GS/HW: Fix tex is fb log for palette conversion.

Tex is fb setting is done in Texture sampler which resulted
in hw blend being applied before with barriers on.

Tex is fb can use sw blend so this optimizes that with no cost.

Also fixes an issue where hw blend was on with fbfetch.

pcsx2/GS/Renderers/Common/GSDevice.cpp

@@ -493,6 +493,7 @@ GSTexture* GSDevice::FetchSurface(GSTexture::Type type, int width, int height, i
 			{
 				ERROR_LOG("GS: Memory allocation failure for {}x{} texture. Purging pool and retrying.", size.x, size.y);
 				PurgePool();
+				t = CreateSurface(type, size.x, size.y, levels, format);
 				if (!t)
 				{
 					ERROR_LOG("GS: Memory allocation failure for {}x{} texture after purging pool.", size.x, size.y);

pcsx2/GS/Renderers/HW/GSRendererHW.cpp

@@ -3352,6 +3352,7 @@ void GSRendererHW::Draw()
 				// Z isn't offset but RT is, so we need a temp Z to align it, hopefully nothing will ever write to the Z too, right??
 				if (ds && vertical_offset && (m_cached_ctx.ZBUF.Block() - ds->m_TEX0.TBP0) != (m_cached_ctx.FRAME.Block() - rt->m_TEX0.TBP0))
 				{
+					m_using_temp_z = true;
 					const int z_vertical_offset = ((static_cast<int>(m_cached_ctx.ZBUF.Block() - ds->m_TEX0.TBP0) / 32) / std::max(rt->m_TEX0.TBW, 1U)) * GSLocalMemory::m_psm[m_cached_ctx.ZBUF.PSM].pgs.y;
 					if (g_texture_cache->GetTemporaryZ() != nullptr)
 					{
@@ -3373,19 +3374,24 @@ void GSRendererHW::Draw()
 					if (g_texture_cache->GetTemporaryZ() == nullptr)
 					{
 						m_temp_z_full_copy = false;
-						u32 vertical_size = std::max(rt->m_unscaled_size.y, ds->m_unscaled_size.y);
-						GL_CACHE("HW: RT in RT Z copy on draw %d z_vert_offset %d z_offset %d", s_n, z_vertical_offset, vertical_offset);
-						GSVector4i dRect = GSVector4i(0, vertical_offset * ds->m_scale, ds->m_unscaled_size.x * ds->m_scale, (vertical_offset + ds->m_unscaled_size.y - z_vertical_offset) * ds->m_scale);
+						const u32 vertical_size = std::max(rt->m_unscaled_size.y, ds->m_unscaled_size.y);
+						const GSVector4i dRect = GSVector4i(0, vertical_offset * ds->m_scale, ds->m_unscaled_size.x * ds->m_scale, (vertical_offset + ds->m_unscaled_size.y - z_vertical_offset) * ds->m_scale);
 						const int new_height = std::max(static_cast<int>(vertical_size * ds->m_scale), dRect.w);
-						GSTexture* tex = g_gs_device->CreateDepthStencil(ds->m_unscaled_size.x * ds->m_scale, new_height, GSTexture::Format::DepthStencil, true);
-						g_gs_device->StretchRect(ds->m_texture, GSVector4(0.0f, z_vertical_offset / static_cast<float>(ds->m_unscaled_size.y), 1.0f, (ds->m_unscaled_size.y - z_vertical_offset) / static_cast<float>(ds->m_unscaled_size.y)), tex, GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
-						g_perfmon.Put(GSPerfMon::TextureCopies, 1);
-						g_texture_cache->SetTemporaryZ(tex);
-						g_texture_cache->SetTemporaryZInfo(ds->m_TEX0.TBP0, vertical_offset - z_vertical_offset);
-						t_size.y = std::max(new_height, t_size.y);
+						if (GSTexture* tex = g_gs_device->CreateDepthStencil(ds->m_unscaled_size.x * ds->m_scale, new_height, GSTexture::Format::DepthStencil, true))
+						{
+							GL_CACHE("HW: RT in RT Z copy on draw %d z_vert_offset %d z_offset %d", s_n, z_vertical_offset, vertical_offset);
+							g_gs_device->StretchRect(ds->m_texture, GSVector4(0.0f, z_vertical_offset / static_cast<float>(ds->m_unscaled_size.y), 1.0f, (ds->m_unscaled_size.y - z_vertical_offset) / static_cast<float>(ds->m_unscaled_size.y)), tex, GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
+							g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+							g_texture_cache->SetTemporaryZ(tex);
+							g_texture_cache->SetTemporaryZInfo(ds->m_TEX0.TBP0, vertical_offset - z_vertical_offset);
+							t_size.y = std::max(new_height, t_size.y);
+						}
+						else
+						{
+							DevCon.Warning("HW: Temporary depth buffer creation failed.");
+							m_using_temp_z = false;
+						}
 					}
-					m_using_temp_z = true;
-
 				}
 			}
 			// Don't resize if the BPP don't match.
@@ -3985,13 +3991,16 @@ void GSRendererHW::Draw()
 
