GS/HW: Add/adjust logs for failed texture creation.

Bonus, add missing texture copies info when doing shader copies.

.github/workflows/linux_build_flatpak.yml

@@ -132,7 +132,7 @@ jobs:
 
       - name: Push to Flathub (beta)
         if: ${{ inputs.publish == true && (inputs.stableBuild == false || inputs.stableBuild == 'false') }}
-        uses: flatpak/flatpak-github-actions/flatpak-builder@10a3c29f0162516f0f68006be14c92f34bd4fa6c
+        uses: flatpak/flatpak-github-actions/flat-manager@10a3c29f0162516f0f68006be14c92f34bd4fa6c
         with:
           flat-manager-url: https://hub.flathub.org/
           repository: beta
@@ -141,7 +141,7 @@ jobs:
 
       - name: Push to Flathub (stable)
         if: ${{ inputs.publish == true && (inputs.stableBuild == true || inputs.stableBuild == 'true') }}
-        uses: flatpak/flatpak-github-actions/flatpak-builder@10a3c29f0162516f0f68006be14c92f34bd4fa6c
+        uses: flatpak/flatpak-github-actions/flat-manager@10a3c29f0162516f0f68006be14c92f34bd4fa6c
         with:
           flat-manager-url: https://hub.flathub.org/
           repository: stable

pcsx2-qt/Translations/pcsx2-qt_en.ts

@@ -17254,7 +17254,7 @@ The saves will not be recoverable.</source>
     <name>MemoryCard</name>
     <message>
         <location filename="../../pcsx2/SIO/Memcard/MemoryCardFile.cpp" line="289"/>
-        <location filename="../../pcsx2/SIO/Memcard/MemoryCardFile.cpp" line="969"/>
+        <location filename="../../pcsx2/SIO/Memcard/MemoryCardFile.cpp" line="970"/>
         <source>Memory Card Creation Failed</source>
         <translation type="unfinished"></translation>
     </message>
@@ -17287,7 +17287,7 @@ Close any other instances of PCSX2, or restart your computer.
         <translation type="unfinished"></translation>
     </message>
     <message>
-        <location filename="../../pcsx2/SIO/Memcard/MemoryCardFile.cpp" line="970"/>
+        <location filename="../../pcsx2/SIO/Memcard/MemoryCardFile.cpp" line="971"/>
         <source>Failed to create memory card. The error was:
 {}</source>
         <translation type="unfinished"></translation>

pcsx2/FPU.cpp

@@ -249,7 +249,6 @@ void CTC1() {
 
 void CVT_S() {
 	_FdValf_ = (float)_FsValSl_;
-	_FdValf_ = fpuDouble( _FdValUl_ );
 }
 
 void CVT_W() {

pcsx2/GS/Renderers/DX11/GSDevice11.cpp

@@ -2507,15 +2507,17 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 		{
 			config.colclip_update_area = config.drawarea;
 
-			const GSVector4 dRect = GSVector4((config.colclip_mode == GSHWDrawConfig::ColClipMode::ConvertOnly) ? GSVector4i::loadh(rtsize) : config.drawarea);
-			const GSVector4 sRect = dRect / GSVector4(rtsize.x, rtsize.y).xyxy();
 			colclip_rt = CreateRenderTarget(rtsize.x, rtsize.y, GSTexture::Format::ColorClip);
 			if (!colclip_rt)
+			{
+				Console.Warning("D3D11: Failed to allocate ColorClip render target, aborting draw.");
 				return;
+			}
 
 			g_gs_device->SetColorClipTexture(colclip_rt);
-			// Warning: StretchRect must be called before BeginScene otherwise
-			// vertices will be overwritten. Trust me you don't want to do that.
+
+			const GSVector4 dRect = GSVector4((config.colclip_mode == GSHWDrawConfig::ColClipMode::ConvertOnly) ? GSVector4i::loadh(rtsize) : config.drawarea);
+			const GSVector4 sRect = dRect / GSVector4(rtsize.x, rtsize.y).xyxy();
 			StretchRect(config.rt, sRect, colclip_rt, dRect, ShaderConvert::COLCLIP_INIT, false);
 			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 		}
@@ -2526,7 +2528,10 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 	{
 		primid_tex = CreateRenderTarget(rtsize.x, rtsize.y, GSTexture::Format::PrimID, false);
 		if (!primid_tex)
+		{
+			Console.WriteLn("D3D11: Failed to allocate DATE image, aborting draw.");
 			return;
+		}
 
