GS/HW: Enable feedback loop if channel shuffle enabled barriers.

Also changed LookupSource/Depth to receive the whole FRAME struct, not just the pointer.

.github/PULL_REQUEST_TEMPLATE.md

@@ -6,3 +6,9 @@
 
 ### Suggested Testing Steps
 <!-- If applicable, including examples you've already tested with / recommendations for how to test further is very helpful! -->
+
+### Did you use AI to help find, test, or implement this issue or feature?
+<!-- Insert an X in one of the boxes. This is a required field. -->
+- [ ] No, I did not use AI.
+
+- [ ] Yes (please explain briefly):

pcsx2-qt/Translations/pcsx2-qt_en.ts

@@ -10786,67 +10786,67 @@ Do you want to load this save and continue?</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="452"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="464"/>
         <source>Saving screenshot to &apos;{}&apos;.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="464"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="476"/>
         <source>Saved screenshot to &apos;{}&apos;.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="471"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="483"/>
         <source>Failed to save screenshot to &apos;{}&apos;.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="542"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="554"/>
         <source>Host GPU device encountered an error and was recovered. This may have broken rendering.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="647"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="659"/>
         <source>CAS is not available, your graphics driver does not support the required functionality.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="702"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="714"/>
         <source>with no compression</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="709"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="721"/>
         <source>with LZMA compression</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="716"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="728"/>
         <source>with Zstandard compression</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="722"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="734"/>
         <source>Saving {0} GS dump {1} to &apos;{2}&apos;</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="723"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="735"/>
         <source>single frame</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="723"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="735"/>
         <source>multi-frame</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="743"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="755"/>
         <source>Failed to render/download screenshot.</source>
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="754"/>
+        <location filename="../../pcsx2/GS/Renderers/Common/GSRenderer.cpp" line="766"/>
         <source>Saved GS dump to &apos;{}&apos;.</source>
         <translation type="unfinished"></translation>
     </message>

pcsx2/GS/Renderers/Common/GSDevice.h

@@ -372,6 +372,7 @@ struct alignas(16) GSHWDrawConfig
 
 				// Others ways to fetch the texture
 				u32 channel : 3;
+				u32 channel_fb : 1;
 
 				// Dithering
 				u32 dither : 2;
@@ -410,7 +411,7 @@ struct alignas(16) GSHWDrawConfig
 		{
 			const u32 sw_blend_bits = blend_a | blend_b | blend_d;
 			const bool sw_blend_needs_rt = (sw_blend_bits != 0 && ((sw_blend_bits | blend_c) & 1u)) || ((a_masked & blend_c) != 0);
-			return tex_is_fb || fbmask || (date > 0 && date != 3) || sw_blend_needs_rt;
+			return channel_fb || tex_is_fb || fbmask || (date > 0 && date != 3) || sw_blend_needs_rt;
 		}
 
 		/// Disables color output from the pixel shader, this is done when all channels are masked.

pcsx2/GS/Renderers/HW/GSRendererHW.cpp

@@ -2920,9 +2920,9 @@ void GSRendererHW::Draw()
 		}
 		else
 		{
-			src = tex_psm.depth ? g_texture_cache->LookupDepthSource(true, TEX0, env.TEXA, MIP_CLAMP, tmm.coverage, possible_shuffle, m_vt.IsLinear(), m_cached_ctx.FRAME.Block(), req_color, req_alpha) :
+			src = tex_psm.depth ? g_texture_cache->LookupDepthSource(true, TEX0, env.TEXA, MIP_CLAMP, tmm.coverage, possible_shuffle, m_vt.IsLinear(), m_cached_ctx.FRAME, req_color, req_alpha) :
 								  g_texture_cache->LookupSource(true, TEX0, env.TEXA, MIP_CLAMP, tmm.coverage, (GSConfig.HWMipmap || GSConfig.TriFilter == TriFiltering::Forced) ? &hash_lod_range : nullptr,
-					possible_shuffle, m_vt.IsLinear(), m_cached_ctx.FRAME.Block(), req_color, req_alpha);
+					possible_shuffle, m_vt.IsLinear(), m_cached_ctx.FRAME, req_color, req_alpha);
 