 				if (z_width != new_w || z_height != new_h)
 				{
-					GSVector4i dRect = GSVector4i(0, 0, g_texture_cache->GetTemporaryZ()->GetWidth(), g_texture_cache->GetTemporaryZ()->GetHeight());
-
-					GSTexture* tex = g_gs_device->CreateDepthStencil(new_w * ds->m_scale, new_h * ds->m_scale, GSTexture::Format::DepthStencil, true);
-					g_gs_device->StretchRect(g_texture_cache->GetTemporaryZ(), GSVector4(0.0f, 0.0f, 1.0f, 1.0f), tex, GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
-					g_perfmon.Put(GSPerfMon::TextureCopies, 1);
-					g_texture_cache->InvalidateTemporaryZ();
-					g_texture_cache->SetTemporaryZ(tex);
+					if (GSTexture* tex = g_gs_device->CreateDepthStencil(new_w * ds->m_scale, new_h * ds->m_scale, GSTexture::Format::DepthStencil, true))
+					{
+						const GSVector4i dRect = GSVector4i(0, 0, g_texture_cache->GetTemporaryZ()->GetWidth(), g_texture_cache->GetTemporaryZ()->GetHeight());
+						g_gs_device->StretchRect(g_texture_cache->GetTemporaryZ(), GSVector4(0.0f, 0.0f, 1.0f, 1.0f), tex, GSVector4(dRect), ShaderConvert::DEPTH_COPY, false);
+						g_perfmon.Put(GSPerfMon::TextureCopies, 1);
+						g_texture_cache->InvalidateTemporaryZ();
+						g_texture_cache->SetTemporaryZ(tex);
+					}
+					else
+						DevCon.Warning("HW: Temporary depth buffer creation failed.");
 				}
 			}
 			if (!m_texture_shuffle && !m_channel_shuffle)
@@ -5929,9 +5938,6 @@ __ri void GSRendererHW::EmulateTextureSampler(const GSTextureCache::Target* rt,
 
 	m_conf.cb_ps.ScaleFactor = GSVector4(scale_factor * (1.0f / 16.0f), 1.0f / scale_factor, scale_rt, 0.0f);
 
-	if ((m_conf.ps.tex_is_fb && rt->m_rt_alpha_scale) || (tex->m_target && tex->m_from_target && tex->m_target_direct && tex->m_from_target->m_rt_alpha_scale))
-		m_conf.ps.rta_source_correction = 1;
-
 	// Warning fetch the texture PSM format rather than the context format. The latter could have been corrected in the texture cache for depth.
 	//const GSLocalMemory::psm_t &psm = GSLocalMemory::m_psm[m_cached_ctx.TEX0.PSM];
 	const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[tex->m_TEX0.PSM];
@@ -6638,7 +6644,7 @@ bool GSRendererHW::CanUseTexIsFB(const GSTextureCache::Target* rt, const GSTextu
 		const GSLocalMemory::psm_t& rt_psm = GSLocalMemory::m_psm[rt->m_TEX0.PSM];
 		if (tex_psm.pal > 0 && tex_psm.bpp < rt_psm.bpp)
 		{
-			GL_CACHE("HW: Disabling tex-is-fb due to palette conversion.");
+			GL_CACHE("HW: Enabling tex-is-fb for palette conversion.");
 			return true;
 		}
 
