LibGL+LibSoftGPU: Implement the stencil buffer

This implements an 8-bit front stencil buffer. Stencil operations are
SIMD optimized. LibGL changes include:

* New `glStencilMask` and `glStencilMaskSeparate` functions
* New context parameter `GL_STENCIL_CLEAR_VALUE`

12 files changed, 420 insertions, 78 deletions

diff --git a/Userland/Libraries/LibGL/GL/gl.h b/Userland/Libraries/LibGL/GL/gl.h
index 60bd67261d..af297e3740 100644
--- a/Userland/Libraries/LibGL/GL/gl.h
+++ b/Userland/Libraries/LibGL/GL/gl.h
@@ -98,6 +98,7 @@ extern "C" {
 #define GL_COLOR_MATERIAL 0x0B57
 #define GL_FOG_START 0x0B63
 #define GL_FOG_END 0x0B64
+#define GL_STENCIL_CLEAR_VALUE 0x0B91
 #define GL_MATRIX_MODE 0x0BA0
 #define GL_NORMALIZE 0x0BA1
 #define GL_VIEWPORT 0x0BA2
@@ -605,9 +606,10 @@ GLAPI void glLightModelfv(GLenum pname, GLfloat const* params);
 GLAPI void glLightModeli(GLenum pname, GLint param);
 GLAPI void glStencilFunc(GLenum func, GLint ref, GLuint mask);
 GLAPI void glStencilFuncSeparate(GLenum face, GLenum func, GLint ref, GLuint mask);
+GLAPI void glStencilMask(GLuint mask);
+GLAPI void glStencilMaskSeparate(GLenum face, GLuint mask);
 GLAPI void glStencilOp(GLenum sfail, GLenum dpfail, GLenum dppass);
 GLAPI void glStencilOpSeparate(GLenum face, GLenum sfail, GLenum dpfail, GLenum dppass);
-GLAPI void glStencilMask(GLuint mask);
 GLAPI void glNormal3f(GLfloat nx, GLfloat ny, GLfloat nz);
 GLAPI void glNormal3fv(GLfloat const* v);
 GLAPI void glNormalPointer(GLenum type, GLsizei stride, void const* pointer);
diff --git a/Userland/Libraries/LibGL/GLContext.h b/Userland/Libraries/LibGL/GLContext.h
index bf189197d9..1eecd49830 100644
--- a/Userland/Libraries/LibGL/GLContext.h
+++ b/Userland/Libraries/LibGL/GLContext.h
@@ -97,6 +97,7 @@ public:
     virtual void gl_pixel_storei(GLenum pname, GLint param) = 0;
     virtual void gl_scissor(GLint x, GLint y, GLsizei width, GLsizei height) = 0;
     virtual void gl_stencil_func_separate(GLenum face, GLenum func, GLint ref, GLuint mask) = 0;
+    virtual void gl_stencil_mask_separate(GLenum face, GLuint mask) = 0;
     virtual void gl_stencil_op_separate(GLenum face, GLenum sfail, GLenum dpfail, GLenum dppass) = 0;
     virtual void gl_normal(GLfloat nx, GLfloat ny, GLfloat nz) = 0;
     virtual void gl_normal_pointer(GLenum type, GLsizei stride, void const* pointer) = 0;
diff --git a/Userland/Libraries/LibGL/GLStencil.cpp b/Userland/Libraries/LibGL/GLStencil.cpp
index 80ded3ef5d..7e2916ef08 100644
--- a/Userland/Libraries/LibGL/GLStencil.cpp
+++ b/Userland/Libraries/LibGL/GLStencil.cpp
@@ -36,5 +36,10 @@ void glStencilOpSeparate(GLenum face, GLenum sfail, GLenum dpfail, GLenum dppass
 
