(const GIFRegTEX0& TEX0, const GSVector4i& rect, int layer = 0);
@@ -322,7 +362,7 @@ protected:
std::unique_ptr<GSDownloadTexture> m_uint16_download_texture;
std::unique_ptr<GSDownloadTexture> m_uint32_download_texture;
- Source* CreateSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Target* t, bool half_right, int x_offset, int y_offset, const GSVector2i* lod, const GSVector4i* src_range, GSTexture* gpu_clut);
+ Source* CreateSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, Target* t, bool half_right, int x_offset, int y_offset, const GSVector2i* lod, const GSVector4i* src_range, GSTexture* gpu_clut, SourceRegion region);
Target* CreateTarget(const GIFRegTEX0& TEX0, int w, int h, int type, const bool clear);
/// Expands a target when the block pointer for a display framebuffer is within another target, but the read offset
@@ -332,10 +372,10 @@ protected:
/// Resizes the download texture if needed.
bool PrepareDownloadTexture(u32 width, u32 height, GSTexture::Format format, std::unique_ptr<GSDownloadTexture>* tex);
- HashCacheEntry* LookupHashCache(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, bool& paltex, const u32* clut, const GSVector2i* lod);
+ HashCacheEntry* LookupHashCache(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, bool& paltex, const u32* clut, const GSVector2i* lod, SourceRegion region);
- static void PreloadTexture(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, GSLocalMemory& mem, bool paltex, GSTexture* tex, u32 level);
- static HashType HashTexture(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA);
+ static void PreloadTexture(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, SourceRegion region, GSLocalMemory& mem, bool paltex, GSTexture* tex, u32 level);
+ static HashType HashTexture(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, SourceRegion region);
// TODO: virtual void Write(Source* s, const GSVector4i& r) = 0;
// TODO: virtual void Write(Target* t, const GSVector4i& r) = 0;
@@ -358,8 +398,8 @@ public:
GSTexture* LookupPaletteSource(u32 CBP, u32 CPSM, u32 CBW, GSVector2i& offset, const GSVector2i& size);
- Source* LookupSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GSVector4i& r, const GSVector2i* lod);
- Source* LookupDepthSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GSVector4i& r, bool palette = false);
+ Source* LookupSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, const GSVector2i* lod);
+ Source* LookupDepthSource(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, bool palette = false);
Target* LookupTarget(const GIFRegTEX0& TEX0, const GSVector2i& size, int type, bool used, u32 fbmask = 0, const bool is_frame = false, const int real_w = 0, const int real_h = 0, bool preload = GSConfig.PreloadFrameWithGSData);
Target* LookupDisplayTarget(const GIFRegTEX0& TEX0, const GSVector2i& size, const int real_w, const int real_h);
diff --git a/pcsx2/GS/Renderers/HW/GSTextureReplacements.cpp b/pcsx2/GS/Renderers/HW/GSTextureReplacements.cpp
index 2cee44524f..30abd4030f 100644
--- a/pcsx2/GS/Renderers/HW/GSTextureReplacements.cpp
+++ b/pcsx2/GS/Renderers/HW/GSTextureReplacements.cpp
@@ -42,6 +42,8 @@
// this is a #define instead of a variable to avoid warnings from non-literal format strings
#define TEXTURE_FILENAME_FORMAT_STRING "%" PRIx64 "-%08x"
#define TEXTURE_FILENAME_CLUT_FORMAT_STRING "%" PRIx64 "-%" PRIx64 "-%08x"
+#define TEXTURE_FILENAME_REGION_FORMAT_STRING "%" PRIx64 "-r%" PRIx64 "-" "-%08x"
+#define TEXTURE_FILENAME_REGION_CLUT_FORMAT_STRING "%" PRIx64 "-%" PRIx64 "-r%" PRIx64 "-%08x"
