rsx: Fix depth clipping

- Fix special case where n=f making (f-n) = 0
- Dynamically update depth range by setting dirty bits
- Fix depth bounds when n=f and bounds test is disabled

rpcs3/Emu/RSX/Common/GLSLCommon.h

@@ -576,19 +576,34 @@ namespace glsl
 		if (props.domain == glsl::program_domain::glsl_vertex_program && props.emulate_zclip_transform)
 		{
 			OS <<
-			"vec4 apply_zclip_xform(const in vec4 pos, const in float near_plane, const in float far_plane)\n"
+			"double rcp_precise(double x)\n"
 			"{\n"
-			"	float d = pos.z / pos.w;\n";
+			"	double scaled = x * 0.0009765625;\n"
+			"	double inv = 1.0 / scaled;\n"
+			"	return inv * 0.0009765625;\n"
+			"}\n"
+			"\n"
+			"vec4 apply_zclip_xform(const in vec4 pos, const in float near_plane, const in float far_plane)\n"
+			"{\n";
 
 			if (!props.emulate_depth_clip_only)
 			{
 				OS <<
+				"	float d = float(pos.z * rcp_precise(pos.w));\n"
 				"	if (d < 0.f && d >= near_plane)\n"
-				"		d = 0.f;\n" //force clamp negative values
+				"	{\n"
+				"		// Clamp\n"
+				"		d = 0.f;\n"
+				"	}\n"
 				"	else if (d > 1.f && d <= far_plane)\n"
+				"	{\n"
+				"		// Compress Z and store towards highest end of the range\n"
 				"		d = min(1., 0.99 + (0.01 * (pos.z - near_plane) / (far_plane - near_plane)));\n"
+				"	}\n"
 				"	else\n"
-				"		return pos; //d = (0.99 * d);\n" //range compression for normal values is disabled until a solution to ops comparing z is found
+				"	{\n"
+				"		return pos;\n"
+				"	}\n"
 				"\n"
 				"	return vec4(pos.x, pos.y, d * pos.w, pos.w);\n";
 			}
@@ -597,10 +612,18 @@ namespace glsl
 				// Technically the depth value here is the 'final' depth that should be stored in the Z buffer.
 				// Forward mapping eqn is d' = d * (f - n) + n, where d' is the stored Z value (this) and d is the normalized API value.
 				OS <<
-				"	double inv_range = double(1.0) / double(far_plane - near_plane);\n"
+				"	if (far_plane != 0.0)\n"
-				"	double new_d = (double(d) - double(near_plane)) * inv_range;\n"
+				"	{\n"
-				"\n"
+				"		double z_range = (far_plane > near_plane)? (far_plane - near_plane) : far_plane;\n"
-				"	return vec4(pos.x, pos.y, float(new_d * pos.w), pos.w);\n";
+				"		double inv_range = rcp_precise(z_range);\n"
+				"		float d = float(pos.z * rcp_precise(pos.w));\n"
+				"		float new_d = (d - near_plane) * float(inv_range);\n"
+				"		return vec4(pos.x, pos.y, (new_d * pos.w), pos.w);\n"
+				"	}\n"
+				"	else\n"
+				"	{\n"
+				"		return pos;\n" // Only values where Z=0 can ever pass this clip
+				"	}\n";
 			}
 
 			OS <<

rpcs3/Emu/RSX/GL/GLDraw.cpp

@@ -257,15 +257,15 @@ void GLGSRender::update_draw_state()
 	//NV4097_SET_TWO_SIDE_LIGHT_EN
 	//NV4097_SET_FLAT_SHADE_OP
 	//NV4097_SET_EDGE_FLAG
-
-
-
 	//NV4097_SET_COLOR_KEY_COLOR
 	//NV4097_SET_SHADER_CONTROL
 	//NV4097_SET_ZMIN_MAX_CONTROL
 	//NV4097_SET_ANTI_ALIASING_CONTROL
 	//NV4097_SET_CLIP_ID_TEST_ENABLE
 
+	// For OGL Z range is updated every draw as it is separate from viewport config
+	m_graphics_state &= ~(rsx::pipeline_state::zclip_config_state_dirty);
+
 	m_frame_stats.setup_time += m_profiler.duration();
 }
 

rpcs3/Emu/RSX/RSXThread.h

@@ -78,14 +78,16 @@ namespace rsx
 		fragment_texture_state_dirty = 0x80, // Fragment texture parameters changed
 		vertex_texture_state_dirty = 0x100,  // Fragment texture parameters changed
 		scissor_config_state_dirty = 0x200,  // Scissor region changed
+		zclip_config_state_dirty = 0x400,    // Viewport Z clip changed
 
-		scissor_setup_invalid = 0x400,       // Scissor configuration is broken
+		scissor_setup_invalid = 0x800,       // Scissor configuration is broken
-		scissor_setup_clipped = 0x800,       // Scissor region is cropped by viewport constraint
+		scissor_setup_clipped = 0x1000,      // Scissor region is cropped by viewport constraint
 