 		StretchRect(colclip_rt ? colclip_rt : config.rt, GSVector4(config.drawarea) / GSVector4(rtsize).xyxy(),
 			primid_tex, GSVector4(config.drawarea), m_date.primid_init_ps[static_cast<u8>(config.datm)].get(), nullptr, false);
@@ -2617,6 +2622,9 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 			if (config.tex && config.tex == config.rt)
 				PSSetShaderResource(0, draw_rt_clone);
 		}
+		else
+			Console.Warning("D3D11: Failed to allocate temp texture for RT copy.");
+
 	}
 
 	GSTexture* draw_ds_clone = nullptr;
@@ -2630,6 +2638,8 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 			CopyRect(config.ds, draw_ds_clone, config.drawarea, config.drawarea.left, config.drawarea.top);
 			PSSetShaderResource(0, draw_ds_clone);
 		}
+		else
+			Console.Warning("D3D11: Failed to allocate temp texture for DS copy.");
 	}
 
 	SetupVS(config.vs, &config.cb_vs);

pcsx2/GS/Renderers/DX12/GSDevice12.cpp

@@ -3849,7 +3849,7 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 			colclip_rt->TransitionToState(D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE);
 
 			draw_rt = static_cast<GSTexture12*>(config.rt);
-			OMSetRenderTargets(draw_rt, draw_ds, config.scissor);
+			OMSetRenderTargets(draw_rt, draw_ds, config.colclip_update_area);
 
 			// if this target was cleared and never drawn to, perform the clear as part of the resolve here.
 			BeginRenderPass(GetLoadOpForTexture(draw_rt), D3D12_RENDER_PASS_ENDING_ACCESS_TYPE_PRESERVE,
@@ -3917,7 +3917,7 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 		{
 			EndRenderPass();
 
-			GL_PUSH("Copy RT to temp texture {%d,%d %dx%d}", config.drawarea.left, config.drawarea.top,
+			GL_PUSH("D3D12: Copy RT to temp texture {%d,%d %dx%d}", config.drawarea.left, config.drawarea.top,
 				config.drawarea.width(), config.drawarea.height());
 
 			draw_rt_clone->SetState(GSTexture::State::Invalidated);
@@ -3927,6 +3927,8 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 			if (config.tex && config.tex == config.rt)
 				PSSetShaderResource(0, draw_rt_clone, true);
 		}
+		else
+			Console.Warning("D3D12: Failed to allocate temp texture for RT copy.");
 	}
 
 	if (config.tex && config.tex == config.ds)
@@ -3937,13 +3939,15 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 		{
 			EndRenderPass();
 
-			GL_PUSH("Copy RT to temp texture {%d,%d %dx%d}", config.drawarea.left, config.drawarea.top,
+			GL_PUSH("D3D12: Copy DS to temp texture {%d,%d %dx%d}", config.drawarea.left, config.drawarea.top,
 				config.drawarea.width(), config.drawarea.height());
 
 			draw_ds_clone->SetState(GSTexture::State::Invalidated);
 			CopyRect(config.ds, draw_ds_clone, config.drawarea, config.drawarea.left, config.drawarea.top);
 			PSSetShaderResource(0, draw_ds_clone, true);
 		}
+		else
+			Console.Warning("D3D12: Failed to allocate temp texture for DS copy.");
 	}
 
 	// Switch to colclip target for colclip hw rendering
@@ -3958,7 +3962,7 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 			colclip_rt = static_cast<GSTexture12*>(CreateRenderTarget(rtsize.x, rtsize.y, GSTexture::Format::ColorClip, false));
 			if (!colclip_rt)
 			{
-				Console.WriteLn("D3D12: Failed to allocate ColorClip render target, aborting draw.");
+				Console.Warning("D3D12: Failed to allocate ColorClip render target, aborting draw.");
 
 				if (date_image)
 					Recycle(date_image);
@@ -4038,6 +4042,7 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 	// rt -> colclip hw blit if enabled
 	if (colclip_rt && (config.colclip_mode == GSHWDrawConfig::ColClipMode::ConvertOnly || config.colclip_mode == GSHWDrawConfig::ColClipMode::ConvertAndResolve) && config.rt->GetState() == GSTexture::State::Dirty)
 	{
+		OMSetRenderTargets(draw_rt, draw_ds, GSVector4i::loadh(rtsize));
 		SetUtilityTexture(static_cast<GSTexture12*>(config.rt), m_point_sampler_cpu);
 		SetPipeline(m_colclip_setup_pipelines[pipe.ds].get());
 