 void glStencilMask(GLuint mask)
 {
-    dbgln("(STUBBED) glStencilMask(0x{:08x})", mask);
+    g_gl_context->gl_stencil_mask_separate(GL_FRONT_AND_BACK, mask);
+}
+
+void glStencilMaskSeparate(GLenum face, GLuint mask)
+{
+    g_gl_context->gl_stencil_mask_separate(face, mask);
 }
diff --git a/Userland/Libraries/LibGL/SoftwareGLContext.cpp b/Userland/Libraries/LibGL/SoftwareGLContext.cpp
index 225ddc016c..0ca7b2f897 100644
--- a/Userland/Libraries/LibGL/SoftwareGLContext.cpp
+++ b/Userland/Libraries/LibGL/SoftwareGLContext.cpp
@@ -165,7 +165,9 @@ Optional<ContextParameter> SoftwareGLContext::get_context_parameter(GLenum name)
         return ContextParameter { .type = GL_BOOL, .is_capability = true, .value = { .boolean_value = scissor_enabled } };
     }
     case GL_STENCIL_BITS:
-        return ContextParameter { .type = GL_INT, .value = { .integer_value = sizeof(float) * 8 } };
+        return ContextParameter { .type = GL_INT, .value = { .integer_value = m_device_info.stencil_bits } };
+    case GL_STENCIL_CLEAR_VALUE:
+        return ContextParameter { .type = GL_INT, .value = { .integer_value = m_clear_stencil } };
     case GL_STENCIL_TEST:
         return ContextParameter { .type = GL_BOOL, .is_capability = true, .value = { .boolean_value = m_stencil_test_enabled } };
     case GL_TEXTURE_1D:
@@ -239,9 +241,8 @@ void SoftwareGLContext::gl_clear(GLbitfield mask)
     if (mask & GL_DEPTH_BUFFER_BIT)
         m_rasterizer.clear_depth(static_cast<float>(m_clear_depth));
 
-    // FIXME: implement GL_STENCIL_BUFFER_BIT
     if (mask & GL_STENCIL_BUFFER_BIT)
-        dbgln_if(GL_DEBUG, "gl_clear(): GL_STENCIL_BUFFER_BIT is unimplemented");
+        m_rasterizer.clear_stencil(m_clear_stencil);
 }
 
 void SoftwareGLContext::gl_clear_color(GLclampf red, GLclampf green, GLclampf blue, GLclampf alpha)
@@ -268,9 +269,7 @@ void SoftwareGLContext::gl_clear_stencil(GLint s)
 
     RETURN_WITH_ERROR_IF(m_in_draw_state, GL_INVALID_OPERATION);
 
-    // FIXME: "s is masked with 2^m - 1 , where m is the number of bits in the stencil buffer"
-
-    m_clear_stencil = s;
+    m_clear_stencil = static_cast<u8>(s & ((1 << m_device_info.stencil_bits) - 1));
 }
 
 void SoftwareGLContext::gl_color(GLdouble r, GLdouble g, GLdouble b, GLdouble a)
@@ -697,6 +696,8 @@ void SoftwareGLContext::gl_enable(GLenum capability)
         break;
     case GL_STENCIL_TEST:
         m_stencil_test_enabled = true;
+        rasterizer_options.enable_stencil_test = true;
+        update_rasterizer_options = true;
         break;
     case GL_TEXTURE_1D:
         m_active_texture_unit->set_texture_1d_enabled(true);
@@ -808,6 +809,8 @@ void SoftwareGLContext::gl_disable(GLenum capability)
         break;
     case GL_STENCIL_TEST:
         m_stencil_test_enabled = false;
+        rasterizer_options.enable_stencil_test = false;
+        update_rasterizer_options = true;
         break;
     case GL_TEXTURE_1D:
         m_active_texture_unit->set_texture_1d_enabled(false);
@@ -2631,13 +2634,28 @@ void SoftwareGLContext::gl_stencil_func_separate(GLenum face, GLenum func, GLint
                              || func == GL_ALWAYS),
         GL_INVALID_ENUM);
 
-    // FIXME: "ref is clamped to the range 02^n - 1 , where n is the number of bitplanes in the stencil buffer"
+    ref = clamp(ref, 0, (1 << m_device_info.stencil_bits) - 1);
 