 			if (!src) [[unlikely]]
 			{
@@ -3212,7 +3212,7 @@ void GSRendererHW::Draw()
 				CleanupDraw(true);
 				return;
 			}
-			else if (IsPageCopy() && src->m_from_target && m_cached_ctx.TEX0.TBP0 >= src->m_from_target->m_TEX0.TBP0)
+			else if (IsPageCopy() && src->m_from_target && m_cached_ctx.TEX0.TBP0 >= src->m_from_target->m_TEX0.TBP0 && m_cached_ctx.FRAME.FBW < ((src->m_from_target->m_TEX0.TBW + 1) >> 1))
 			{
 				FRAME_TEX0.TBW = src->m_from_target->m_TEX0.TBW;
 			}
@@ -3264,8 +3264,6 @@ void GSRendererHW::Draw()
 				const GSVector4i new_drect = GSVector4i(0, new_offset * rt->m_scale, new_size.x * rt->m_scale, new_size.y * rt->m_scale);
 				rt->ResizeTexture(new_size.x, new_size.y, true, true, new_drect);
 
-				g_texture_cache->CombineAlignedInsideTargets(rt, src);
-
 				if (src && src->m_from_target && src->m_from_target == rt && src->m_target_direct)
 				{
 					src->m_texture = rt->m_texture;
@@ -3283,6 +3281,8 @@ void GSRendererHW::Draw()
 				rt->m_drawn_since_read.y += new_offset;
 				rt->m_drawn_since_read.w += new_offset;
 
+				g_texture_cache->CombineAlignedInsideTargets(rt, src);
+
 				if (rt->m_dirty.size())
 				{
 					for (int i = 0; i < static_cast<int>(rt->m_dirty.size()); i++)
@@ -5030,6 +5030,8 @@ __ri bool GSRendererHW::EmulateChannelShuffle(GSTextureCache::Target* src, bool
 		// Hitman suffers from this, not sure on the exact scenario at the moment, but we need the barrier.
 		if (PRIM->ABE && m_context->ALPHA.IsCdInBlend())
 		{
+			// Needed to enable IsFeedbackLoop.
+			m_conf.ps.channel_fb = 1;
 			// Assume no overlap when it's a channel shuffle, no need for full barriers.
 			m_conf.require_one_barrier = true;
 		}

pcsx2/GS/Renderers/HW/GSTextureCache.cpp

@@ -998,7 +998,7 @@ __ri static GSTextureCache::Source* FindSourceInMap(const GIFRegTEX0& TEX0, cons
 	return nullptr;
 }
 
-GSTextureCache::Source* GSTextureCache::LookupDepthSource(const bool is_depth, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, const bool possible_shuffle, const bool linear, const u32 frame_fbp, bool req_color, bool req_alpha, bool palette)
+GSTextureCache::Source* GSTextureCache::LookupDepthSource(const bool is_depth, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, const bool possible_shuffle, const bool linear, const GIFRegFRAME& frame, bool req_color, bool req_alpha, bool palette)
 {
 	if (GSConfig.UserHacks_DisableDepthSupport)
 	{
@@ -1183,7 +1183,7 @@ GSTextureCache::Source* GSTextureCache::LookupDepthSource(const bool is_depth, c
 	else
 	{
 		// This is a bit of a worry, since it could load junk from local memory... but it's better than skipping the draw.
-		return is_depth ? LookupSource(false, TEX0, TEXA, CLAMP, r, nullptr, possible_shuffle, linear, frame_fbp, req_color, req_alpha) : nullptr;
+		return is_depth ? LookupSource(false, TEX0, TEXA, CLAMP, r, nullptr, possible_shuffle, linear, frame, req_color, req_alpha) : nullptr;
 	}
 
 	pxAssert(src->m_texture);
@@ -1192,7 +1192,7 @@ GSTextureCache::Source* GSTextureCache::LookupDepthSource(const bool is_depth, c
 	return src;
 }
 
-GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, const GSVector2i* lod, const bool possible_shuffle, const bool linear, const u32 frame_fbp, bool req_color, bool req_alpha)
+GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, const GSVector2i* lod, const bool possible_shuffle, const bool linear, const GIFRegFRAME& frame, bool req_color, bool req_alpha)
 {
 	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);
 
@@ -1456,7 +1456,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 				else
 				{
 					rect_clean = t->m_dirty.empty();
-					if (!possible_shuffle && frame_fbp != t->m_TEX0.TBP0 && rect_clean && bp == t->m_TEX0.TBP0 && t && GSUtil::HasCompatibleBits(psm, t->m_TEX0.PSM) && width_match && real_fmt_match)
+					if (!possible_shuffle && frame.Block() != t->m_TEX0.TBP0 && rect_clean && bp == t->m_TEX0.TBP0 && t && GSUtil::HasCompatibleBits(psm, t->m_TEX0.PSM) && width_match && real_fmt_match)
 					{
 						if (!tex_merge_rt && t->Overlaps(bp, bw, psm, req_rect))
 						{
@@ -1485,7 +1485,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 					}
 				}
 
-				if (t->m_TEX0.TBP0 != frame_fbp && !possible_shuffle && bp > t->m_TEX0.TBP0 && t->Overlaps(bp, bw, psm, req_rect) && GSUtil::GetChannelMask(psm) == GSUtil::GetChannelMask(t->m_TEX0.PSM) && !width_match)
+				if (t->m_TEX0.TBP0 != frame.Block() && !possible_shuffle && bp > t->m_TEX0.TBP0 && t->Overlaps(bp, bw, psm, req_rect) && GSUtil::GetChannelMask(psm) == GSUtil::GetChannelMask(t->m_TEX0.PSM) && !width_match)
 				{
 					GSVector4i new_rect = req_rect;
 
@@ -1710,11 +1710,12 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 
 						GSVector4i new_rect = (GSLocalMemory::m_psm[color_psm].bpp != GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp && (psm & 0x7) != PSMCT16) ? block_boundary_rect : rect;
 
-						// Check if it is possible to hit with valid <x,y> offset on the given Target.
-						// Fixes Jak eyes rendering.
-						// Fixes Xenosaga 3 last dungeon graphic bug.
-						// Fixes Pause menu in The Getaway.
-						const bool can_translate = CanTranslate(bp, bw, src_psm, new_rect, t->m_TEX0.TBP0, t->m_TEX0.PSM, t->m_TEX0.TBW);
+						// If the sizing is completely wrong on the frame vs the source when reading from alpha then it's likely the target has 2 different sizes for rgb and alpha.
+						// This is just changing the target width for the rect translation, it has no bearing on the actual source read or the target itself.
+						// Hitman Blood Money is an example of this in the theatre.
+						const u32 rt_tbw = (possible_shuffle || bw == 1 || GSUtil::GetChannelMask(psm) != 0x8 || frame.FBW <= bw || frame.FBW == t->m_TEX0.TBW) ? t->m_TEX0.TBW : frame.FBW;
+
+						const bool can_translate = CanTranslate(bp, bw, src_psm, new_rect, t->m_TEX0.TBP0, t->m_TEX0.PSM, rt_tbw);
 						if (can_translate)
 						{
 							const bool swizzle_match = GSLocalMemory::m_psm[src_psm].depth == GSLocalMemory::m_psm[t->m_TEX0.PSM].depth;
@@ -1724,7 +1725,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 
 							if (swizzle_match)
 							{
-								rect = TranslateAlignedRectByPage(t, bp, src_psm, bw, new_rect);
+								rect = TranslateAlignedRectByPage(t->m_TEX0.TBP0, t->m_end_block, rt_tbw, t->m_TEX0.PSM, bp, src_psm, bw, new_rect);
 								rect.x -= new_rect.x;
 								rect.y -= new_rect.y;
 							}
@@ -1773,7 +1774,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 							dst = t;
 							tex_merge_rt = false;
 							found_t = true;
-							if (dst->m_TEX0.TBP0 == frame_fbp && possible_shuffle)
+							if (dst->m_TEX0.TBP0 == frame.Block() && possible_shuffle)
 								break;
 							else
 								continue;
@@ -1799,7 +1800,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 								tex_merge_rt = false;
 								found_t = true;
 								// Keep looking, just in case there is an exact match (Situation: Target frame drawn inside target frame, current makes a separate texture)
-								if (dst->m_TEX0.TBP0 == frame_fbp && possible_shuffle)
+								if (dst->m_TEX0.TBP0 == frame.Block() && possible_shuffle)
 									break;
 								else
 									continue;
@@ -1856,7 +1857,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 								break;
 							}
 
