GS/TC: Correct some preload behaviour on merging targets

bin/resources/shaders/dx11/tfx.fx

@@ -168,6 +168,7 @@ cbuffer cb1
 	float4 LODParams;
 	float4 STRange;
 	int4 ChannelShuffle;
+	float2 ChannelShuffleOffset;
 	float2 TC_OffsetHack;
 	float2 STScale;
 	float4x4 DitherMatrix;
@@ -757,17 +758,17 @@ float4 ps_color(PS_INPUT input)
 #endif
 
 #if PS_CHANNEL_FETCH == 1
-	float4 T = fetch_red(int2(input.p.xy));
+	float4 T = fetch_red(int2(input.p.xy + ChannelShuffleOffset));
 #elif PS_CHANNEL_FETCH == 2
-	float4 T = fetch_green(int2(input.p.xy));
+	float4 T = fetch_green(int2(input.p.xy + ChannelShuffleOffset));
 #elif PS_CHANNEL_FETCH == 3
-	float4 T = fetch_blue(int2(input.p.xy));
+	float4 T = fetch_blue(int2(input.p.xy + ChannelShuffleOffset));
 #elif PS_CHANNEL_FETCH == 4
-	float4 T = fetch_alpha(int2(input.p.xy));
+	float4 T = fetch_alpha(int2(input.p.xy + ChannelShuffleOffset));
 #elif PS_CHANNEL_FETCH == 5
-	float4 T = fetch_rgb(int2(input.p.xy));
+	float4 T = fetch_rgb(int2(input.p.xy + ChannelShuffleOffset));
 #elif PS_CHANNEL_FETCH == 6
-	float4 T = fetch_gXbY(int2(input.p.xy));
+	float4 T = fetch_gXbY(int2(input.p.xy + ChannelShuffleOffset));
 #elif PS_DEPTH_FMT > 0
 	float4 T = sample_depth(st_int, input.p.xy);
 #else

bin/resources/shaders/opengl/tfx_fs.glsl

@@ -49,6 +49,7 @@ layout(std140, binding = 0) uniform cb21
 	vec4 STRange;
 
 	ivec4 ChannelShuffle;
+	vec2 ChannelShuffleOffset;
 
 	vec2 TC_OffsetHack;
 	vec2 STScale;
@@ -315,7 +316,7 @@ int fetch_raw_depth()
 #if PS_TEX_IS_FB == 1
 	return int(sample_from_rt().r * multiplier);
 #else
-	return int(texelFetch(TextureSampler, ivec2(gl_FragCoord.xy), 0).r * multiplier);
+	return int(texelFetch(TextureSampler, ivec2(gl_FragCoord.xy + ChannelShuffleOffset), 0).r * multiplier);
 #endif
 }
 
@@ -324,7 +325,7 @@ vec4 fetch_raw_color()
 #if PS_TEX_IS_FB == 1
 	return sample_from_rt();
 #else
-	return texelFetch(TextureSampler, ivec2(gl_FragCoord.xy), 0);
+	return texelFetch(TextureSampler, ivec2(gl_FragCoord.xy + ChannelShuffleOffset), 0);
 #endif
 }
 

bin/resources/shaders/vulkan/tfx.glsl

@@ -316,6 +316,7 @@ layout(std140, set = 0, binding = 1) uniform cb1
 	vec4 LODParams;
 	vec4 STRange;
 	ivec4 ChannelShuffle;
+	vec2 ChannelShuffleOffset;
 	vec2 TC_OffsetHack;
 	vec2 STScale;
 	mat4 DitherMatrix;
@@ -926,17 +927,17 @@ vec4 ps_color()
 #if !NEEDS_TEX
 	vec4 T = vec4(0.0f);
 #elif PS_CHANNEL_FETCH == 1
-	vec4 T = fetch_red(ivec2(gl_FragCoord.xy));
+	vec4 T = fetch_red(ivec2(gl_FragCoord.xy + ChannelShuffleOffset));
 #elif PS_CHANNEL_FETCH == 2
-	vec4 T = fetch_green(ivec2(gl_FragCoord.xy));
+	vec4 T = fetch_green(ivec2(gl_FragCoord.xy + ChannelShuffleOffset));
 #elif PS_CHANNEL_FETCH == 3
-	vec4 T = fetch_blue(ivec2(gl_FragCoord.xy));
+	vec4 T = fetch_blue(ivec2(gl_FragCoord.xy + ChannelShuffleOffset));
 #elif PS_CHANNEL_FETCH == 4
-	vec4 T = fetch_alpha(ivec2(gl_FragCoord.xy));
+	vec4 T = fetch_alpha(ivec2(gl_FragCoord.xy + ChannelShuffleOffset));
 #elif PS_CHANNEL_FETCH == 5
-	vec4 T = fetch_rgb(ivec2(gl_FragCoord.xy));
+	vec4 T = fetch_rgb(ivec2(gl_FragCoord.xy + ChannelShuffleOffset));
 #elif PS_CHANNEL_FETCH == 6
-	vec4 T = fetch_gXbY(ivec2(gl_FragCoord.xy));
+	vec4 T = fetch_gXbY(ivec2(gl_FragCoord.xy + ChannelShuffleOffset));
 #elif PS_DEPTH_FMT > 0
 	vec4 T = sample_depth(st_int, ivec2(gl_FragCoord.xy));
 #else