#define TEXTURE_REPLACEMENT_SUBDIRECTORY_NAME "replacements"
#define TEXTURE_DUMP_SUBDIRECTORY_NAME "dumps"
@@ -51,6 +53,7 @@ namespace
{
u64 TEX0Hash;
u64 CLUTHash;
+ GSTextureCache::SourceRegion region;
union
{
@@ -68,9 +71,10 @@ namespace
};
u32 miplevel;
- __fi u32 Width() const { return (1u << TEX0_TW); }
- __fi u32 Height() const { return (1u << TEX0_TH); }
+ __fi u32 Width() const { return (region.HasX() ? region.GetWidth() : (1u << TEX0_TW)); }
+ __fi u32 Height() const { return (region.HasY() ? region.GetWidth() : (1u << TEX0_TH)); }
__fi bool HasPalette() const { return (GSLocalMemory::m_psm[TEX0_PSM].pal > 0); }
+ __fi bool HasRegion() const { return region.HasEither(); }
__fi GSVector2 ReplacementScale(const GSTextureReplacements::ReplacementTexture& rtex) const
{
@@ -79,14 +83,27 @@ namespace
__fi GSVector2 ReplacementScale(u32 rwidth, u32 rheight) const
{
- return GSVector2(static_cast<float>(rwidth) / static_cast<float>(Width()), static_cast<float>(rheight) / static_cast<float>(Height()));
+ return GSVector2(static_cast<float>(rwidth) / static_cast<float>(Width()),
+ static_cast<float>(rheight) / static_cast<float>(Height()));
}
- __fi bool operator==(const TextureName& rhs) const { return std::tie(TEX0Hash, CLUTHash, bits) == std::tie(rhs.TEX0Hash, rhs.CLUTHash, rhs.bits); }
- __fi bool operator!=(const TextureName& rhs) const { return std::tie(TEX0Hash, CLUTHash, bits) != std::tie(rhs.TEX0Hash, rhs.CLUTHash, rhs.bits); }
- __fi bool operator<(const TextureName& rhs) const { return std::tie(TEX0Hash, CLUTHash, bits) < std::tie(rhs.TEX0Hash, rhs.CLUTHash, rhs.bits); }
+ __fi bool operator==(const TextureName& rhs) const
+ {
+ return std::tie(TEX0Hash, CLUTHash, region.bits, bits) ==
+ std::tie(rhs.TEX0Hash, rhs.CLUTHash, region.bits, rhs.bits);
+ }
+ __fi bool operator!=(const TextureName& rhs) const
+ {
+ return std::tie(TEX0Hash, CLUTHash, region.bits, bits) !=
+ std::tie(rhs.TEX0Hash, rhs.CLUTHash, region.bits, rhs.bits);
+ }
+ __fi bool operator<(const TextureName& rhs) const
+ {
+ return std::tie(TEX0Hash, CLUTHash, region.bits, bits) <
+ std::tie(rhs.TEX0Hash, rhs.CLUTHash, region.bits, rhs.bits);
+ }
};
- static_assert(sizeof(TextureName) == 24, "ReplacementTextureName is expected size");
+ static_assert(sizeof(TextureName) == 32, "ReplacementTextureName is expected size");
} // namespace
namespace std
@@ -97,7 +114,7 @@ namespace std
std::size_t operator()(const TextureName& val) const
{
std::size_t h = 0;
- HashCombine(h, val.TEX0Hash, val.CLUTHash, val.bits, val.miplevel);
+ HashCombine(h, val.TEX0Hash, val.CLUTHash, val.region.bits, val.bits, val.miplevel);
return h;
}
};
@@ -169,6 +186,7 @@ TextureName GSTextureReplacements::CreateTextureName(const GSTextureCache::HashC
name.TEX0Hash = hash.TEX0Hash;
name.CLUTHash = name.HasPalette() ? hash.CLUTHash : 0;
name.miplevel = miplevel;
+ name.region = hash.region;
return name;
}
@@ -184,6 +202,7 @@ GSTextureCache::HashCacheKey GSTextureReplacements::HashCacheKeyFromTextureName(
key.TEXA.TA1 = tn.TEXA_TA1;
key.TEX0Hash = tn.TEX0Hash;
key.CLUTHash = tn.HasPalette() ? tn.CLUTHash : 0;
+ key.region = tn.region;
return key;
}
@@ -192,15 +211,38 @@ std::optional<TextureName> GSTextureReplacements::ParseReplacementName(const std
TextureName ret;
ret.miplevel = 0;
- // TODO(Stenzek): Make this better.
char extension_dot;
- if (std::sscanf(filename.c_str(), TEXTURE_FILENAME_CLUT_FORMAT_STRING "%c", &ret.TEX0Hash, &ret.CLUTHash, &ret.bits, &extension_dot) != 4 || extension_dot != '.')