@@ -4047,8 +4052,10 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 		g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 
 		GL_POP();
-	}
 
+		// Restore original scissor, not sure if needed since the render pass has already been started. But to be safe.
+		OMSetRenderTargets(draw_rt, draw_ds, config.scissor);
+	}
 	// VB/IB upload, if we did DATE setup and it's not colclip hw this has already been done
 	SetPrimitiveTopology(s_primitive_topology_mapping[static_cast<u8>(config.topology)]);
 	if (!date_image || colclip_rt)
@@ -4111,7 +4118,7 @@ void GSDevice12::RenderHW(GSHWDrawConfig& config)
 			colclip_rt->TransitionToState(D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE);
 
 			draw_rt = static_cast<GSTexture12*>(config.rt);
-			OMSetRenderTargets(draw_rt, draw_ds, config.scissor);
+			OMSetRenderTargets(draw_rt, draw_ds, config.colclip_update_area);
 
 			// if this target was cleared and never drawn to, perform the clear as part of the resolve here.
 			BeginRenderPass(GetLoadOpForTexture(draw_rt), D3D12_RENDER_PASS_ENDING_ACCESS_TYPE_PRESERVE,

pcsx2/GS/Renderers/HW/GSTextureCache.cpp

@@ -1588,14 +1588,13 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
 					//	DevCon.Warning("Expected %x Got %x shuffle %d draw %d", psm, t_psm, possible_shuffle, GSState::s_n);
 					if (match)
 					{
-						// It is a complex to convert the code in shader. As a reference, let's do it on the CPU, it will be slow but
-						// 1/ it just works :)
-						// 2/ even with upscaling
-						// 3/ for both Direct3D and OpenGL
-						if (psm == PSMT4 || (GSConfig.UserHacks_CPUFBConversion && psm == PSMT8))
+						// It is a complex to convert the code in shader. As a reference, let's do it on the CPU,
+						// it will be slow but can work even with upscaling, also fine tune it so it's not enabled when not needed.
+						if (psm == PSMT4 || (GSConfig.UserHacks_CPUFBConversion && psm == PSMT8 && (!possible_shuffle || GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp != 32)))
 						{
 							// Forces 4-bit and 8-bit frame buffer conversion to be done on the CPU instead of the GPU, but performance will be slower.
-							// There is no dedicated shader to handle 4-bit conversion (Stuntman has been confirmed to use 4-bit).
+							// There is no dedicated shader to handle 4-bit conversion (Beyond Good and Evil and Stuntman).
+							// Note: Stuntman no longer hits the PSMT4 code path.
 							// Direct3D10/11 and OpenGL support 8-bit fb conversion but don't render some corner cases properly (Harry Potter games).
 							// The hack can fix glitches in some games.
 							if (!t->m_drawn_since_read.rempty())

pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp

@@ -2424,7 +2424,7 @@ void GSDeviceOGL::RenderHW(GSHWDrawConfig& config)
 			const GSVector4 dRect(config.colclip_update_area);
 			const GSVector4 sRect = dRect / GSVector4(size.x, size.y).xyxy();
 			StretchRect(colclip_rt, sRect, config.rt, dRect, ShaderConvert::COLCLIP_RESOLVE, false);
-
+			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			Recycle(colclip_rt);
 
 			g_gs_device->SetColorClipTexture(nullptr);
@@ -2444,6 +2444,14 @@ void GSDeviceOGL::RenderHW(GSHWDrawConfig& config)
 			config.colclip_update_area = config.drawarea;
 
 			colclip_rt = CreateRenderTarget(rtsize.x, rtsize.y, GSTexture::Format::ColorClip, false);
+
+			if (!colclip_rt)
+			{
+				Console.Warning("GL: Failed to allocate ColorClip render target, aborting draw.");
+
+				return;
+			}
+
 			OMSetRenderTargets(colclip_rt, config.ds, nullptr);
 
 			g_gs_device->SetColorClipTexture(colclip_rt);
@@ -2451,6 +2459,7 @@ void GSDeviceOGL::RenderHW(GSHWDrawConfig& config)
 			const GSVector4 dRect = GSVector4((config.colclip_mode == GSHWDrawConfig::ColClipMode::ConvertOnly) ? GSVector4i::loadh(rtsize) : config.drawarea);
 			const GSVector4 sRect = dRect / GSVector4(rtsize.x, rtsize.y).xyxy();
 			StretchRect(config.rt, sRect, colclip_rt, dRect, ShaderConvert::COLCLIP_INIT, false);
+			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 		}
 	}
 