-		polygon_stipple_pattern_dirty = 0x1000,  // Rasterizer stippling pattern changed
+		polygon_stipple_pattern_dirty = 0x2000,  // Rasterizer stippling pattern changed
-		line_stipple_pattern_dirty = 0x2000,     // Line stippling pattern changed
+		line_stipple_pattern_dirty = 0x4000,     // Line stippling pattern changed
 
 		invalidate_pipeline_bits = fragment_program_dirty | vertex_program_dirty,
+		invalidate_zclip_bits = vertex_state_dirty | zclip_config_state_dirty,
 		memory_barrier_bits = framebuffer_reads_dirty,
 		all_dirty = ~0u
 	};

rpcs3/Emu/RSX/VK/VKDraw.cpp

@@ -117,8 +117,9 @@ void VKGSRender::update_draw_state()
 		}
 		else
 		{
-			bounds_min = rsx::method_registers.clip_min();
+			// Avoid special case where min=max and depth bounds (incorrectly) fails
-			bounds_max = rsx::method_registers.clip_max();
+			bounds_min = std::min(0.f, rsx::method_registers.clip_min());
+			bounds_max = std::max(1.f, rsx::method_registers.clip_max());
 		}
 
 		if (!m_device->get_unrestricted_depth_range_support())

rpcs3/Emu/RSX/VK/VKGSRender.cpp

@@ -974,6 +974,8 @@ void VKGSRender::set_viewport()
 		m_viewport.minDepth = 0.f;
 		m_viewport.maxDepth = 1.f;
 	}
+
+	m_graphics_state &= ~(rsx::pipeline_state::zclip_config_state_dirty);
 }
 
 void VKGSRender::set_scissor(bool clip_viewport)
@@ -990,6 +992,17 @@ void VKGSRender::set_scissor(bool clip_viewport)
 
 void VKGSRender::bind_viewport()
 {
+	if (m_graphics_state & rsx::pipeline_state::zclip_config_state_dirty)
+	{
+		if (m_device->get_unrestricted_depth_range_support())
+		{
+			m_viewport.minDepth = rsx::method_registers.clip_min();
+			m_viewport.maxDepth = rsx::method_registers.clip_max();
+		}
+
+		m_graphics_state &= ~(rsx::pipeline_state::zclip_config_state_dirty);
+	}
+
 	vkCmdSetViewport(*m_current_command_buffer, 0, 1, &m_viewport);
 	vkCmdSetScissor(*m_current_command_buffer, 0, 1, &m_scissor);
 }

rpcs3/Emu/RSX/rsx_methods.cpp

@@ -887,18 +887,6 @@ namespace rsx
 				}
 			}
 		};
-
-		template<u32 index>
-		struct set_viewport_dirty_bit
-		{
-			static void impl(thread* rsx, u32 _reg, u32 arg)
-			{
-				if (arg != method_registers.register_previous_value)
-				{
-					rsx->m_graphics_state |= rsx::pipeline_state::vertex_state_dirty;
-				}
-			}
-		};
 	}
 
 	namespace nv308a
@@ -3169,8 +3157,8 @@ namespace rsx
 		bind<NV4097_SET_VERTEX_DATA_BASE_INDEX, nv4097::set_index_base_offset>();
 		bind<NV4097_SET_USER_CLIP_PLANE_CONTROL, nv4097::notify_state_changed<vertex_state_dirty>>();
 		bind<NV4097_SET_TRANSFORM_BRANCH_BITS, nv4097::notify_state_changed<vertex_state_dirty>>();
-		bind<NV4097_SET_CLIP_MIN, nv4097::notify_state_changed<vertex_state_dirty>>();
+		bind<NV4097_SET_CLIP_MIN, nv4097::notify_state_changed<invalidate_zclip_bits>>();
-		bind<NV4097_SET_CLIP_MAX, nv4097::notify_state_changed<vertex_state_dirty>>();
+		bind<NV4097_SET_CLIP_MAX, nv4097::notify_state_changed<invalidate_zclip_bits>>();
 		bind<NV4097_SET_POINT_SIZE, nv4097::notify_state_changed<vertex_state_dirty>>();
 		bind<NV4097_SET_ALPHA_FUNC, nv4097::notify_state_changed<fragment_state_dirty>>();
 		bind<NV4097_SET_ALPHA_REF, nv4097::notify_state_changed<fragment_state_dirty>>();
@@ -3184,8 +3172,8 @@ namespace rsx
 		bind<NV4097_SET_VIEWPORT_HORIZONTAL, nv4097::notify_state_changed<scissor_config_state_dirty>>();
 		bind<NV4097_SET_VIEWPORT_VERTICAL, nv4097::notify_state_changed<scissor_config_state_dirty>>();
 		bind_array<NV4097_SET_FOG_PARAMS, 1, 2, nv4097::notify_state_changed<fragment_state_dirty>>();
-		bind_range<NV4097_SET_VIEWPORT_SCALE, 1, 3, nv4097::set_viewport_dirty_bit>();
+		bind_array<NV4097_SET_VIEWPORT_SCALE, 1, 3, nv4097::notify_state_changed<vertex_state_dirty>>();
-		bind_range<NV4097_SET_VIEWPORT_OFFSET, 1, 3, nv4097::set_viewport_dirty_bit>();
+		bind_array<NV4097_SET_VIEWPORT_OFFSET, 1, 3, nv4097::notify_state_changed<vertex_state_dirty>>();
 		bind<NV4097_SET_INDEX_ARRAY_DMA, nv4097::check_index_array_dma>();
 		bind<NV4097_SET_BLEND_EQUATION, nv4097::set_blend_equation>();
 		bind<NV4097_SET_BLEND_FUNC_SFACTOR, nv4097::set_blend_factor>();