pcsx2/GS/GSState.cpp

@@ -1202,7 +1202,7 @@ void GSState::ApplyTEX0(GIFRegTEX0& TEX0)
 		// Urban Chaos writes to the memory backing the CLUT in the middle of a shuffle, and
 		// it's unclear whether the CLUT would actually get reloaded in that case.
 		if (TEX0.CBP != m_mem.m_clut.GetCLUTCBP())
-			m_channel_shuffle_abort = true;
+			m_channel_shuffle_finish = true;
 	}
 
 	TEX0.CPSM &= 0xa; // 1010b

pcsx2/GS/GSState.h

@@ -258,7 +258,7 @@ public:
 	bool m_using_temp_z = false;
 	bool m_temp_z_full_copy = false;
 	bool m_in_target_draw = false;
-	bool m_channel_shuffle_abort = false;
+	bool m_channel_shuffle_finish = false;
 
 	u32 m_target_offset = 0;
 	u8 m_scanmask_used = 0;

pcsx2/GS/Renderers/Common/GSDevice.h

@@ -622,6 +622,7 @@ struct alignas(16) GSHWDrawConfig
 		GSVector4 LODParams;
 		GSVector4 STRange;
 		GSVector4i ChannelShuffle;
+		GSVector2 ChannelShuffleOffset;
 		GSVector2 TCOffsetHack;
 		GSVector2 STScale;
 

pcsx2/GS/Renderers/HW/GSRendererHW.cpp

@@ -2345,9 +2345,18 @@ void GSRendererHW::Draw()
 		// Fortunately, it seems to change the FBMSK along the way, so this check alone is sufficient.
 		// Tomb Raider: Underworld does similar, except with R, G, B in separate palettes, therefore
 		// we need to split on those too.
-		m_channel_shuffle = !m_channel_shuffle_abort && IsPossibleChannelShuffle() && m_last_channel_shuffle_fbmsk == m_context->FRAME.FBMSK &&
-		                    m_last_channel_shuffle_fbp <= m_context->FRAME.Block() && m_last_channel_shuffle_end_block > m_context->FRAME.Block() &&
-		                    m_last_channel_shuffle_tbp <= m_context->TEX0.TBP0;
+		const bool is_hle_skip = m_conf.ps.urban_chaos_hle || m_conf.ps.tales_of_abyss_hle;
+		const u32 max_skip = ((m_channel_shuffle_finish || !m_channel_shuffle_width) ? std::max(m_context->FRAME.FBW, 1U) : m_channel_shuffle_width) << 5;
+		const bool shuffle_detect = IsPossibleChannelShuffle() && m_last_channel_shuffle_fbmsk == m_context->FRAME.FBMSK &&
+		                            m_last_channel_shuffle_fbp <= m_context->FRAME.Block() && (m_last_channel_shuffle_fbp + max_skip) >= m_context->FRAME.Block() && 
+									m_last_channel_shuffle_end_block > m_context->FRAME.Block() && m_last_channel_shuffle_tbp <= m_context->TEX0.TBP0
+									&& (m_last_channel_shuffle_tbp + max_skip) >= m_context->TEX0.TBP0;
+
+		const bool shuffle_detect_loose = IsPossibleChannelShuffle() && m_last_channel_shuffle_fbmsk == m_context->FRAME.FBMSK &&
+		                            m_last_channel_shuffle_fbp <= m_context->FRAME.Block() &&
+		                            m_last_channel_shuffle_end_block > m_context->FRAME.Block() && m_last_channel_shuffle_tbp <= m_context->TEX0.TBP0;
+
+		m_channel_shuffle = !m_channel_shuffle_finish && ((!is_hle_skip && shuffle_detect) || (is_hle_skip && shuffle_detect_loose));
 