@@ -2462,6 +2471,11 @@ void GSDeviceOGL::RenderHW(GSHWDrawConfig& config)
 			break; // No setup
 		case GSHWDrawConfig::DestinationAlphaMode::PrimIDTracking:
 			primid_texture = InitPrimDateTexture(colclip_rt ? colclip_rt : config.rt, config.drawarea, config.datm);
+			if (!primid_texture)
+			{
+				Console.WriteLn("GL: Failed to allocate DATE image, aborting draw.");
+				return;
+			}
 			break;
 		case GSHWDrawConfig::DestinationAlphaMode::StencilOne:
 			if (m_features.texture_barrier)
@@ -2498,6 +2512,8 @@ void GSDeviceOGL::RenderHW(GSHWDrawConfig& config)
 				config.drawarea.width(), config.drawarea.height());
 			CopyRect(colclip_rt ? colclip_rt : config.rt, draw_rt_clone, config.drawarea, config.drawarea.left, config.drawarea.top);
 		}
+		else
+			Console.Warning("GL: Failed to allocate temp texture for RT copy.");
 	}
 
 	IASetVertexBuffer(config.verts, config.nverts);
@@ -2713,7 +2729,7 @@ void GSDeviceOGL::RenderHW(GSHWDrawConfig& config)
 			const GSVector4 dRect(config.colclip_update_area);
 			const GSVector4 sRect = dRect / GSVector4(size.x, size.y).xyxy();
 			StretchRect(colclip_rt, sRect, config.rt, dRect, ShaderConvert::COLCLIP_RESOLVE, false);
-
+			g_perfmon.Put(GSPerfMon::TextureCopies, 1);
 			Recycle(colclip_rt);
 
 			g_gs_device->SetColorClipTexture(nullptr);

pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp

@@ -5625,7 +5625,7 @@ void GSDeviceVK::RenderHW(GSHWDrawConfig& config)
 		date_image = SetupPrimitiveTrackingDATE(config);
 		if (!date_image)
 		{
-			Console.WriteLn("Failed to allocate DATE image, aborting draw.");
+			Console.WriteLn("VK: Failed to allocate DATE image, aborting draw.");
 			return;
 		}
 
@@ -5726,12 +5726,14 @@ void GSDeviceVK::RenderHW(GSHWDrawConfig& config)
 		{
 			EndRenderPass();
 
-			GL_PUSH("Copy RT to temp texture for fbmask {%d,%d %dx%d}", config.drawarea.left, config.drawarea.top,
+			GL_PUSH("VK: Copy RT to temp texture for fbmask {%d,%d %dx%d}", config.drawarea.left, config.drawarea.top,
 				config.drawarea.width(), config.drawarea.height());
 
 			CopyRect(draw_rt, draw_rt_clone, config.drawarea, config.drawarea.left, config.drawarea.top);
 			PSSetShaderResource(2, draw_rt_clone, true);
 		}
+		else
+			Console.Warning("VK: Failed to allocate temp texture for RT copy.");
 	}
 
 	// Switch to colclip target for colclip hw rendering
@@ -5744,7 +5746,7 @@ void GSDeviceVK::RenderHW(GSHWDrawConfig& config)
 			colclip_rt = static_cast<GSTextureVK*>(CreateRenderTarget(rtsize.x, rtsize.y, GSTexture::Format::ColorClip, false));
 			if (!colclip_rt)
 			{
-				Console.WriteLn("Failed to allocate ColorClip render target, aborting draw.");
+				Console.Warning("VK: Failed to allocate ColorClip render target, aborting draw.");
 
 				if (date_image)
 					Recycle(date_image);

pcsx2/SIO/Memcard/MemoryCardFile.cpp

@@ -887,6 +887,7 @@ std::vector<AvailableMcdInfo> FileMcd_GetAvailableCards(bool include_in_use_card
 		}
 	}
 
+	std::sort(mcds.begin(), mcds.end(), [](auto& a, auto& b) { return a.name < b.name; });
 	return mcds;
 }
 

pcsx2/VUops.cpp

@@ -879,12 +879,10 @@ static __fi u32 floatToInt(u32 uvalue)
 	float fvalue = std::bit_cast<float>(uvalue);
 	if (Offset)
 		fvalue *= std::bit_cast<float>(0x3f800000 + (Offset << 23));
-	s32 svalue = std::bit_cast<s32>(fvalue);
+	uvalue = std::bit_cast<u32>(fvalue);
 