+ if (std::sscanf(filename.c_str(), TEXTURE_FILENAME_REGION_CLUT_FORMAT_STRING "%c", &ret.TEX0Hash, &ret.CLUTHash,
+ &ret.region.bits, &ret.bits, &extension_dot) == 5 &&
+ extension_dot == '.')
{
- if (std::sscanf(filename.c_str(), TEXTURE_FILENAME_FORMAT_STRING "%c", &ret.TEX0Hash, &ret.bits, &extension_dot) != 3 || extension_dot != '.')
- return std::nullopt;
+ return ret;
}
- return ret;
+ if (std::sscanf(filename.c_str(), TEXTURE_FILENAME_REGION_FORMAT_STRING "%c", &ret.TEX0Hash, &ret.region.bits,
+ &ret.bits, &extension_dot) == 4 &&
+ extension_dot == '.')
+ {
+ return ret;
+ }
+
+ ret.region.bits = 0;
+
+ if (std::sscanf(filename.c_str(), TEXTURE_FILENAME_CLUT_FORMAT_STRING "%c", &ret.TEX0Hash, &ret.CLUTHash, &ret.bits,
+ &extension_dot) == 4 &&
+ extension_dot == '.')
+ {
+ return ret;
+ }
+
+ if (std::sscanf(filename.c_str(), TEXTURE_FILENAME_FORMAT_STRING "%c", &ret.TEX0Hash, &ret.bits, &extension_dot) ==
+ 3 &&
+ extension_dot == '.')
+ {
+ return ret;
+ }
+
+ return std::nullopt;
}
std::string GSTextureReplacements::GetGameTextureDirectory()
@@ -229,23 +271,45 @@ std::string GSTextureReplacements::GetDumpFilename(const TextureName& name, u32
const std::string game_subdir(Path::Combine(game_dir, TEXTURE_DUMP_SUBDIRECTORY_NAME));
- if (name.HasPalette())
+ std::string filename;
+ if (name.HasRegion())
{
- const std::string filename(
- (level > 0) ?
- StringUtil::StdStringFromFormat(TEXTURE_FILENAME_CLUT_FORMAT_STRING "-mip%u.png", name.TEX0Hash, name.CLUTHash, name.bits, level) :
- StringUtil::StdStringFromFormat(TEXTURE_FILENAME_CLUT_FORMAT_STRING ".png", name.TEX0Hash, name.CLUTHash, name.bits));
- ret = Path::Combine(game_subdir, filename);
+ if (name.HasPalette())
+ {
+ filename = (level > 0) ?
+ StringUtil::StdStringFromFormat(TEXTURE_FILENAME_REGION_CLUT_FORMAT_STRING "-mip%u.png",
+ name.TEX0Hash, name.CLUTHash, name.region.bits, name.bits, level) :
+ StringUtil::StdStringFromFormat(TEXTURE_FILENAME_REGION_CLUT_FORMAT_STRING ".png",
+ name.TEX0Hash, name.CLUTHash, name.region.bits, name.bits);
+ }
+ else
+ {
+ filename = (level > 0) ? StringUtil::StdStringFromFormat(
+ TEXTURE_FILENAME_FORMAT_STRING "-mip%u.png", name.TEX0Hash, name.bits, level) :
+ StringUtil::StdStringFromFormat(
+ TEXTURE_FILENAME_FORMAT_STRING ".png", name.TEX0Hash, name.bits);
+ }
}
else
{
- const std::string filename(
- (level > 0) ?
