GS: Move expanded dither storage to software renderer

Hardware doesn't use it.

pcsx2/GS/GSState.cpp

@@ -154,8 +154,6 @@ void GSState::Reset(bool hardware_reset)
 
 	UpdateVertexKick();
 
-	UpdateDIMX();
-
 	for (u32 i = 0; i < 2; i++)
 	{
 		m_env.CTXT[i].UpdateScissor();
@@ -1207,9 +1205,6 @@ void GSState::GIFRegHandlerDIMX(const GIFReg* RESTRICT r)
 
 	m_env.DIMX = r->DIMX;
 
-	if (update)
-		UpdateDIMX();
-
 	if (m_prev_env.DIMX != m_env.DIMX)
 		m_dirty_gs_regs |= (1 << DIRTY_REG_DIMX);
 	else
@@ -1454,18 +1449,12 @@ void GSState::Flush(GSFlushReason reason)
 			PRIM = &m_prev_env.PRIM;
 			UpdateContext();
 
-			if (m_dirty_gs_regs & (1 << DIRTY_REG_DIMX))
-				UpdateDIMX();
-
 			FlushPrim();
 
 			m_draw_env = &m_env;
 			PRIM = &m_env.PRIM;
 			UpdateContext();
 
-			if (m_dirty_gs_regs & (1 << DIRTY_REG_DIMX))
-				UpdateDIMX();
-
 			m_backed_up_ctx = -1;
 		}
 		else
@@ -2526,8 +2515,6 @@ int GSState::Defrost(const freezeData* fd)
 
 	UpdateVertexKick();
 
-	UpdateDIMX();
-
 	for (u32 i = 0; i < 2; i++)
 	{
 		m_env.CTXT[i].UpdateScissor();
@@ -2566,7 +2553,6 @@ void GSState::UpdateContext()
 	if (ctx_switch)
 		GL_REG("Context Switch %d", PRIM->CTXT);
 
-	// TODO: Don't mutate context (looking at you, HW)
 	m_context = const_cast<GSDrawingContext*>(&m_draw_env->CTXT[PRIM->CTXT]);
 
 	UpdateScissor();
@@ -3814,19 +3800,6 @@ bool GSState::IsCoverageAlpha()
 	return !PRIM->ABE && PRIM->AA1 && (m_vt.m_primclass == GS_LINE_CLASS || m_vt.m_primclass == GS_TRIANGLE_CLASS);
 }
 
-void GSState::UpdateDIMX()
-{
-	const GIFRegDIMX& DIMX = m_draw_env->DIMX;
-	dimx[1] = GSVector4i(DIMX.DM00, 0, DIMX.DM01, 0, DIMX.DM02, 0, DIMX.DM03, 0);
-	dimx[0] = dimx[1].xxzzlh();
-	dimx[3] = GSVector4i(DIMX.DM10, 0, DIMX.DM11, 0, DIMX.DM12, 0, DIMX.DM13, 0);
-	dimx[2] = dimx[3].xxzzlh();
-	dimx[5] = GSVector4i(DIMX.DM20, 0, DIMX.DM21, 0, DIMX.DM22, 0, DIMX.DM23, 0);
-	dimx[4] = dimx[5].xxzzlh();
-	dimx[7] = GSVector4i(DIMX.DM30, 0, DIMX.DM31, 0, DIMX.DM32, 0, DIMX.DM33, 0);
-	dimx[6] = dimx[7].xxzzlh();
-}
-
 GIFRegTEX0 GSState::GetTex0Layer(u32 lod)
 {
 	// Shortcut

pcsx2/GS/GSState.h

@@ -205,7 +205,6 @@ protected:
 	bool IsMipMapDraw();
 	bool IsMipMapActive();
 	bool IsCoverageAlpha();
-	void UpdateDIMX();
 
 public:
 	struct GSUploadQueue
@@ -224,7 +223,6 @@ public:
 	const GSDrawingEnvironment* m_draw_env = &m_env;
 	GSDrawingContext* m_context = nullptr;
 	GSVector4i temp_draw_rect = {};
-	GSVector4i dimx[8] = {};
 	u32 m_crc = 0;
 	CRC::Game m_game = {};
 	std::unique_ptr<GSDumpBase> m_dump;
@@ -870,6 +868,9 @@ public:
 	/// Returns the appropriate directory for draw dumping.
 	static std::string GetDrawDumpPath(const char* format, ...);
 
+	/// Expands dither matrix, suitable for software renderer.
+	static void ExpandDIMX(GSVector4i* dimx, const GIFRegDIMX DIMX);
+
 	void ResetHandlers();
 	void ResetPCRTC();
 
