GS/HW: Make static variables members where possible.

Fixes a bug with batch runner mode with split channel shuffle.

Make variables constexpr where possible.

Author: TJnotJT

pcsx2/GS/Renderers/Common/GSDevice.cpp

@@ -339,6 +339,10 @@ void GSDevice::ResetRenderState()
 {
 	ClearCurrent();
 	PurgePool();
+
+#ifdef PCSX2_DEVBUILD
+	s_texture_counts.fill(0);
+#endif
 }
 
 bool GSDevice::AcquireWindow(bool recreate_window)
@@ -790,7 +794,6 @@ void GSDevice::Merge(GSTexture* sTex[3], GSVector4* sRect, GSVector4* dRect, con
 
 void GSDevice::Interlace(const GSVector2i& ds, int field, int mode, float yoffset)
 {
-	static int bufIdx = 0;
 	float offset = yoffset * static_cast<float>(field);
 	offset = GSConfig.DisableInterlaceOffset ? 0.0f : offset;
 
@@ -840,14 +843,14 @@ void GSDevice::Interlace(const GSVector2i& ds, int field, int mode, float yoffse
 			m_current = m_blend;
 			break;
 		case 3: // FastMAD Motion Adaptive Deinterlacing
-			bufIdx++;
-			bufIdx &= ~1;
-			bufIdx |= field;
-			bufIdx &= 3;
+			m_interlace_bufIdx++;
+			m_interlace_bufIdx &= ~1;
+			m_interlace_bufIdx |= field;
+			m_interlace_bufIdx &= 3;
 			ResizeRenderTarget(&m_mad, ds.x, ds.y * 2.0f, true, false);
-			do_interlace(m_merge, m_mad, ShaderInterlace::MAD_BUFFER, false, offset, bufIdx);
+			do_interlace(m_merge, m_mad, ShaderInterlace::MAD_BUFFER, false, offset, m_interlace_bufIdx);
 			ResizeRenderTarget(&m_weavebob, ds.x, ds.y, true, false);
-			do_interlace(m_mad, m_weavebob, ShaderInterlace::MAD_RECONSTRUCT, false, 0, bufIdx);
+			do_interlace(m_mad, m_weavebob, ShaderInterlace::MAD_RECONSTRUCT, false, 0, m_interlace_bufIdx);
 			m_current = m_weavebob;
 			break;
 		default:

pcsx2/GS/Renderers/Common/GSDevice.h

@@ -903,6 +903,8 @@ class GSDevice : public GSAlignedClass<32>
 	GSTexture* m_cas = nullptr;
 	GSTexture* m_colclip_rt = nullptr; ///< Temp hw colclip texture
 
+	int m_interlace_bufIdx = 0;
+
 	bool AcquireWindow(bool recreate_window);
 
 	virtual GSTexture* CreateSurface(GSTexture::Type type, int width, int height, int levels, GSTexture::Format format) = 0;

pcsx2/GS/Renderers/DX11/GSDevice11.cpp

@@ -1929,12 +1929,12 @@ void GSDevice11::SetupOM(OMDepthStencilSelector dssel, OMBlendSelector bsel, u8
 
 		if (dssel.ztst != ZTST_ALWAYS || dssel.zwe)
 		{
-			static const D3D11_COMPARISON_FUNC ztst[] =
+			static constexpr D3D11_COMPARISON_FUNC ztst[] =
 			{
-					D3D11_COMPARISON_NEVER,
-					D3D11_COMPARISON_ALWAYS,
-					D3D11_COMPARISON_GREATER_EQUAL,
-					D3D11_COMPARISON_GREATER
+				D3D11_COMPARISON_NEVER,
+				D3D11_COMPARISON_ALWAYS,
+				D3D11_COMPARISON_GREATER_EQUAL,
+				D3D11_COMPARISON_GREATER
 			};
 
 			dsd.DepthEnable = true;
@@ -2027,7 +2027,7 @@ bool GSDevice11::CreateCASShaders()
 