- StringUtil::StdStringFromFormat(TEXTURE_FILENAME_FORMAT_STRING "-mip%u.png", name.TEX0Hash, name.bits, level) :
- StringUtil::StdStringFromFormat(TEXTURE_FILENAME_FORMAT_STRING ".png", name.TEX0Hash, name.bits));
- ret = Path::Combine(game_subdir, filename);
+ if (name.HasPalette())
+ {
+ filename = (level > 0) ? StringUtil::StdStringFromFormat(TEXTURE_FILENAME_CLUT_FORMAT_STRING "-mip%u.png",
+ name.TEX0Hash, name.CLUTHash, name.bits, level) :
+ StringUtil::StdStringFromFormat(TEXTURE_FILENAME_CLUT_FORMAT_STRING ".png",
+ name.TEX0Hash, name.CLUTHash, name.bits);
+ }
+ else
+ {
+ filename = (level > 0) ? StringUtil::StdStringFromFormat(
+ TEXTURE_FILENAME_FORMAT_STRING "-mip%u.png", name.TEX0Hash, name.bits, level) :
+ StringUtil::StdStringFromFormat(
+ TEXTURE_FILENAME_FORMAT_STRING ".png", name.TEX0Hash, name.bits);
+ }
}
+ ret = Path::Combine(game_subdir, filename);
+
return ret;
}
@@ -569,7 +633,8 @@ void GSTextureReplacements::ProcessAsyncLoadedTextures()
s_async_loaded_textures.clear();
}
-void GSTextureReplacements::DumpTexture(const GSTextureCache::HashCacheKey& hash, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, GSLocalMemory& mem, u32 level)
+void GSTextureReplacements::DumpTexture(const GSTextureCache::HashCacheKey& hash, const GIFRegTEX0& TEX0,
+ const GIFRegTEXA& TEXA, GSTextureCache::SourceRegion region, GSLocalMemory& mem, u32 level)
{
// check if it's been dumped or replaced already
const TextureName name(CreateTextureName(hash, level));
@@ -589,12 +654,12 @@ void GSTextureReplacements::DumpTexture(const GSTextureCache::HashCacheKey& hash
// compute width/height
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[TEX0.PSM];
const GSVector2i& bs = psm.bs;
- const int tw = 1 << TEX0.TW;
- const int th = 1 << TEX0.TH;
- const GSVector4i rect(0, 0, tw, th);
+ const int tw = region.HasX() ? region.GetWidth() : (1 << TEX0.TW);
+ const int th = region.HasY() ? region.GetHeight() : (1 << TEX0.TH);
+ const GSVector4i rect(region.GetRect(tw, th));
const GSVector4i block_rect(rect.ralign<Align_Outside>(bs));
- const int read_width = std::max(tw, psm.bs.x);
- const int read_height = std::max(th, psm.bs.y);
+ const int read_width = block_rect.width();
+ const int read_height = block_rect.height();
const u32 pitch = static_cast<u32>(read_width) * sizeof(u32);
// use per-texture buffer so we can compress the texture asynchronously and not block the GS thread
@@ -603,8 +668,9 @@ void GSTextureReplacements::DumpTexture(const GSTextureCache::HashCacheKey& hash
psm.rtx(mem, mem.GetOffset(TEX0.TBP0, TEX0.TBW, TEX0.PSM), block_rect, buffer.GetPtr(), pitch, TEXA);
// okay, now we can actually dump it
- QueueWorkerThreadItem([filename = std::move(filename), tw, th, pitch, buffer = std::move(buffer)]() {
- if (!SavePNGImage(filename.c_str(), tw, th, buffer.GetPtr(), pitch))
+ const u32 buffer_offset = ((rect.top - block_rect.top) * pitch) + ((rect.left - block_rect.left) * sizeof(u32));
+ QueueWorkerThreadItem([filename = std::move(filename), tw, th, pitch, buffer = std::move(buffer), buffer_offset]() {
+ if (!SavePNGImage(filename.c_str(), tw, th, buffer.GetPtr() + buffer_offset, pitch))
Console.Error("Failed to dump texture to '%s'.", filename.c_str());
});
}
diff --git a/pcsx2/GS/Renderers/HW/GSTextureReplacements.h b/pcsx2/GS/Renderers/HW/GSTextureReplacements.h
index cc61be7702..39a8e0c8e7 100644
--- a/pcsx2/GS/Renderers/HW/GSTextureReplacements.h
+++ b/pcsx2/GS/Renderers/HW/GSTextureReplacements.h
@@ -52,7 +52,8 @@ namespace GSTextureReplacements
GSTexture* CreateReplacementTexture(const ReplacementTexture& rtex, const GSVector2& scale, bool mipmap);
void ProcessAsyncLoadedTextures();
- void DumpTexture(const GSTextureCache::HashCacheKey& hash, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, GSLocalMemory& mem, u32 level);
+ void DumpTexture(const GSTextureCache::HashCacheKey& hash, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA,
+ GSTextureCache::SourceRegion region, GSLocalMemory& mem, u32 level);
void ClearDumpedTextureList();
/// Loader will take a filename and interpret the format (e.g. DDS, PNG, etc).