 		if (m_channel_shuffle)
 		{
@@ -2415,12 +2424,21 @@ void GSRendererHW::Draw()
 				CleanupDraw(false);
 			}
 		}
+
+		if (!shuffle_detect)
+		{
+			m_last_channel_shuffle_fbp = 0xffff;
+			m_last_channel_shuffle_tbp = 0xffff;
+			m_last_channel_shuffle_end_block = 0xffff;
+		}
 #ifdef ENABLE_OGL_DEBUG
 		if (num_skipped_channel_shuffle_draws > 0)
 			GL_CACHE("HW: Skipped %d channel shuffle draws ending at %d", num_skipped_channel_shuffle_draws, s_n);
 #endif
 		num_skipped_channel_shuffle_draws = 0;
-
+	}
+	else
+	{
 		m_last_channel_shuffle_fbp = 0xffff;
 		m_last_channel_shuffle_tbp = 0xffff;
 		m_last_channel_shuffle_end_block = 0xffff;
@@ -2429,7 +2447,7 @@ void GSRendererHW::Draw()
 	m_last_rt = nullptr;
 	m_channel_shuffle_width = 0;
 	m_full_screen_shuffle = false;
-	m_channel_shuffle_abort = false;
+	m_channel_shuffle_finish = false;
 	m_channel_shuffle_src_valid = GSVector4i::zero();
 
 	GL_PUSH("HW: Draw %d (Context %u)", s_n, PRIM->CTXT);
@@ -3905,7 +3923,6 @@ void GSRendererHW::Draw()
 
 		if (m_channel_shuffle)
 		{
-			m_last_channel_shuffle_fbp = rt->m_TEX0.TBP0;
 			m_last_channel_shuffle_tbp = src->m_TEX0.TBP0;
 
 			// If it's a new target, we don't know where the end is as it's starting on a shuffle, so just do every shuffle following.
@@ -4042,7 +4059,6 @@ void GSRendererHW::Draw()
 			m_last_channel_shuffle_fbmsk = m_context->FRAME.FBMSK;
 			if (rt)
 			{
-				m_last_channel_shuffle_fbp = rt->m_TEX0.TBP0;
 				m_last_channel_shuffle_tbp = src->m_TEX0.TBP0;
 				// Urban Chaos goes from Z16 to C32, so let's just use the rt's original end block.
 				if (!src->m_from_target || GSLocalMemory::m_psm[src->m_from_target_TEX0.PSM].bpp != GSLocalMemory::m_psm[rt->m_TEX0.PSM].bpp)
@@ -4741,6 +4757,10 @@ void GSRendererHW::Draw()
 		// Limit to 2x the vertical height of the resolution (for double buffering)
 		rt->UpdateValidity(real_rect, !frame_masked && (can_update_size || (real_rect.w <= (resolution.y * 2) && !m_texture_shuffle)));
 
+		if (m_channel_shuffle)
+		{
+			m_last_channel_shuffle_fbp = rt->m_TEX0.TBP0;
+		}
 	}
 
 	if (ds)
@@ -5547,7 +5567,6 @@ __ri u32 GSRendererHW::EmulateChannelShuffle(GSTextureCache::Target* src, bool t
 		s[1].V = 16384;
 
 		m_r = GSVector4i(0, 0, 1024, 1024);
-
 		// We need to count the pages that get shuffled to, some games (like Hitman Blood Money dialogue blur effects) only do half the screen.
 		if (!m_full_screen_shuffle && !m_conf.ps.urban_chaos_hle && !m_conf.ps.tales_of_abyss_hle && src)
 		{
@@ -5557,21 +5576,14 @@ __ri u32 GSRendererHW::EmulateChannelShuffle(GSTextureCache::Target* src, bool t
 
 			m_channel_shuffle_width = src->m_TEX0.TBW;
 		}
+
+		m_channel_shuffle_finish = false;
 	}
 	else
 	{
 		const u32 frame_page_offset = std::max(static_cast<int>(((m_r.x / frame_psm.pgs.x) + (m_r.y / frame_psm.pgs.y) * rt->m_TEX0.TBW)), 0);
 		m_r = GSVector4i(m_r.x & ~(frame_psm.pgs.x - 1), m_r.y & ~(frame_psm.pgs.y - 1), (m_r.z + (frame_psm.pgs.x - 1)) & ~(frame_psm.pgs.x - 1), (m_r.w + (frame_psm.pgs.y - 1)) & ~(frame_psm.pgs.y - 1));
 
