Merge pull request #1078 from PCSX2/gsdx-unsafe-fbmask

Gsdx unsafe fbmask

plugins/GSdx/GSLinuxDialog.cpp

@@ -351,6 +351,7 @@ void populate_hack_table(GtkWidget* hack_table)
 	GtkWidget* hack_tco_entry      = CreateTextBox("UserHacks_TCOffset");
 	GtkWidget* align_sprite_check  = CreateCheckBox("Align sprite hack", "UserHacks_align_sprite_X");
 	GtkWidget* preload_gs_check    = CreateCheckBox("Preload Frame", "preload_frame_with_gs_data");
+	GtkWidget* hack_unsafe_fbmask  = CreateCheckBox("Fast Accurate Blending", "UserHacks_unsafe_fbmask");
 
 	GtkWidget* hack_sprite_box     = CreateComboBoxFromVector(theApp.m_gs_hack, "UserHacks_SpriteHack");
 	GtkWidget* hack_sprite_label   = left_label("Alpha-Sprite Hack:");
@@ -375,6 +376,7 @@ void populate_hack_table(GtkWidget* hack_table)
 	InsertWidgetInTable(hack_table , hack_enble_check);
 	InsertWidgetInTable(hack_table , hack_wild_check     , align_sprite_check);
 	InsertWidgetInTable(hack_table , hack_offset_check   , preload_gs_check);
+	InsertWidgetInTable(hack_table , hack_unsafe_fbmask);
 	InsertWidgetInTable(hack_table , hack_sprite_label   , hack_sprite_box );
 	InsertWidgetInTable(hack_table , stretch_hack_label  , stretch_hack_box );
 	InsertWidgetInTable(hack_table , hack_skipdraw_label , hack_skipdraw_spin);

plugins/GSdx/GSRendererOGL.cpp

@@ -34,6 +34,7 @@ GSRendererOGL::GSRendererOGL()
 	UserHacks_TCOffset       = theApp.GetConfig("UserHacks_TCOffset", 0);
 	UserHacks_TCO_x          = (UserHacks_TCOffset & 0xFFFF) / -1000.0f;
 	UserHacks_TCO_y          = ((UserHacks_TCOffset >> 16) & 0xFFFF) / -1000.0f;
+	UserHacks_unsafe_fbmask  = theApp.GetConfig("UserHacks_unsafe_fbmask", false);
 
 	m_prim_overlap = PRIM_OVERLAP_UNKNOW;
 
@@ -41,6 +42,7 @@ GSRendererOGL::GSRendererOGL()
 		UserHacks_TCOffset       = 0;
 		UserHacks_TCO_x          = 0;
 		UserHacks_TCO_y          = 0;
+		UserHacks_unsafe_fbmask  = false;
 	}
 }
 
@@ -299,12 +301,39 @@ bool GSRendererOGL::EmulateTextureShuffleAndFbmask(GSDeviceOGL::PSSelector& ps_s
 		ps_sel.fbmask = m_sw_blending && (~ff_fbmask & ~zero_fbmask & 0xF);
 
 		if (ps_sel.fbmask) {
+			ps_cb.FbMask = fbmask_v.u8to32();
+			// Only alpha is special here, I think we can take a very unsafe shortcut
+			// Alpha isn't blended on the GS but directly copyied into the RT.
+			//
+			// Behavior is clearly undefined however there is a high probability that
+			// it will work. Masked bit will be constant and normally the same everywhere
+			// RT/FS output/Cached value.
+			//
+			// Just to be sure let's add a new safe hack for unsafe access :)
+			//
+			// Here the GL spec quote to emphasize the unexpected behavior.
+			/*
+			   - If a texel has been written, then in order to safely read the result
+			   a texel fetch must be in a subsequent Draw separated by the command
+
+			   void TextureBarrier(void);
+
+			   TextureBarrier() will guarantee that writes have completed and caches
+			   have been invalidated before subsequent Draws are executed.
+			 */
+			if (!(~ff_fbmask & ~zero_fbmask & 0x7) && UserHacks_unsafe_fbmask) {
+				GL_INS("FBMASK Unsafe SW emulated fb_mask:%x on %d bits format", m_context->FRAME.FBMSK,
+						(GSLocalMemory::m_psm[m_context->FRAME.PSM].fmt == 2) ? 16 : 32);
+				m_unsafe_fbmask = true;
+				require_barrier = false;
+			} else {
+				// The safe and accurate path (but slow)
 				GL_INS("FBMASK SW emulated fb_mask:%x on %d bits format", m_context->FRAME.FBMSK,
 						(GSLocalMemory::m_psm[m_context->FRAME.PSM].fmt == 2) ? 16 : 32);
-			ps_cb.FbMask = fbmask_v.u8to32();
 				require_barrier = true;
 			}
 		}
+	}
 