 bool GSDevice11::DoCAS(GSTexture* sTex, GSTexture* dTex, bool sharpen_only, const std::array<u32, NUM_CAS_CONSTANTS>& constants)
 {
-	static const int threadGroupWorkRegionDim = 16;
+	static constexpr int threadGroupWorkRegionDim = 16;
 	const int dispatchX = (dTex->GetWidth() + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim;
 	const int dispatchY = (dTex->GetHeight() + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim;
 

pcsx2/GS/Renderers/DX12/GSDevice12.cpp

@@ -3030,7 +3030,7 @@ GSDevice12::ComPtr<ID3D12PipelineState> GSDevice12::CreateTFXPipeline(const Pipe
 	// DepthStencil
 	if (p.ds)
 	{
-		static const D3D12_COMPARISON_FUNC ztst[] = {D3D12_COMPARISON_FUNC_NEVER, D3D12_COMPARISON_FUNC_ALWAYS,
+		static constexpr D3D12_COMPARISON_FUNC ztst[] = {D3D12_COMPARISON_FUNC_NEVER, D3D12_COMPARISON_FUNC_ALWAYS,
 			D3D12_COMPARISON_FUNC_GREATER_EQUAL, D3D12_COMPARISON_FUNC_GREATER};
 		gpb.SetDepthState((p.dss.ztst != ZTST_ALWAYS || p.dss.zwe), p.dss.zwe, ztst[p.dss.ztst]);
 		if (p.dss.date)

pcsx2/GS/Renderers/HW/GSRendererHW.cpp

@@ -2322,8 +2322,6 @@ void GSRendererHW::RoundSpriteOffset()
 
 void GSRendererHW::Draw()
 {
-	static u32 num_skipped_channel_shuffle_draws = 0;
-
 	// We mess with this state as an optimization, so take a copy and use that instead.
 	const GSDrawingContext* context = m_context;
 	m_cached_ctx.TEX0 = context->TEX0;
@@ -9464,17 +9462,13 @@ GSHWDrawConfig& GSRendererHW::BeginHLEHardwareDraw(
 	std::memset(&config.cb_vs, 0, sizeof(config.cb_vs));
 	std::memset(&config.cb_ps, 0, sizeof(config.cb_ps));
 
-	// Reused between draws, since the draw config is shared, you can't have multiple draws in flight anyway.
-	static GSVertex vertices[4];
-	static constexpr u16 indices[6] = {0, 1, 2, 2, 1, 3};
-
 #define V(i, x, y, u, v) \
 	do \
 	{ \
-		vertices[i].XYZ.X = x; \
-		vertices[i].XYZ.Y = y; \
-		vertices[i].U = u; \
-		vertices[i].V = v; \
+		m_hle_vertices[i].XYZ.X = x; \
+		m_hle_vertices[i].XYZ.Y = y; \
+		m_hle_vertices[i].U = u; \
+		m_hle_vertices[i].V = v; \
 	} while (0)
 
 	const GSVector4i fp_rect = unscaled_rect.sll32<4>();
@@ -9490,10 +9484,10 @@ GSHWDrawConfig& GSRendererHW::BeginHLEHardwareDraw(
 	config.ds = ds;
 	config.tex = tex;
 	config.pal = nullptr;
-	config.indices = indices;
-	config.verts = vertices;
-	config.nverts = static_cast<u32>(std::size(vertices));
-	config.nindices = static_cast<u32>(std::size(indices));
+	config.indices = m_hle_indices;
+	config.verts = m_hle_vertices;
+	config.nverts = static_cast<u32>(std::size(m_hle_vertices));
+	config.nindices = static_cast<u32>(std::size(m_hle_indices));
 	config.indices_per_prim = 3;
 	config.drawlist = nullptr;
 	config.scissor = rt_or_ds->GetRect();

pcsx2/GS/Renderers/HW/GSRendererHW.h

@@ -183,6 +183,7 @@ class GSRendererHW : public GSRenderer
 	u32 m_channel_shuffle_width = 0;
 	GSVector4i m_channel_shuffle_src_valid = GSVector4i::zero();
 	bool m_full_screen_shuffle = false;
+	u32 num_skipped_channel_shuffle_draws = 0;
 