-		// Hitman suffers from this, not sure on the exact scenario at the moment, but we need the barrier.
-		if (NeedsBlending() && 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;
-		}
-
 		// This is for offsetting the texture, however if the texture has a region clamp, we don't want to move it.
 		// A good two test games for this is Ghost in the Shell (no region clamp) and Tekken 5 (offset clamp on shadows)
 		if (rt && rt->m_TEX0.TBP0 == m_cached_ctx.FRAME.Block())
@@ -5595,7 +5607,6 @@ __ri u32 GSRendererHW::EmulateChannelShuffle(GSTextureCache::Target* src, bool t
 		s[1].U = m_r.z << 4;
 		s[0].V = m_r.y << 4;
 		s[1].V = m_r.w << 4;
-		m_last_channel_shuffle_fbmsk = 0xFFFFFFFF;
 
 		// If we're doing per page copying, then set the valid 1 frame ahead if we're continuing, as this will save the target lookup making a new target for the new row.
 		const u32 frame_offset = m_cached_ctx.FRAME.Block() + (IsPageCopy() ? 0x20 : 0);
@@ -5614,13 +5625,15 @@ __ri u32 GSRendererHW::EmulateChannelShuffle(GSTextureCache::Target* src, bool t
 
 		new_valid.w = std::max(new_valid.w, offset_height);
 		rt->UpdateValidity(new_valid, true);
+
+		m_channel_shuffle_finish = true;
 	}
 
 	m_vertex.head = m_vertex.tail = m_vertex.next = 2;
 	m_index.tail = 2;
 
 	m_primitive_covers_without_gaps = NoGapsType::FullCover;
-	m_channel_shuffle_abort = false;
+	m_conf.cb_ps.ChannelShuffleOffset = GSVector2(0, 0);
 
 	return true;
 }
@@ -6426,7 +6439,10 @@ __ri void GSRendererHW::EmulateTextureSampler(const GSTextureCache::Target* rt,
 {
 	// don't overwrite the texture when using channel shuffle, but keep the palette
 	if (!m_channel_shuffle)
+	{
+		m_conf.cb_ps.ChannelShuffleOffset = GSVector2(0, 0);
 		m_conf.tex = tex->m_texture;
+	}
 	m_conf.pal = tex->m_palette;
 
 	// Hazard handling (i.e. reading from the current RT/DS).
@@ -6818,7 +6834,7 @@ __ri void GSRendererHW::HandleTextureHazards(const GSTextureCache::Target* rt, c
 	const GSTextureCache::Target* src_target = nullptr;
 	if (!m_downscale_source || !tex->m_from_target)
 	{
-		if (rt && m_conf.tex == m_conf.rt && !(m_channel_shuffle && tex && (tex_diff != frame_diff || target_region)))
+		if (rt && m_conf.tex == m_conf.rt && !(m_channel_shuffle && tex && tex_diff != frame_diff))
 		{
 			// Can we read the framebuffer directly? (i.e. sample location matches up).
 			if (CanUseTexIsFB(rt, tex, tmm))
@@ -6901,37 +6917,63 @@ __ri void GSRendererHW::HandleTextureHazards(const GSTextureCache::Target* rt, c
 		GSVector4i::storel(&copy_dst_offset, copy_range);
 		if (m_channel_shuffle && (tex_diff || frame_diff))
 		{
+			const int page_offset = (m_cached_ctx.TEX0.TBP0 - src_target->m_TEX0.TBP0) >> 5;
+			const int horizontal_offset = ((page_offset % src_target->m_TEX0.TBW) * GSLocalMemory::m_psm[src_target->m_TEX0.PSM].pgs.x);
+			const int vertical_offset = ((page_offset / src_target->m_TEX0.TBW) * GSLocalMemory::m_psm[src_target->m_TEX0.PSM].pgs.y);
 