@@ -7123,6 +7129,19 @@ __ri void GSRendererHW::DrawPrims(GSTextureCache::Target* rt, GSTextureCache::Ta
 		new_scale_rt_alpha = rt->m_rt_alpha_scale;
 	}
 
+	GSDevice::RecycledTexture tex_copy;
+	if (tex)
+	{
+		EmulateTextureSampler(rt, ds, tex, tmm, tex_copy);
+	}
+	else
+	{
+		const float scale_factor = rt ? rt->GetScale() : ds->GetScale();
+		m_conf.cb_ps.ScaleFactor = GSVector4(scale_factor * (1.0f / 16.0f), 1.0f / scale_factor, scale_factor, 0.0f);
+
+		m_conf.ps.tfx = 4;
+	}
+
 	// AA1: Set alpha source to coverage 128 when there is no alpha blending.
 	m_conf.ps.fixed_one_a = IsCoverageAlpha();
 
@@ -7146,6 +7165,9 @@ __ri void GSRendererHW::DrawPrims(GSTextureCache::Target* rt, GSTextureCache::Ta
 		}
 	}
 
+	if ((m_conf.ps.tex_is_fb && rt && rt->m_rt_alpha_scale) || (tex && tex->m_from_target && tex->m_target_direct && tex->m_from_target->m_rt_alpha_scale))
+		m_conf.ps.rta_source_correction = 1;
+
 	if (req_src_update && tex->m_texture != rt->m_texture)
 		tex->m_texture = rt->m_texture;
 
@@ -7278,8 +7300,25 @@ __ri void GSRendererHW::DrawPrims(GSTextureCache::Target* rt, GSTextureCache::Ta
 	if (m_conf.destination_alpha >= GSHWDrawConfig::DestinationAlphaMode::Stencil &&
 		m_conf.destination_alpha <= GSHWDrawConfig::DestinationAlphaMode::StencilOne && !m_conf.ds)
 	{
-		temp_ds.reset(g_gs_device->CreateDepthStencil(m_conf.rt->GetWidth(), m_conf.rt->GetHeight(), GSTexture::Format::DepthStencil, false));
-		m_conf.ds = temp_ds.get();
+		const bool is_one_barrier = (m_conf.require_full_barrier && (m_prim_overlap == PRIM_OVERLAP_NO || m_conf.ps.shuffle || m_channel_shuffle));
+		if ((temp_ds.reset(g_gs_device->CreateDepthStencil(m_conf.rt->GetWidth(), m_conf.rt->GetHeight(), GSTexture::Format::DepthStencil, false)), temp_ds))
+		{
+			m_conf.ds = temp_ds.get();
+		}
+		else if (features.primitive_id && !(m_conf.ps.scanmsk & 2) && (!m_conf.require_full_barrier || is_one_barrier))
+		{
+			DATE_one = false;
+			DATE_PRIMID = true;
+			m_conf.destination_alpha = GSHWDrawConfig::DestinationAlphaMode::PrimIDTracking;
+			DevCon.Warning("HW: Depth buffer creation failed for Stencil Date. Fallback to PrimIDTracking.");
+		}
+		else
+		{
+			DATE = false;
+			DATE_one = false;
+			m_conf.destination_alpha = GSHWDrawConfig::DestinationAlphaMode::Off;
+			DevCon.Warning("HW: Depth buffer creation failed for Stencil Date.");
+		}
 	}
 
 	// vs
@@ -7412,18 +7451,7 @@ __ri void GSRendererHW::DrawPrims(GSTextureCache::Target* rt, GSTextureCache::Ta
 		m_conf.cb_ps.FogColor_AREF = fc.blend32<8>(m_conf.cb_ps.FogColor_AREF);
 	}
 
-	GSDevice::RecycledTexture tex_copy;
-	if (tex)
-	{
-		EmulateTextureSampler(rt, ds, tex, tmm, tex_copy);
-	}
-	else
-	{
-		const float scale_factor = rt ? rt->GetScale() : ds->GetScale();
-		m_conf.cb_ps.ScaleFactor = GSVector4(scale_factor * (1.0f / 16.0f), 1.0f / scale_factor, scale_factor, 0.0f);
 
-		m_conf.ps.tfx = 4;
-	}
 
 	// Update RT scaled alpha flag, nothing's going to read it anymore.
 	if (rt)