 	GSTextureCache::Target* m_last_rt;
 
@@ -205,6 +206,10 @@ class GSRendererHW : public GSRenderer
 	std::unique_ptr<GSTextureCacheSW::Texture> m_sw_texture[7 + 1];
 	std::unique_ptr<GSVirtualAlignedClass<32>> m_sw_rasterizer;
 
+	// HLE state
+	GSVertex m_hle_vertices[4] = {};
+	static constexpr u16 m_hle_indices[6] = {0, 1, 2, 2, 1, 3};
+
 public:
 	GSRendererHW();
 	virtual ~GSRendererHW() override;

pcsx2/GS/Renderers/HW/GSTextureCache.cpp

@@ -30,9 +30,6 @@ std::unique_ptr<GSTextureCache> g_texture_cache;
 
 static u8* s_unswizzle_buffer;
 
-/// List of candidates for purging when the hash cache gets too large.
-static std::vector<std::pair<GSTextureCache::HashCacheMap::iterator, s32>> s_hash_cache_purge_list;
-
 #ifdef PCSX2_DEVBUILD
 // We can only set one texture name per command buffer, which would break our fancy texture cache RT/DS/texture naming.
 // So, when debug device is enabled, don't reuse any textures that are drawable.
@@ -62,7 +59,6 @@ GSTextureCache::~GSTextureCache()
 {
 	RemoveAll(true, true, true);
 
-	s_hash_cache_purge_list = {};
 	_aligned_free(s_unswizzle_buffer);
 }
 
@@ -109,6 +105,8 @@ void GSTextureCache::RemoveAll(bool sources, bool targets, bool hash_cache)
 		m_hash_cache.clear();
 		m_hash_cache_memory_usage = 0;
 		m_hash_cache_replacement_memory_usage = 0;
+
+		m_hash_cache_purge_list.clear();
 	}
 }
 
@@ -6811,7 +6809,7 @@ void GSTextureCache::AgeHashCache()
 
 	bool might_need_cache_purge = (m_hash_cache.size() > MAX_HASH_CACHE_SIZE);
 	if (might_need_cache_purge)
-		s_hash_cache_purge_list.clear();
+		m_hash_cache_purge_list.clear();
 
 	for (auto it = m_hash_cache.begin(); it != m_hash_cache.end();)
 	{
@@ -6833,7 +6831,7 @@ void GSTextureCache::AgeHashCache()
 		{
 			might_need_cache_purge = (m_hash_cache.size() > MAX_HASH_CACHE_SIZE);
 			if (might_need_cache_purge)
-				s_hash_cache_purge_list.emplace_back(it, static_cast<s32>(e.age));
+				m_hash_cache_purge_list.emplace_back(it, static_cast<s32>(e.age));
 		}
 
 		++it;
@@ -6842,13 +6840,13 @@ void GSTextureCache::AgeHashCache()
 	// Pushing to a list, sorting, and removing ends up faster than re-iterating the map.
 	if (might_need_cache_purge)
 	{
-		std::sort(s_hash_cache_purge_list.begin(), s_hash_cache_purge_list.end(),
+		std::sort(m_hash_cache_purge_list.begin(), m_hash_cache_purge_list.end(),
 			[](const auto& lhs, const auto& rhs) { return lhs.second > rhs.second; });
 
 		const u32 entries_to_purge = std::min(static_cast<u32>(m_hash_cache.size() - MAX_HASH_CACHE_SIZE),
-			static_cast<u32>(s_hash_cache_purge_list.size()));
+			static_cast<u32>(m_hash_cache_purge_list.size()));
 		for (u32 i = 0; i < entries_to_purge; i++)
-			RemoveFromHashCache(s_hash_cache_purge_list[i].first);
+			RemoveFromHashCache(m_hash_cache_purge_list[i].first);
 	}
 }
 