     StencilFunctionOptions new_options = { func, ref, mask };
     if (face == GL_FRONT || face == GL_FRONT_AND_BACK)
         m_stencil_function[Face::Front] = new_options;
     if (face == GL_BACK || face == GL_FRONT_AND_BACK)
         m_stencil_function[Face::Back] = new_options;
+
+    m_stencil_configuration_dirty = true;
+}
+
+void SoftwareGLContext::gl_stencil_mask_separate(GLenum face, GLuint mask)
+{
+    APPEND_TO_CALL_LIST_AND_RETURN_IF_NEEDED(gl_stencil_mask_separate, face, mask);
+    RETURN_WITH_ERROR_IF(m_in_draw_state, GL_INVALID_OPERATION);
+
+    if (face == GL_FRONT || face == GL_FRONT_AND_BACK)
+        m_stencil_operation[Face::Front].write_mask = mask;
+    if (face == GL_BACK || face == GL_FRONT_AND_BACK)
+        m_stencil_operation[Face::Back].write_mask = mask;
+
+    m_stencil_configuration_dirty = true;
 }
 
 void SoftwareGLContext::gl_stencil_op_separate(GLenum face, GLenum sfail, GLenum dpfail, GLenum dppass)
@@ -2647,39 +2665,26 @@ void SoftwareGLContext::gl_stencil_op_separate(GLenum face, GLenum sfail, GLenum
 
     RETURN_WITH_ERROR_IF(!(face == GL_FRONT || face == GL_BACK || face == GL_FRONT_AND_BACK), GL_INVALID_ENUM);
 
-    RETURN_WITH_ERROR_IF(!(sfail == GL_KEEP
-                             || sfail == GL_ZERO
-                             || sfail == GL_REPLACE
-                             || sfail == GL_INCR
-                             || sfail == GL_INCR_WRAP
-                             || sfail == GL_DECR
-                             || sfail == GL_DECR_WRAP
-                             || sfail == GL_INVERT),
-        GL_INVALID_ENUM);
-    RETURN_WITH_ERROR_IF(!(dpfail == GL_KEEP
-                             || dpfail == GL_ZERO
-                             || dpfail == GL_REPLACE
-                             || dpfail == GL_INCR
-                             || dpfail == GL_INCR_WRAP
-                             || dpfail == GL_DECR
-                             || dpfail == GL_DECR_WRAP
-                             || dpfail == GL_INVERT),
-        GL_INVALID_ENUM);
-    RETURN_WITH_ERROR_IF(!(dppass == GL_KEEP
-                             || dppass == GL_ZERO
-                             || dppass == GL_REPLACE
-                             || dppass == GL_INCR
-                             || dppass == GL_INCR_WRAP
-                             || dppass == GL_DECR
-                             || dppass == GL_DECR_WRAP
-                             || dppass == GL_INVERT),
-        GL_INVALID_ENUM);
-
-    StencilOperationOptions new_options = { sfail, dpfail, dppass };
+    auto is_valid_op = [](GLenum op) -> bool {
+        return op == GL_KEEP || op == GL_ZERO || op == GL_REPLACE || op == GL_INCR || op == GL_INCR_WRAP
+            || op == GL_DECR || op == GL_DECR_WRAP || op == GL_INVERT;
+    };
+    RETURN_WITH_ERROR_IF(!is_valid_op(sfail), GL_INVALID_ENUM);
+    RETURN_WITH_ERROR_IF(!is_valid_op(dpfail), GL_INVALID_ENUM);
+    RETURN_WITH_ERROR_IF(!is_valid_op(dppass), GL_INVALID_ENUM);
+
+    auto update_stencil_operation = [&](Face face, GLenum sfail, GLenum dpfail, GLenum dppass) {
+        auto& stencil_operation = m_stencil_operation[face];
+        stencil_operation.op_fail = sfail;
+        stencil_operation.op_depth_fail = dpfail;
+        stencil_operation.op_pass = dppass;
+    };
     if (face == GL_FRONT || face == GL_FRONT_AND_BACK)
-        m_stencil_operation[Face::Front] = new_options;
+        update_stencil_operation(Face::Front, sfail, dpfail, dppass);
     if (face == GL_BACK || face == GL_FRONT_AND_BACK)
-        m_stencil_operation[Face::Back] = new_options;
+        update_stencil_operation(Face::Back, sfail, dpfail, dppass);
+
+    m_stencil_configuration_dirty = true;
 }
 
 void SoftwareGLContext::gl_normal(GLfloat nx, GLfloat ny, GLfloat nz)
@@ -2908,6 +2913,7 @@ void SoftwareGLContext::sync_device_config()
     sync_device_sampler_config();
     sync_device_texcoord_config();
     sync_light_state();
+    sync_stencil_configuration();
 }
 
 void SoftwareGLContext::sync_device_sampler_config()
@@ -3170,6 +3176,74 @@ void SoftwareGLContext::sync_device_texcoord_config()
     m_rasterizer.set_options(options);
 }
 