 	return require_barrier;
 }
@@ -358,7 +387,7 @@ bool GSRendererOGL::EmulateBlending(GSDeviceOGL::PSSelector& ps_sel, bool DATE_G
 												((ALPHA.C == 0 && m_vt.m_alpha.max > 128) || (ALPHA.C == 2 && ALPHA.FIX > 128u));
 		case ACC_BLEND_CCLIP_DALPHA:    sw_blending |= (ALPHA.C == 1) || (m_env.COLCLAMP.CLAMP == 0);
 		case ACC_BLEND_SPRITE:          sw_blending |= m_vt.m_primclass == GS_SPRITE_CLASS;
-		case ACC_BLEND_FREE:            sw_blending |= ps_sel.fbmask || impossible_or_free_blend;
+		case ACC_BLEND_FREE:            sw_blending |= (ps_sel.fbmask  && !m_unsafe_fbmask) || impossible_or_free_blend; // blending is only free when we use slow fbmask
 		default:                        sw_blending |= accumulation_blend;
 	}
 	// SW Blending
@@ -566,7 +595,12 @@ void GSRendererOGL::SendDraw(bool require_barrier)
 {
 	GSDeviceOGL* dev = (GSDeviceOGL*)m_dev;
 
-	if (!require_barrier) {
+	if (!require_barrier && m_unsafe_fbmask) {
+		// Not safe but still worth to take some precautions.
+		ASSERT(GLLoader::found_GL_ARB_texture_barrier);
+		glTextureBarrier();
+		dev->DrawIndexedPrimitive();
+	} else if (!require_barrier) {
 		dev->DrawIndexedPrimitive();
 	} else if (m_prim_overlap == PRIM_OVERLAP_NO) {
 		ASSERT(GLLoader::found_GL_ARB_texture_barrier);
@@ -621,6 +655,7 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
 	bool DATE_GL45 = false;
 
 	bool require_barrier = false; // For accurate option
+	m_unsafe_fbmask = false;
 
 	ASSERT(m_dev != NULL);
 

plugins/GSdx/GSRendererOGL.h

@@ -48,9 +48,11 @@ class GSRendererOGL : public GSRendererHW
 		bool m_accurate_date;
 		int m_sw_blending;
 		PRIM_OVERLAP m_prim_overlap;
+		bool m_unsafe_fbmask;
 
 		unsigned int UserHacks_TCOffset;
 		float UserHacks_TCO_x, UserHacks_TCO_y;
+		bool UserHacks_unsafe_fbmask;
 
 		GSDeviceOGL::VSConstantBuffer vs_cb;
 		GSDeviceOGL::PSConstantBuffer ps_cb;

plugins/GSdx/GSSetting.cpp

@@ -102,6 +102,9 @@ const char* dialog_message(int ID, bool* updateText) {
 				"High:\nExtend it to destination alpha blending and color wrapping. (help shadow and fog effect). A good CPU is required.\n\n"
 				"Full:\nExcept few cases, the blending unit will be fully emulated by the shader. It is ultra slow! It is intended for debug.\n\n"
 				"Ultra:\nThe blending unit will be completely emulated by the shader. It is ultra slow! It is intended for debug.";
+		case IDC_UNSAFE_FBMASK:
+			return "Relies on undefined hardware behavior to accelerate accurate blending operation.\n\n"
+				"It provides a major boost on the Xenosaga Series.\n";
 		case IDC_TC_DEPTH:
 			return "Allows the conversion of Depth buffer from/to Color buffer. It is used for blur & depth of field effects";
 		case IDC_AFCOMBO:

plugins/GSdx/GSSetting.h

@@ -60,6 +60,7 @@ enum {
 	IDC_TCOFFSETY2,
 	IDC_PALTEX,
 	IDC_ACCURATE_BLEND_UNIT,
+	IDC_UNSAFE_FBMASK,
 	IDC_ACCURATE_DATE,
 	IDC_TC_DEPTH,
 	IDC_CRC_LEVEL,

plugins/GSdx/GSSettingsDlg.cpp

@@ -658,6 +658,8 @@ void GSHacksDlg::OnInit()
 	CheckDlgButton(m_hWnd, IDC_ALPHASTENCIL, theApp.GetConfig("UserHacks_AlphaStencil", 0));
 	CheckDlgButton(m_hWnd, IDC_PRELOAD_GS, theApp.GetConfig("preload_frame_with_gs_data", 0));
 	CheckDlgButton(m_hWnd, IDC_ALIGN_SPRITE, theApp.GetConfig("UserHacks_align_sprite_X", 0));
+	CheckDlgButton(m_hWnd, IDC_UNSAFE_FBMASK, theApp.GetConfig("UserHacks_unsafe_fbmask", 0));
+
 