-	if (svalue >= static_cast<s32>(0x4f000000))
-		return 0x7fffffff;
-	else if (svalue <= static_cast<s32>(0xcf000000))
-		return 0x80000000;
+	if ((uvalue & 0x7f800000) >= 0x4f000000)
+		return (uvalue & 0x80000000) ? 0x80000000 : 0x7fffffff;
 	else
 		return static_cast<u32>(static_cast<s32>(fvalue));
 }

pcsx2/x86/iFPU.cpp

@@ -26,8 +26,6 @@ namespace DOUBLE
 	void recC_EQ_xmm(int info);
 	void recC_LE_xmm(int info);
 	void recC_LT_xmm(int info);
-	void recCVT_S_xmm(int info);
-	void recCVT_W();
 	void recDIV_S_xmm(int info);
 	void recMADD_S_xmm(int info);
 	void recMADDA_S_xmm(int info);
@@ -993,15 +991,16 @@ void recCVT_S_xmm(int info)
 	}
 }
 
-FPURECOMPILE_CONSTCODE(CVT_S, XMMINFO_WRITED | XMMINFO_READS);
+void recCVT_S()
+{
+	// Float version is fully accurate, no double version
+	eeFPURecompileCode(recCVT_S_xmm, R5900::Interpreter::OpcodeImpl::COP1::CVT_S, XMMINFO_WRITED | XMMINFO_READS);
+}
 
 void recCVT_W()
 {
-	if (CHECK_FPU_FULL)
-	{
-		DOUBLE::recCVT_W();
-		return;
-	}
+	// Float version is fully accurate, no double version
+
 	// If we have the following EmitOP() on the top then it'll get calculated twice when CHECK_FPU_FULL is true
 	// as we also have an EmitOP() at recCVT_W() on iFPUd.cpp.  hence we have it below the possible return.
 	EE::Profiler.EmitOp(eeOpcode::CVTW);
@@ -1010,26 +1009,23 @@ void recCVT_W()
 
 	if (regs >= 0)
 	{
-		if (CHECK_FPU_EXTRA_OVERFLOW)
-			fpuFloat2(regs);
 		xCVTTSS2SI(eax, xRegisterSSE(regs));
-		xMOVMSKPS(edx, xRegisterSSE(regs)); //extract the signs
-		xAND(edx, 1); // keep only LSB
+		xMOVD(edx, xRegisterSSE(regs));
 	}
 	else
 	{
 		xCVTTSS2SI(eax, ptr32[&fpuRegs.fpr[_Fs_]]);
 		xMOV(edx, ptr[&fpuRegs.fpr[_Fs_]]);
-		xSHR(edx, 31); // mov sign to lsb
 	}
 
 	//kill register allocation for dst because we write directly to fpuRegs.fpr[_Fd_]
 	_deleteFPtoXMMreg(_Fd_, DELETE_REG_FREE_NO_WRITEBACK);
 
-	xADD(edx, 0x7FFFFFFF); // 0x7FFFFFFF if positive, 0x8000 0000 if negative
-
-	xCMP(eax, 0x80000000); // If the result is indefinitive
-	xCMOVE(eax, edx);      // Saturate it
+	// cvttss2si converts unrepresentable values to 0x80000000, so negative values are already handled.
+	// So we just need to handle positive values.
+	xCMP(edx, 0x4f000000); // If the input is greater than INT_MAX
+	xMOV(edx, 0x7fffffff);
+	xCMOVGE(eax, edx);     // Saturate it
 
 	//Write the result
 	xMOV(ptr[&fpuRegs.fpr[_Fd_]], eax);

pcsx2/x86/iFPUd.cpp

@@ -540,57 +540,10 @@ FPURECOMPILE_CONSTCODE(C_LT, XMMINFO_READS | XMMINFO_READT);
 //------------------------------------------------------------------
 // CVT.x XMM
 //------------------------------------------------------------------
-void recCVT_S_xmm(int info)
-{
-	EE::Profiler.EmitOp(eeOpcode::CVTS_F);
 