+void SoftwareGLContext::sync_stencil_configuration()
+{
+    if (!m_stencil_configuration_dirty)
+        return;
+    m_stencil_configuration_dirty = false;
+
+    auto set_device_stencil = [&](SoftGPU::Face face, StencilFunctionOptions func, StencilOperationOptions op) {
+        SoftGPU::StencilConfiguration device_configuration;
+
+        // Stencil test function
+        auto map_func = [](GLenum func) -> SoftGPU::StencilTestFunction {
+            switch (func) {
+            case GL_ALWAYS:
+                return SoftGPU::StencilTestFunction::Always;
+            case GL_EQUAL:
+                return SoftGPU::StencilTestFunction::Equal;
+            case GL_GEQUAL:
+                return SoftGPU::StencilTestFunction::GreaterOrEqual;
+            case GL_GREATER:
+                return SoftGPU::StencilTestFunction::Greater;
+            case GL_LESS:
+                return SoftGPU::StencilTestFunction::Less;
+            case GL_LEQUAL:
+                return SoftGPU::StencilTestFunction::LessOrEqual;
+            case GL_NEVER:
+                return SoftGPU::StencilTestFunction::Never;
+            case GL_NOTEQUAL:
+                return SoftGPU::StencilTestFunction::NotEqual;
+            }
+            VERIFY_NOT_REACHED();
+        };
+        device_configuration.test_function = map_func(func.func);
+        device_configuration.reference_value = func.reference_value;
+        device_configuration.test_mask = func.mask;
+
+        // Stencil operation
+        auto map_operation = [](GLenum operation) -> SoftGPU::StencilOperation {
+            switch (operation) {
+            case GL_DECR:
+                return SoftGPU::StencilOperation::Decrement;
+            case GL_DECR_WRAP:
+                return SoftGPU::StencilOperation::DecrementWrap;
+            case GL_INCR:
+                return SoftGPU::StencilOperation::Increment;
+            case GL_INCR_WRAP:
+                return SoftGPU::StencilOperation::IncrementWrap;
+            case GL_INVERT:
+                return SoftGPU::StencilOperation::Invert;
+            case GL_KEEP:
+                return SoftGPU::StencilOperation::Keep;
+            case GL_REPLACE:
+                return SoftGPU::StencilOperation::Replace;
+            case GL_ZERO:
+                return SoftGPU::StencilOperation::Zero;
+            }
+            VERIFY_NOT_REACHED();
+        };
+        device_configuration.on_stencil_test_fail = map_operation(op.op_fail);
+        device_configuration.on_depth_test_fail = map_operation(op.op_depth_fail);
+        device_configuration.on_pass = map_operation(op.op_pass);
+        device_configuration.write_mask = op.write_mask;
+
+        m_rasterizer.set_stencil_configuration(face, device_configuration);
+    };
+    set_device_stencil(SoftGPU::Face::Front, m_stencil_function[Face::Front], m_stencil_operation[Face::Front]);
+    set_device_stencil(SoftGPU::Face::Back, m_stencil_function[Face::Back], m_stencil_operation[Face::Back]);
+}
+
 void SoftwareGLContext::gl_lightf(GLenum light, GLenum pname, GLfloat param)
 {
     APPEND_TO_CALL_LIST_AND_RETURN_IF_NEEDED(gl_lightf, light, pname, param);
diff --git a/Userland/Libraries/LibGL/SoftwareGLContext.h b/Userland/Libraries/LibGL/SoftwareGLContext.h
index 5494642944..ff5ea056be 100644
--- a/Userland/Libraries/LibGL/SoftwareGLContext.h
+++ b/Userland/Libraries/LibGL/SoftwareGLContext.h
@@ -128,6 +128,7 @@ public:
     virtual void gl_pixel_storei(GLenum pname, GLint param) override;
     virtual void gl_scissor(GLint x, GLint y, GLsizei width, GLsizei height) override;
     virtual void gl_stencil_func_separate(GLenum face, GLenum func, GLint ref, GLuint mask) override;
+    virtual void gl_stencil_mask_separate(GLenum face, GLuint mask) override;
     virtual void gl_stencil_op_separate(GLenum face, GLenum sfail, GLenum dpfail, GLenum dppass) override;
     virtual void gl_normal(GLfloat nx, GLfloat ny, GLfloat nz) override;
     virtual void gl_normal_pointer(GLenum type, GLsizei stride, void const* pointer) override;
@@ -154,6 +155,7 @@ private:
     void sync_device_sampler_config();
     void sync_device_texcoord_config();
     void sync_light_state();
+    void sync_stencil_configuration();
 