@@ -920,3 +921,16 @@ public:
 	PRIM_OVERLAP PrimitiveOverlap();
 	GIFRegTEX0 GetTex0Layer(u32 lod);
 };
+
+// We put this in the header because of Multi-ISA.
+inline void GSState::ExpandDIMX(GSVector4i* dimx, const GIFRegDIMX DIMX)
+{
+	dimx[1] = GSVector4i(DIMX.DM00, 0, DIMX.DM01, 0, DIMX.DM02, 0, DIMX.DM03, 0);
+	dimx[0] = dimx[1].xxzzlh();
+	dimx[3] = GSVector4i(DIMX.DM10, 0, DIMX.DM11, 0, DIMX.DM12, 0, DIMX.DM13, 0);
+	dimx[2] = dimx[3].xxzzlh();
+	dimx[5] = GSVector4i(DIMX.DM20, 0, DIMX.DM21, 0, DIMX.DM22, 0, DIMX.DM23, 0);
+	dimx[4] = dimx[5].xxzzlh();
+	dimx[7] = GSVector4i(DIMX.DM30, 0, DIMX.DM31, 0, DIMX.DM32, 0, DIMX.DM33, 0);
+	dimx[6] = dimx[7].xxzzlh();
+}

pcsx2/GS/Renderers/HW/GSRendererHWMultiISA.cpp

@@ -36,6 +36,10 @@ void CURRENT_ISA::GSRendererHWPopulateFunctions(GSRendererHW& renderer)
 	GSRendererHWFunctions::Populate(renderer);
 }
 
+// since there's no overlapping draws, we can just keep this intact
+static GSVector4i s_dimx_storage[8];
+static GIFRegDIMX s_last_dimx;
+
 bool GSRendererHWFunctions::SwPrimRender(GSRendererHW& hw, bool invalidate_tc)
 {
 	GSVertexTrace& vt = hw.m_vt;
@@ -47,9 +51,6 @@ bool GSRendererHWFunctions::SwPrimRender(GSRendererHW& hw, bool invalidate_tc)
 	GSRasterizerData data;
 	GSScanlineGlobalData& gd = data.global;
 
-	u32 clut_storage[256];
-	GSVector4i dimx_storage[8];
-
 	hw.m_sw_vertex_buffer.resize(((hw.m_vertex.next + 1) & ~1));
 
 	data.primclass = vt.m_primclass;
@@ -180,9 +181,7 @@ bool GSRendererHWFunctions::SwPrimRender(GSRendererHW& hw, bool invalidate_tc)
 			{
 				gd.sel.tlu = 1;
 
-				gd.clut = clut_storage; // FIXME: might address uninitialized data of the texture (0xCD) that is not in 0-15 range for 4-bpp formats
+				gd.clut = const_cast<u32*>(static_cast<const u32*>(hw.m_mem.m_clut));
-
-				memcpy(gd.clut, (const u32*)hw.m_mem.m_clut, sizeof(u32) * GSLocalMemory::m_psm[context->TEX0.PSM].pal);
 			}
 
 			gd.sel.wms = context->CLAMP.WMS;
@@ -452,10 +451,12 @@ bool GSRendererHWFunctions::SwPrimRender(GSRendererHW& hw, bool invalidate_tc)
 		if (env.DTHE.DTHE)
 		{
 			gd.sel.dthe = 1;
-
+			gd.dimx = s_dimx_storage;
-			gd.dimx = dimx_storage;
+			if (s_last_dimx != env.DIMX)
-
+			{
-			memcpy(gd.dimx, hw.dimx, sizeof(hw.dimx));
+				s_last_dimx = env.DIMX;
+				GSState::ExpandDIMX(s_dimx_storage, env.DIMX);
+			}
 		}
 	}
 

pcsx2/GS/Renderers/SW/GSRendererSW.cpp

@@ -1327,9 +1327,15 @@ bool GSRendererSW::GetScanlineGlobalData(SharedData* data)
 		{
 			gd.sel.dthe = 1;
 
-			gd.dimx = (GSVector4i*)m_vertex_heap.alloc(sizeof(dimx), 32);
+			if (m_last_dimx != env.DIMX)
+			{
+				m_last_dimx = env.DIMX;
+				ExpandDIMX(m_dimx, env.DIMX);
+			}
 
-			memcpy(gd.dimx, dimx, sizeof(dimx));
+			gd.dimx = (GSVector4i*)m_vertex_heap.alloc(sizeof(m_dimx), 32);
+
+			std::memcpy(gd.dimx, m_dimx, sizeof(m_dimx));
 		}
 	}
 

pcsx2/GS/Renderers/SW/GSRendererSW.h

@@ -69,6 +69,8 @@ protected:
 	u32 m_fzb_cur_pages[16];
 	std::atomic<u32> m_fzb_pages[512]; // u16 frame/zbuf pages interleaved
 	std::atomic<u16> m_tex_pages[512];
+	GIFRegDIMX m_last_dimx = {};
+	GSVector4i m_dimx[8] = {};
 
 	void Reset(bool hardware_reset) override;
 	void VSync(u32 field, bool registers_written) override;