-	if (info & PROCESS_EE_D)
-	{
-		if (info & PROCESS_EE_S)
-			xCVTDQ2PS(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
-		else
-			xCVTSI2SS(xRegisterSSE(EEREC_D), ptr32[&fpuRegs.fpr[_Fs_]]);
-	}
-	else
-	{
-		const int temp = _allocTempXMMreg(XMMT_FPS);
-		xCVTSI2SS(xRegisterSSE(temp), ptr32[&fpuRegs.fpr[_Fs_]]);
-		xMOVSS(ptr32[&fpuRegs.fpr[_Fd_]], xRegisterSSE(temp));
-		_freeXMMreg(temp);
-	}
-}
+// CVT.S: Identical to non-double variant, omitted
+// CVT.W: Identical to non-double variant, omitted
 
-FPURECOMPILE_CONSTCODE(CVT_S, XMMINFO_WRITED | XMMINFO_READS);
-
-void recCVT_W() //called from iFPU.cpp's recCVT_W
-{
-	EE::Profiler.EmitOp(eeOpcode::CVTW);
-	int regs = _checkXMMreg(XMMTYPE_FPREG, _Fs_, MODE_READ);
-
-	if (regs >= 0)
-	{
-		xCVTTSS2SI(eax, xRegisterSSE(regs));
-		xMOVMSKPS(edx, xRegisterSSE(regs)); // extract the signs
-		xAND(edx, 1);                       // keep only LSB
-	}
-	else
-	{
-		xCVTTSS2SI(eax, ptr32[&fpuRegs.fpr[_Fs_]]);
-		xMOV(edx, ptr[&fpuRegs.fpr[_Fs_]]);
-		xSHR(edx, 31); //mov sign to lsb
-	}
-
-	//kill register allocation for dst because we write directly to fpuRegs.fpr[_Fd_]
-	_deleteFPtoXMMreg(_Fd_, DELETE_REG_FREE_NO_WRITEBACK);
-
-	xADD(edx, 0x7FFFFFFF); // 0x7FFFFFFF if positive, 0x8000 0000 if negative
-
-	xCMP(eax, 0x80000000); // If the result is indefinitive
-	xCMOVE(eax, edx);      // Saturate it
-
-	//Write the result
-	xMOV(ptr[&fpuRegs.fpr[_Fd_]], eax);
-}
 //------------------------------------------------------------------
 
 

pcsx2/x86/microVU_Misc.h

@@ -42,6 +42,7 @@ struct mVU_Globals
 	u32   E4      [4] = __four(0x3933e553);
 	u32   E5      [4] = __four(0x36b63510);
 	u32   E6      [4] = __four(0x353961ac);
+	u32   I32MAXF [4] = __four(0x4effffff);
 	float FTOI_4  [4] = __four(16.0);
 	float FTOI_12 [4] = __four(4096.0);
 	float FTOI_15 [4] = __four(32768.0);

pcsx2/x86/microVU_Upper.inl

@@ -484,23 +484,19 @@ static void mVU_FTOIx(mP, const float* addr, microOpcode opEnum)
 			return;
 		const xmm& Fs = mVU.regAlloc->allocReg(_Fs_, _Ft_, _X_Y_Z_W, !((_Fs_ == _Ft_) && (_X_Y_Z_W == 0xf)));
 		const xmm& t1 = mVU.regAlloc->allocReg();
-		const xmm& t2 = mVU.regAlloc->allocReg();
 
-		// Note: For help understanding this algorithm see recVUMI_FTOI_Saturate()
-		xMOVAPS(t1, Fs);
+		// cvttps2dq returns 0x8000000 for any unrepresentable values.
+		// We want it to return 0x8000000 for negative and 0x7fffffff for positive.
+		// So for unrepresentable positive values, xor with 0xffffffff to turn 0x80000000 into 0x7fffffff.
 		if (addr)
 			xMUL.PS(Fs, ptr128[addr]);
+		xMOVAPS(t1, Fs);
+		xPCMP.GTD(t1, ptr128[mVUglob.I32MAXF]);
 		xCVTTPS2DQ(Fs, Fs);
-		xPXOR(t1, ptr128[mVUglob.signbit]);
-		xPSRA.D(t1, 31);
-		xMOVAPS(t2, Fs);
-		xPCMP.EQD(t2, ptr128[mVUglob.signbit]);
-		xAND.PS(t1, t2);
-		xPADD.D(Fs, t1);
+		xPXOR(Fs, t1);
 
 		mVU.regAlloc->clearNeeded(Fs);
 		mVU.regAlloc->clearNeeded(t1);
-		mVU.regAlloc->clearNeeded(t2);
 		mVU.profiler.EmitOp(opEnum);
 	}
 	pass3