-			const u32 page_offset = (m_cached_ctx.TEX0.TBP0 - src_target->m_TEX0.TBP0) >> 5;
-			const u32 horizontal_offset = (page_offset % src_target->m_TEX0.TBW) * GSLocalMemory::m_psm[src_target->m_TEX0.PSM].pgs.x;
-			const u32 vertical_offset = (page_offset / src_target->m_TEX0.TBW) * GSLocalMemory::m_psm[src_target->m_TEX0.PSM].pgs.y;
-
-			copy_range.x += horizontal_offset;
-			copy_range.y += vertical_offset;
-			copy_range.z += horizontal_offset;
-			copy_range.w += vertical_offset;
-
-			if (!m_channel_shuffle)
+			if (g_gs_device->Features().texture_barrier || g_gs_device->Features().multidraw_fb_copy)
 			{
-				copy_size.y -= vertical_offset;
-				copy_size.x -= horizontal_offset;
-			}
-			target_region = false;
-			source_region.bits = 0;
-			//copied_rt = tex->m_from_target != nullptr;
-			if (m_in_target_draw && (page_offset || frame_diff))
-			{
-				copy_range.z = copy_range.x + m_r.width();
-				copy_range.w = copy_range.y + m_r.height();
+				const u32 max_skip = ((m_channel_shuffle_finish || !m_channel_shuffle_width) ? 1 : m_channel_shuffle_width) << 5;
+				const bool new_shuffle = !(m_last_channel_shuffle_fbmsk == m_context->FRAME.FBMSK &&
+										   m_last_channel_shuffle_fbp <= m_context->FRAME.Block() && (m_last_channel_shuffle_fbp + max_skip) >= m_context->FRAME.Block() &&
+										   m_last_channel_shuffle_end_block > m_context->FRAME.Block() && m_last_channel_shuffle_tbp <= m_context->TEX0.TBP0 && (m_last_channel_shuffle_tbp + max_skip) >= m_context->TEX0.TBP0);
 
-				if (tex_diff != frame_diff)
+				if (rt == tex->m_from_target && new_shuffle)
 				{
-					GSVector4i::storel(&copy_dst_offset, m_r);
+					if (m_prim_overlap == PRIM_OVERLAP_NO)
+						m_conf.require_one_barrier = true;
+					else
+						m_conf.require_full_barrier = true;
 				}
-			}
 
-			copy_range.z = std::min(copy_range.z, src_target->m_unscaled_size.x);
-			copy_range.w = std::min(copy_range.w, src_target->m_unscaled_size.y);
+				m_conf.cb_ps.ChannelShuffleOffset = GSVector2((horizontal_offset - m_r.x) * tex->GetScale(), (vertical_offset - m_r.y) * tex->GetScale());
+				m_conf.ps.channel_fb = 1;
+				target_region = false;
+				source_region.bits = 0;
+
+				unscaled_size = src_target->GetUnscaledSize();
+				scale = src_target->GetScale();
+				return;
+			}
+			else
+			{
+				copy_range.x += horizontal_offset;
+				copy_range.y += vertical_offset;
+				copy_range.z += horizontal_offset;
+				copy_range.w += vertical_offset;
+
+				if (!m_channel_shuffle)
+				{
+					copy_size.y -= vertical_offset;
+					copy_size.x -= horizontal_offset;
+				}
+				target_region = false;
+				source_region.bits = 0;
+				//copied_rt = tex->m_from_target != nullptr;
+				if (m_in_target_draw && (page_offset || frame_diff))
+				{
+					copy_range.z = copy_range.x + m_r.width();
+					copy_range.w = copy_range.y + m_r.height();
+
+					if (tex_diff != frame_diff)
+					{
+						GSVector4i::storel(&copy_dst_offset, m_r);
+					}
+				}
+
+				copy_range.z = std::min(copy_range.z, src_target->m_unscaled_size.x);
+				copy_range.w = std::min(copy_range.w, src_target->m_unscaled_size.y);
+			}
 		}
 	}
 	else