-							if (dst->m_TEX0.TBP0 == frame_fbp && possible_shuffle)
+							if (dst->m_TEX0.TBP0 == frame.Block() && possible_shuffle)
 								break;
 							else
 								continue;
@@ -1873,7 +1874,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 
 							// Prefer a target inside over a target outside.
 							found_t = false;
-							if (dst->m_TEX0.TBP0 == frame_fbp && possible_shuffle)
+							if (dst->m_TEX0.TBP0 == frame.Block() && possible_shuffle)
 								break;
 							else
 								continue;
@@ -1934,7 +1935,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 							GIFRegTEX0 depth_TEX0;
 							depth_TEX0.U32[0] = TEX0.U32[0] | (0x30u << 20u);
 							depth_TEX0.U32[1] = TEX0.U32[1];
-							src = LookupDepthSource(false, depth_TEX0, TEXA, CLAMP, block_boundary_rect, possible_shuffle, linear, frame_fbp, req_color, req_alpha);
+							src = LookupDepthSource(false, depth_TEX0, TEXA, CLAMP, block_boundary_rect, possible_shuffle, linear, frame, req_color, req_alpha);
 
 							if (src != nullptr)
 							{
@@ -1956,7 +1957,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 							}
 							else
 							{
-								src = LookupDepthSource(false, TEX0, TEXA, CLAMP, block_boundary_rect, possible_shuffle, linear, frame_fbp, req_color, req_alpha, true);
+								src = LookupDepthSource(false, TEX0, TEXA, CLAMP, block_boundary_rect, possible_shuffle, linear, frame, req_color, req_alpha, true);
 
 								if (src != nullptr)
 								{
@@ -2096,7 +2097,7 @@ void GSTextureCache::CombineAlignedInsideTargets(Target* target, GSTextureCache:
 		if (t != target)
 		{
 			// Target not contained, skip it.
-			if (t->m_TEX0.TBP0 < target->m_TEX0.TBP0 || t->m_end_block > target->m_end_block)
+			if (t->m_TEX0.TBP0 < target->m_TEX0.TBP0 || t->UnwrappedEndBlock() > target->UnwrappedEndBlock())
 			{
 				i++;
 				continue;
@@ -2112,16 +2113,24 @@ void GSTextureCache::CombineAlignedInsideTargets(Target* target, GSTextureCache:
 				{
 					if (t->m_last_draw > target->m_last_draw || t->m_valid.rintersect(target->m_valid).rempty())
 					{
-						t->Update();
-						const u32 vertical_offset = (((t->m_TEX0.TBP0 - target->m_TEX0.TBP0) >> 5) / std::max(1U, t->m_TEX0.TBW)) * t_psm.pgs.y;
-						const u32 horizontal_offset = page_offset * t_psm.pgs.x;
-						const GSVector4i target_drect_unscaled = t->m_valid + GSVector4i(horizontal_offset, vertical_offset).xyxy();
-						const GSVector4 target_drect = GSVector4(target_drect_unscaled) * target->m_scale;
-						const bool valid_color = t->m_valid_rgb;
-						const bool valid_alpha = (t->m_valid_alpha_high | t->m_valid_alpha_low) && (GSUtil::GetChannelMask(t->m_TEX0.PSM) & 0x8);
-						g_gs_device->StretchRect(t->m_texture, GSVector4(0, 0, 1, 1), target->m_texture, target_drect, valid_color, valid_color, valid_color, valid_alpha, (target->m_type == RenderTarget) ? ShaderConvert::COPY : ShaderConvert::DEPTH_COPY);
+						if (!t->m_drawn_since_read.rempty())
+						{
+							t->Update();
 