diff --git a/pcsx2/GS/Renderers/Metal/GSDeviceMTL.mm b/pcsx2/GS/Renderers/Metal/GSDeviceMTL.mm
index 85f011b148..a2987d1b72 100644
--- a/pcsx2/GS/Renderers/Metal/GSDeviceMTL.mm
+++ b/pcsx2/GS/Renderers/Metal/GSDeviceMTL.mm
@@ -1374,6 +1374,8 @@ void GSDeviceMTL::MRESetHWPipelineState(GSHWDrawConfig::VSSelector vssel, GSHWDr
setFnConstantB(m_fn_constants, pssel.tcc, GSMTLConstantIndex_PS_TCC);
setFnConstantI(m_fn_constants, pssel.wms, GSMTLConstantIndex_PS_WMS);
setFnConstantI(m_fn_constants, pssel.wmt, GSMTLConstantIndex_PS_WMT);
+ setFnConstantB(m_fn_constants, pssel.adjs, GSMTLConstantIndex_PS_ADJS);
+ setFnConstantB(m_fn_constants, pssel.adjt, GSMTLConstantIndex_PS_ADJT);
setFnConstantB(m_fn_constants, pssel.ltf, GSMTLConstantIndex_PS_LTF);
setFnConstantB(m_fn_constants, pssel.shuffle, GSMTLConstantIndex_PS_SHUFFLE);
setFnConstantB(m_fn_constants, pssel.read_ba, GSMTLConstantIndex_PS_READ_BA);
@@ -1403,7 +1405,6 @@ void GSDeviceMTL::MRESetHWPipelineState(GSHWDrawConfig::VSSelector vssel, GSHWDr
setFnConstantB(m_fn_constants, pssel.automatic_lod, GSMTLConstantIndex_PS_AUTOMATIC_LOD);
setFnConstantB(m_fn_constants, pssel.manual_lod, GSMTLConstantIndex_PS_MANUAL_LOD);
setFnConstantB(m_fn_constants, pssel.point_sampler, GSMTLConstantIndex_PS_POINT_SAMPLER);
- setFnConstantB(m_fn_constants, pssel.invalid_tex0, GSMTLConstantIndex_PS_INVALID_TEX0);
setFnConstantI(m_fn_constants, pssel.scanmsk, GSMTLConstantIndex_PS_SCANMSK);
auto newps = LoadShader(@"ps_main");
ps = newps;
@@ -1594,10 +1595,10 @@ static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, WH) == of
static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, TA_MaxDepth_Af.x) == offsetof(GSMTLMainPSUniform, ta));
static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, TA_MaxDepth_Af.z) == offsetof(GSMTLMainPSUniform, max_depth));
static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, TA_MaxDepth_Af.w) == offsetof(GSMTLMainPSUniform, alpha_fix));
-static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, MskFix) == offsetof(GSMTLMainPSUniform, uv_msk_fix));
static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, FbMask) == offsetof(GSMTLMainPSUniform, fbmask));
static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, HalfTexel) == offsetof(GSMTLMainPSUniform, half_texel));
static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, MinMax) == offsetof(GSMTLMainPSUniform, uv_min_max));
+static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, STRange) == offsetof(GSMTLMainPSUniform, st_range));
static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, ChannelShuffle) == offsetof(GSMTLMainPSUniform, channel_shuffle));
static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, TCOffsetHack) == offsetof(GSMTLMainPSUniform, tc_offset));