pcsx2/GS/Renderers/HW/GSTextureCache.h

@@ -424,6 +424,8 @@ class GSTextureCache
 	HashCacheMap m_hash_cache;
 	u64 m_hash_cache_memory_usage = 0;
 	u64 m_hash_cache_replacement_memory_usage = 0;
+	/// List of candidates for purging when the hash cache gets too large.
+	std::vector<std::pair<GSTextureCache::HashCacheMap::iterator, s32>> m_hash_cache_purge_list = {};
 
 	FastList<Target*> m_dst[2];
 	FastList<TargetHeightElem> m_target_heights;

pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp

@@ -1268,7 +1268,7 @@ GSDepthStencilOGL* GSDeviceOGL::CreateDepthStencil(OMDepthStencilSelector dssel)
 
 	if (dssel.ztst != ZTST_ALWAYS || dssel.zwe)
 	{
-		static const GLenum ztst[] =
+		static constexpr GLenum ztst[] =
 		{
 			GL_NEVER,
 			GL_ALWAYS,
@@ -2101,7 +2101,7 @@ bool GSDeviceOGL::DoCAS(GSTexture* sTex, GSTexture* dTex, bool sharpen_only, con
 	PSSetShaderResource(0, sTex);
 	glBindImageTexture(0, static_cast<GSTextureOGL*>(dTex)->GetID(), 0, GL_FALSE, 0, GL_WRITE_ONLY, GL_RGBA8);
 
-	static const int threadGroupWorkRegionDim = 16;
+	static constexpr int threadGroupWorkRegionDim = 16;
 	const int dispatchX = (dTex->GetWidth() + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim;
 	const int dispatchY = (dTex->GetHeight() + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim;
 	glDispatchCompute(dispatchX, dispatchY, 1);

pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp

@@ -597,7 +597,7 @@ bool GSDeviceVK::CreateDevice(VkSurfaceKHR surface, bool enable_validation_layer
 	// Enable debug layer on debug builds
 	if (enable_validation_layer)
 	{
-		static const char* layer_names[] = {"VK_LAYER_LUNARG_standard_validation"};
+		static constexpr const char* layer_names[] = {"VK_LAYER_LUNARG_standard_validation"};
 		device_info.enabledLayerCount = 1;
 		device_info.ppEnabledLayerNames = layer_names;
 	}
@@ -4520,7 +4520,7 @@ bool GSDeviceVK::DoCAS(
 	dsub.PushUpdate(cmdbuf, VK_PIPELINE_BIND_POINT_COMPUTE, m_cas_pipeline_layout, 0, false);
 
 	// the actual meat and potatoes! only four commands.
-	static const int threadGroupWorkRegionDim = 16;
+	static constexpr int threadGroupWorkRegionDim = 16;
 	const int dispatchX = (dTex->GetWidth() + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim;
 	const int dispatchY = (dTex->GetHeight() + (threadGroupWorkRegionDim - 1)) / threadGroupWorkRegionDim;
 
@@ -4877,7 +4877,7 @@ VkPipeline GSDeviceVK::CreateTFXPipeline(const PipelineSelector& p)
 	}
 
 	// DepthStencil
-	static const VkCompareOp ztst[] = {
+	static constexpr VkCompareOp ztst[] = {
 		VK_COMPARE_OP_NEVER, VK_COMPARE_OP_ALWAYS, VK_COMPARE_OP_GREATER_OR_EQUAL, VK_COMPARE_OP_GREATER};
 	gpb.SetDepthState((p.dss.ztst != ZTST_ALWAYS || p.dss.zwe), p.dss.zwe, ztst[p.dss.ztst]);
 	if (p.dss.date)