-						target->UpdateValidity(target_drect_unscaled);
+							const u32 vertical_offset = (((t->m_TEX0.TBP0 - target->m_TEX0.TBP0) >> 5) / std::max(1U, t->m_TEX0.TBW)) * t_psm.pgs.y;
+							const u32 horizontal_offset = page_offset * t_psm.pgs.x;
+							const GSVector4i target_drect_unscaled = t->m_drawn_since_read + GSVector4i(horizontal_offset, vertical_offset).xyxy();
+
+							const GSVector4 source_rect = GSVector4(t->m_drawn_since_read) / (GSVector4(t->m_unscaled_size) * t->GetScale());
+							const GSVector4 target_drect = GSVector4(target_drect_unscaled) * target->m_scale;
+
+							const bool valid_color = t->m_valid_rgb;
+							const bool valid_alpha = (t->m_valid_alpha_high | t->m_valid_alpha_low) && (GSUtil::GetChannelMask(t->m_TEX0.PSM) & 0x8);
+
+							g_gs_device->StretchRect(t->m_texture, source_rect, target->m_texture, target_drect, valid_color, valid_color, valid_color, valid_alpha, (target->m_type == RenderTarget) ? ShaderConvert::COPY : ShaderConvert::DEPTH_COPY);
+
+							target->UpdateValidity(target_drect_unscaled);
+						}
 					}
 
 					if (src && src->m_from_target == t)

pcsx2/GS/Renderers/HW/GSTextureCache.h

@@ -494,8 +494,8 @@ public:
 	GSTexture* LookupPaletteSource(u32 CBP, u32 CPSM, u32 CBW, GSVector2i& offset, float* scale, const GSVector2i& size);
 	std::shared_ptr<Palette> LookupPaletteObject(const u32* clut, u16 pal, bool need_gs_texture);
 
-	Source* LookupSource(const bool is_color, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, const GSVector2i* lod, const bool possible_shuffle, const bool linear, const u32 frame_fbp = 0xFFFFFFFF, bool req_color = true, bool req_alpha = true);
-	Source* LookupDepthSource(const bool is_depth, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, const bool possible_shuffle, const bool linear, const u32 frame_fbp = 0xFFFFFFFF, bool req_color = true, bool req_alpha = true, bool palette = false);
+	Source* LookupSource(const bool is_color, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, const GSVector2i* lod, const bool possible_shuffle, const bool linear, const GIFRegFRAME& frame, bool req_color = true, bool req_alpha = true);
+	Source* LookupDepthSource(const bool is_depth, const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const GIFRegCLAMP& CLAMP, const GSVector4i& r, const bool possible_shuffle, const bool linear, const GIFRegFRAME& frame, bool req_color = true, bool req_alpha = true, bool palette = false);
 
 	Target* FindTargetOverlap(Target* target, int type, int psm);
 	void CombineAlignedInsideTargets(Target* target, GSTextureCache::Source* src = nullptr);