static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, STScale) == offsetof(GSMTLMainPSUniform, st_scale));
diff --git a/pcsx2/GS/Renderers/Metal/GSMTLSharedHeader.h b/pcsx2/GS/Renderers/Metal/GSMTLSharedHeader.h
index 6d22d1b2d1..d8c2265c37 100644
--- a/pcsx2/GS/Renderers/Metal/GSMTLSharedHeader.h
+++ b/pcsx2/GS/Renderers/Metal/GSMTLSharedHeader.h
@@ -108,11 +108,15 @@ struct GSMTLMainPSUniform
vector_float2 ta;
float max_depth;
float alpha_fix;
- vector_uint4 uv_msk_fix;
vector_uint4 fbmask;
vector_float4 half_texel;
- vector_float4 uv_min_max;
+ union
+ {
+ vector_float4 uv_min_max;
+ vector_uint4 uv_msk_fix;
+ };
+ vector_float4 st_range;
struct
{
unsigned int blue_mask;
@@ -166,6 +170,8 @@ enum GSMTLFnConstants
GSMTLConstantIndex_PS_TCC,
GSMTLConstantIndex_PS_WMS,
GSMTLConstantIndex_PS_WMT,
+ GSMTLConstantIndex_PS_ADJS,
+ GSMTLConstantIndex_PS_ADJT,
GSMTLConstantIndex_PS_LTF,
GSMTLConstantIndex_PS_SHUFFLE,
GSMTLConstantIndex_PS_READ_BA,
@@ -194,6 +200,5 @@ enum GSMTLFnConstants
GSMTLConstantIndex_PS_AUTOMATIC_LOD,
GSMTLConstantIndex_PS_MANUAL_LOD,
GSMTLConstantIndex_PS_POINT_SAMPLER,
- GSMTLConstantIndex_PS_INVALID_TEX0,
GSMTLConstantIndex_PS_SCANMSK,
};
diff --git a/pcsx2/GS/Renderers/Metal/tfx.metal b/pcsx2/GS/Renderers/Metal/tfx.metal
index 1daa381f84..4a46cceaa8 100644
--- a/pcsx2/GS/Renderers/Metal/tfx.metal
+++ b/pcsx2/GS/Renderers/Metal/tfx.metal
@@ -37,6 +37,8 @@ constant uint PS_TFX [[function_constant(GSMTLConstantIndex_PS_TF
constant bool PS_TCC [[function_constant(GSMTLConstantIndex_PS_TCC)]];
constant uint PS_WMS [[function_constant(GSMTLConstantIndex_PS_WMS)]];
constant uint PS_WMT [[function_constant(GSMTLConstantIndex_PS_WMT)]];
+constant bool PS_ADJS [[function_constant(GSMTLConstantIndex_PS_ADJS)]];
+constant bool PS_ADJT [[function_constant(GSMTLConstantIndex_PS_ADJT)]];
constant bool PS_LTF [[function_constant(GSMTLConstantIndex_PS_LTF)]];
constant bool PS_SHUFFLE [[function_constant(GSMTLConstantIndex_PS_SHUFFLE)]];
constant bool PS_READ_BA [[function_constant(GSMTLConstantIndex_PS_READ_BA)]];
@@ -65,7 +67,6 @@ constant bool PS_TEX_IS_FB [[function_constant(GSMTLConstantIndex_PS_TE
constant bool PS_AUTOMATIC_LOD [[function_constant(GSMTLConstantIndex_PS_AUTOMATIC_LOD)]];
constant bool PS_MANUAL_LOD [[function_constant(GSMTLConstantIndex_PS_MANUAL_LOD)]];
constant bool PS_POINT_SAMPLER [[function_constant(GSMTLConstantIndex_PS_POINT_SAMPLER)]];
-constant bool PS_INVALID_TEX0 [[function_constant(GSMTLConstantIndex_PS_INVALID_TEX0)]];
constant uint PS_SCANMSK [[function_constant(GSMTLConstantIndex_PS_SCANMSK)]];
constant GSMTLExpandType VS_EXPAND_TYPE = static_cast<GSMTLExpandType>(VS_EXPAND_TYPE_RAW);
@@ -321,7 +322,21 @@ struct PSMain
// As of 2018 this issue is still present.