pcsx2/GS/Renderers/HW/GSTextureCache.cpp

@@ -3776,23 +3776,54 @@ bool GSTextureCache::PreloadTarget(GIFRegTEX0 TEX0, const GSVector2i& size, cons
 						if (GSConfig.UserHacks_TextureInsideRt >= GSTextureInRtMode::InsideTargets && t->Inside(dst->m_TEX0.TBP0, dst->m_TEX0.TBW, dst->m_TEX0.PSM, dst->m_valid) &&
 							GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp == GSLocalMemory::m_psm[dst->m_TEX0.PSM].bpp)
 						{
-							dst->m_TEX0.TBP0 = t->m_TEX0.TBP0;
-							dst->m_valid = t->m_valid;
-							dst->m_drawn_since_read = t->m_drawn_since_read;
-							dst->m_end_block = t->m_end_block;
-							dst->m_valid_rgb = true;
-							t->m_valid_rgb = false;
-							t->m_was_dst_matched = true;
+							// if this part is dirty, then let's break it in two.
+							GSVector4i dirty_rect = t->m_dirty.GetTotalRect(t->m_TEX0, t->m_unscaled_size);
+							if (GSLocalMemory::GetStartBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, dirty_rect) >= dst->m_TEX0.TBP0)
+							{
+								t->m_valid.w = dirty_rect.y;
 
-							dst->ResizeTexture(t->m_unscaled_size.x, t->m_unscaled_size.y);
+								if (t->m_valid.rempty())
+								{
+									if (src && src->m_target && src->m_from_target == t)
+									{
+										src->m_from_target = nullptr;
+										src->m_texture = t->m_texture;
+										src->m_target_direct = false;
+										src->m_shared_texture = false;
 
-							const ShaderConvert shader = (GSLocalMemory::m_psm[dst->m_TEX0.PSM].trbpp == 16) ? ShaderConvert::RGB5A1_TO_FLOAT16 :
-							                             (GSLocalMemory::m_psm[dst->m_TEX0.PSM].trbpp == 32) ? ShaderConvert::RGBA8_TO_FLOAT32 :
-							                                                                                   ShaderConvert::RGBA8_TO_FLOAT24;
+										t->m_texture = nullptr;
+										i = list.erase(j);
+										delete t;
+									}
+									else
+									{
+										InvalidateSourcesFromTarget(t);
+										i = list.erase(j);
+										delete t;
+									}
+								}
+								else
+									t->UpdateValidity(t->m_valid, true);
+							}
+							else
+							{
+								dst->m_TEX0.TBP0 = t->m_TEX0.TBP0;
+								dst->m_valid = t->m_valid;
+								dst->m_drawn_since_read = t->m_drawn_since_read;
+								dst->m_end_block = t->m_end_block;
+								dst->m_valid_rgb = true;
+								t->m_valid_rgb = false;
+								t->m_was_dst_matched = true;
 
-							g_gs_device->StretchRect(t->m_texture, GSVector4(0, 0, 1, 1),
-								dst->m_texture, GSVector4(t->GetUnscaledRect()) * GSVector4(dst->GetScale()), shader, false);
+								dst->ResizeTexture(t->m_unscaled_size.x, t->m_unscaled_size.y);
 
+								const ShaderConvert shader = (GSLocalMemory::m_psm[dst->m_TEX0.PSM].trbpp == 16) ? ShaderConvert::RGB5A1_TO_FLOAT16 :
+								                             (GSLocalMemory::m_psm[dst->m_TEX0.PSM].trbpp == 32) ? ShaderConvert::RGBA8_TO_FLOAT32 :
+								                                                                                   ShaderConvert::RGBA8_TO_FLOAT24;
+
+								g_gs_device->StretchRect(t->m_texture, GSVector4(0, 0, 1, 1),
+									dst->m_texture, GSVector4(t->GetUnscaledRect()) * GSVector4(dst->GetScale()), shader, false);
+							}
 							break;
 						}
 						else
@@ -3819,7 +3850,18 @@ bool GSTextureCache::PreloadTarget(GIFRegTEX0 TEX0, const GSVector2i& size, cons
 
 						if (height_adjust < t->m_unscaled_size.y)
 						{
-							t->m_TEX0.TBP0 = GSLocalMemory::GetStartBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, GSVector4i(0, height_adjust, t->m_valid.z, t->m_valid.w));
+							u32 new_base_tbp = GSLocalMemory::GetStartBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, GSVector4i(0, height_adjust, t->m_valid.z, t->m_valid.w));
+							if (t->m_dirty.size() > 0)
+							{
+								u32 dirty_end = GSLocalMemory::GetEndBlockAddress(t->m_TEX0.TBP0, t->m_TEX0.TBW, t->m_TEX0.PSM, t->m_dirty.GetTotalRect(t->m_TEX0, t->m_unscaled_size));
+
+								if (new_base_tbp >= dirty_end)
+									t->m_dirty.clear();
+								else
+									t->Update(true);
+							}
+
+							t->m_TEX0.TBP0 = new_base_tbp;
 							t->m_valid.w -= height_adjust;
 							t->ResizeValidity(t->m_valid);
 