 	ComboBoxInit(IDC_ROUND_SPRITE, theApp.m_gs_hack, theApp.GetConfig("UserHacks_round_sprite_offset", 0));
 	ComboBoxInit(IDC_SPRITEHACK, theApp.m_gs_hack, theApp.GetConfig("UserHacks_SpriteHack", 0));
@@ -673,6 +675,8 @@ void GSHacksDlg::OnInit()
 
 	ShowWindow(GetDlgItem(m_hWnd, IDC_ALPHASTENCIL), ogl ? SW_HIDE : SW_SHOW);
 	ShowWindow(GetDlgItem(m_hWnd, IDC_ALPHAHACK), ogl ? SW_HIDE : SW_SHOW);
+	ShowWindow(GetDlgItem(m_hWnd, IDC_UNSAFE_FBMASK), ogl ? SW_SHOW : SW_HIDE);
+
 
 	AddTooltip(IDC_SKIPDRAWHACKEDIT);
 	AddTooltip(IDC_SKIPDRAWHACK);
@@ -689,6 +693,7 @@ void GSHacksDlg::OnInit()
 	AddTooltip(IDC_TCOFFSETY);
 	AddTooltip(IDC_TCOFFSETY2);
 	AddTooltip(IDC_PRELOAD_GS);
+	AddTooltip(IDC_UNSAFE_FBMASK);
 }
 
 void GSHacksDlg::UpdateControls()
@@ -723,6 +728,8 @@ bool GSHacksDlg::OnMessage(UINT message, WPARAM wParam, LPARAM lParam)
 			theApp.SetConfig("UserHacks_AlphaStencil", (int)IsDlgButtonChecked(m_hWnd, IDC_ALPHASTENCIL));
 			theApp.SetConfig("preload_frame_with_gs_data", (int)IsDlgButtonChecked(m_hWnd, IDC_PRELOAD_GS));
 			theApp.SetConfig("Userhacks_align_sprite_X", (int)IsDlgButtonChecked(m_hWnd, IDC_ALIGN_SPRITE));
+			theApp.SetConfig("UserHacks_unsafe_fbmask", (int)IsDlgButtonChecked(m_hWnd, IDC_UNSAFE_FBMASK));
+
 
 			unsigned int TCOFFSET  =  SendMessage(GetDlgItem(m_hWnd, IDC_TCOFFSETX), UDM_GETPOS, 0, 0) & 0xFFFF;
 						 TCOFFSET |= (SendMessage(GetDlgItem(m_hWnd, IDC_TCOFFSETY), UDM_GETPOS, 0, 0) & 0xFFFF) << 16;

plugins/GSdx/GSdx.rc

@@ -105,6 +105,7 @@ BEGIN
     CONTROL         "Half-pixel Offset",IDC_OFFSETHACK,"Button",BS_AUTOCHECKBOX | WS_TABSTOP,14,51,70,8
     COMBOBOX        IDC_MSAACB,88,17,58,63,CBS_DROPDOWNLIST | WS_VSCROLL | WS_TABSTOP
     CONTROL         "Wild Arms Offset",IDC_WILDHACK,"Button",BS_AUTOCHECKBOX | WS_TABSTOP,14,64,70,8
+    CONTROL         "Fast accurate blending",IDC_UNSAFE_FBMASK,"Button",BS_AUTOCHECKBOX | WS_TABSTOP,14,77,90,8
     CONTROL         "Alpha Stencil",IDC_ALPHASTENCIL,"Button",BS_AUTOCHECKBOX | WS_TABSTOP,14,77,57,8
     CONTROL         "Align Sprite",IDC_ALIGN_SPRITE,"Button",BS_AUTOCHECKBOX | WS_TABSTOP,88,51,58,8
     RTEXT           "TC Offset X:",IDC_STATIC,40,146,44,8

plugins/GSdx/resource.h

@@ -89,6 +89,7 @@
 #define IDC_MIPMAP                      2084
 #define IDC_PRELOAD_GS                  2085
 #define IDC_TVSHADER                    2086
+#define IDC_UNSAFE_FBMASK               2087
 #define IDR_CONVERT_FX                  10000
 #define IDR_TFX_FX                      10001
 #define IDR_MERGE_FX                    10002
@@ -107,7 +108,7 @@
 #ifndef APSTUDIO_READONLY_SYMBOLS
 #define _APS_NEXT_RESOURCE_VALUE        10013
 #define _APS_NEXT_COMMAND_VALUE         32771
-#define _APS_NEXT_CONTROL_VALUE         2087
+#define _APS_NEXT_CONTROL_VALUE         2088
 #define _APS_NEXT_SYMED_VALUE           5000
 #endif
 #endif