uv = (trunc(uv * cb.wh.zw) + 0.5) / cb.wh.zw;
}
- uv *= cb.st_scale;
+ if (!PS_ADJS && !PS_ADJT)
+ {
+ uv *= cb.st_scale;
+ }
+ else
+ {
+ if (PS_ADJS)
+ uv.x = (uv.x - cb.st_range.x) * cb.st_range.z;
+ else
+ uv.x = uv.x * cb.st_scale.x;
+ if (PS_ADJT)
+ uv.y = (uv.y - cb.st_range.y) * cb.st_range.w;
+ else
+ uv.y = uv.y * cb.st_scale.y;
+ }
if (PS_AUTOMATIC_LOD)
{
@@ -360,7 +375,7 @@ struct PSMain
float4 clamp_wrap_uv(float4 uv)
{
float4 uv_out = uv;
- float4 tex_size = PS_INVALID_TEX0 ? cb.wh.zwzw : cb.wh.xyxy;
+ float4 tex_size = cb.wh.xyxy;
if (PS_WMS == PS_WMT)
{
@@ -724,12 +739,7 @@ struct PSMain
float4 ps_color()
{
float2 st, st_int;
- if (!FST && PS_INVALID_TEX0)
- {
- st = (in.t.xy * cb.wh.xy) / (in.t.w * cb.wh.zw);
- st_int = (in.ti.zw * cb.wh.xy) / (in.t.w * cb.wh.zw);
- }
- else if (!FST)
+ if (!FST)
{
st = in.t.xy / in.t.w;
st_int = in.ti.zw / in.t.w;
diff --git a/pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp b/pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp
index 7fb4074408..7523c27170 100644
--- a/pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp
+++ b/pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp
@@ -1029,6 +1029,8 @@ std::string GSDeviceOGL::GetPSSource(const PSSelector& sel)
std::string macro = fmt::format("#define PS_FST {}\n", sel.fst)
+ fmt::format("#define PS_WMS {}\n", sel.wms)
+ fmt::format("#define PS_WMT {}\n", sel.wmt)
+ + fmt::format("#define PS_ADJS {}\n", sel.adjs)
+ + fmt::format("#define PS_ADJT {}\n", sel.adjt)
+ fmt::format("#define PS_AEM_FMT {}\n", sel.aem_fmt)
+ fmt::format("#define PS_PAL_FMT {}\n", sel.pal_fmt)
+ fmt::format("#define PS_DFMT {}\n", sel.dfmt)
@@ -1037,7 +1039,6 @@ std::string GSDeviceOGL::GetPSSource(const PSSelector& sel)
+ fmt::format("#define PS_URBAN_CHAOS_HLE {}\n", sel.urban_chaos_hle)
+ fmt::format("#define PS_TALES_OF_ABYSS_HLE {}\n", sel.tales_of_abyss_hle)
+ fmt::format("#define PS_TEX_IS_FB {}\n", sel.tex_is_fb)
- + fmt::format("#define PS_INVALID_TEX0 {}\n", sel.invalid_tex0)
+ fmt::format("#define PS_AEM {}\n", sel.aem)
+ fmt::format("#define PS_TFX {}\n", sel.tfx)
+ fmt::format("#define PS_TCC {}\n", sel.tcc)
diff --git a/pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp b/pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp
index 1a1ee0155d..0e4bf4f6a5 100644
--- a/pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp
+++ b/pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp
@@ -1948,6 +1948,8 @@ VkShaderModule GSDeviceVK::GetTFXFragmentShader(const GSHWDrawConfig::PSSelector
AddMacro(ss, "PS_FST", sel.fst);
AddMacro(ss, "PS_WMS", sel.wms);
AddMacro(ss, "PS_WMT", sel.wmt);
+ AddMacro(ss, "PS_ADJS", sel.adjs);
+ AddMacro(ss, "PS_ADJT", sel.adjt);
AddMacro(ss, "PS_AEM_FMT", sel.aem_fmt);
AddMacro(ss, "PS_PAL_FMT", sel.pal_fmt);
AddMacro(ss, "PS_DFMT", sel.dfmt);
@@ -1955,7 +1957,6 @@ VkShaderModule GSDeviceVK::GetTFXFragmentShader(const GSHWDrawConfig::PSSelector
AddMacro(ss, "PS_CHANNEL_FETCH", sel.channel);
AddMacro(ss, "PS_URBAN_CHAOS_HLE", sel.urban_chaos_hle);
AddMacro(ss, "PS_TALES_OF_ABYSS_HLE", sel.tales_of_abyss_hle);
- AddMacro(ss, "PS_INVALID_TEX0", sel.invalid_tex0);
AddMacro(ss, "PS_AEM", sel.aem);
AddMacro(ss, "PS_TFX", sel.tfx);
AddMacro(ss, "PS_TCC", sel.tcc);
diff --git a/pcsx2/ShaderCacheVersion.h b/pcsx2/ShaderCacheVersion.h
index 602ed0d8bf..af2c5fe8dd 100644
--- a/pcsx2/ShaderCacheVersion.h
+++ b/pcsx2/ShaderCacheVersion.h
@@ -15,4 +15,4 @@
/// Version number for GS and other shaders. Increment whenever any of the contents of the
/// shaders change, to invalidate the cache.
-static constexpr u32 SHADER_CACHE_VERSION = 11;
+static constexpr u32 SHADER_CACHE_VERSION = 12;