@@ -4330,7 +4372,7 @@ void GSTextureCache::InvalidateContainedTargets(u32 start_bp, u32 end_bp, u32 wr
 				const u32 end_width = write_bw * 64;
 				const u32 end_height = ((end_page_offset / std::max(write_bw, 1U)) * GSLocalMemory::m_psm[write_psm].pgs.y) + GSLocalMemory::m_psm[write_psm].pgs.y;
 				const GSVector4i r = GSVector4i(0, 0, end_width, end_height);
-				const GSVector4i invalidate_r = TranslateAlignedRectByPage(t, start_bp, write_psm, write_bw, r, false).rintersect(t->m_valid); // it is invalidation but we need a real rect.
+				const GSVector4i invalidate_r = TranslateAlignedRectByPage(t, start_bp, write_psm, write_bw, r, true).rintersect(t->m_valid); // it is invalidation but we need a real rect.
 
 				if (offset == 0 || dirty_rect.rempty() || !dirty_rect.rintersect(invalidate_r).rempty())
 				{
@@ -4339,7 +4381,10 @@ void GSTextureCache::InvalidateContainedTargets(u32 start_bp, u32 end_bp, u32 wr
 
 						RGBAMask mask;
 						mask._u32 = GSUtil::GetChannelMask(write_psm);
-						AddDirtyRectTarget(t, invalidate_r, t->m_TEX0.PSM, t->m_TEX0.TBW, mask, false);
+						if (write_bw != t->m_TEX0.TBW)
+							DirtyRectByPage(start_bp, write_psm, write_bw, t, r);
+						else
+							AddDirtyRectTarget(t, invalidate_r, t->m_TEX0.PSM, t->m_TEX0.TBW, mask, false);
 					}
 
 					++i;

pcsx2/GS/Renderers/Metal/GSDeviceMTL.mm

@@ -2081,6 +2081,7 @@ static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, HalfTexel)        == of
 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, ChannelShuffleOffset) == offsetof(GSMTLMainPSUniform, channel_shuffle_offset));
 static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, TCOffsetHack)     == offsetof(GSMTLMainPSUniform, tc_offset));
 static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, STScale)          == offsetof(GSMTLMainPSUniform, st_scale));
 static_assert(offsetof(GSHWDrawConfig::PSConstantBuffer, DitherMatrix)     == offsetof(GSMTLMainPSUniform, dither_matrix));

pcsx2/GS/Renderers/Metal/GSMTLSharedHeader.h

@@ -129,6 +129,7 @@ struct GSMTLMainPSUniform
 		unsigned int green_mask;
 		unsigned int green_shift;
 	} channel_shuffle;
+	vector_float2 channel_shuffle_offset;
 	vector_float2 tc_offset;
 	vector_float2 st_scale;
 	matrix_float4x4 dither_matrix;

pcsx2/GS/Renderers/Metal/tfx.metal

@@ -509,7 +509,7 @@ struct PSMain
 
 	uint fetch_raw_depth()
 	{
-		return tex_depth.read(ushort2(in.p.xy)) * 0x1p32f;
+		return tex_depth.read(ushort2(in.p.xy + cb.channel_shuffle_offset)) * 0x1p32f;
 	}
 
 	float4 fetch_raw_color()
@@ -517,7 +517,7 @@ struct PSMain
 		if (PS_TEX_IS_FB)
 			return current_color;
 		else
-			return tex.read(ushort2(in.p.xy));
+			return tex.read(ushort2(in.p.xy + cb.channel_shuffle_offset));
 	}
 
 	float4 fetch_c(ushort2 uv)

pcsx2/ShaderCacheVersion.h

@@ -3,4 +3,4 @@
 
 /// 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 = 79;
+static constexpr u32 SHADER_CACHE_VERSION = 80;