     template<typename T>
     T* store_in_listing(T value)
@@ -195,7 +197,7 @@ private:
 
     FloatVector4 m_clear_color { 0.0f, 0.0f, 0.0f, 0.0f };
     double m_clear_depth { 1.0 };
-    GLint m_clear_stencil { 0 };
+    u8 m_clear_stencil { 0 };
 
     FloatVector4 m_current_vertex_color = { 1.0f, 1.0f, 1.0f, 1.0f };
     FloatVector4 m_current_vertex_tex_coord = { 0.0f, 0.0f, 0.0f, 1.0f };
@@ -225,6 +227,7 @@ private:
 
     // Stencil configuration
     bool m_stencil_test_enabled { false };
+    bool m_stencil_configuration_dirty { true };
 
     struct StencilFunctionOptions {
         GLenum func { GL_ALWAYS };
@@ -237,6 +240,7 @@ private:
         GLenum op_fail { GL_KEEP };
         GLenum op_depth_fail { GL_KEEP };
         GLenum op_pass { GL_KEEP };
+        GLuint write_mask { NumericLimits<GLuint>::max() };
     };
     Array<StencilOperationOptions, 2u> m_stencil_operation;
 
@@ -360,6 +364,7 @@ private:
             decltype(&SoftwareGLContext::gl_polygon_offset),
             decltype(&SoftwareGLContext::gl_scissor),
             decltype(&SoftwareGLContext::gl_stencil_func_separate),
+            decltype(&SoftwareGLContext::gl_stencil_mask_separate),
             decltype(&SoftwareGLContext::gl_stencil_op_separate),
             decltype(&SoftwareGLContext::gl_normal),
             decltype(&SoftwareGLContext::gl_raster_pos),
diff --git a/Userland/Libraries/LibSoftGPU/CMakeLists.txt b/Userland/Libraries/LibSoftGPU/CMakeLists.txt
index 96c04d4d42..03ee632187 100644
--- a/Userland/Libraries/LibSoftGPU/CMakeLists.txt
+++ b/Userland/Libraries/LibSoftGPU/CMakeLists.txt
@@ -4,6 +4,7 @@ set(SOURCES
     Device.cpp
     Image.cpp
     Sampler.cpp
+    StencilBuffer.cpp
 )
 
 add_compile_options(-Wno-psabi)
diff --git a/Userland/Libraries/LibSoftGPU/Device.cpp b/Userland/Libraries/LibSoftGPU/Device.cpp
index b2f3c1bd66..fa4509b977 100644
--- a/Userland/Libraries/LibSoftGPU/Device.cpp
+++ b/Userland/Libraries/LibSoftGPU/Device.cpp
@@ -1,12 +1,14 @@
 /*
  * Copyright (c) 2021, Stephan Unverwerth <s.unverwerth@serenityos.org>
  * Copyright (c) 2021, Jesse Buhagiar <jooster669@gmail.com>
+ * Copyright (c) 2022, Jelle Raaijmakers <jelle@gmta.nl>
  *
  * SPDX-License-Identifier: BSD-2-Clause
  */
 
 #include <AK/Function.h>
 #include <AK/Math.h>
+#include <AK/NumericLimits.h>
 #include <AK/SIMDExtras.h>
 #include <AK/SIMDMath.h>
 #include <LibCore/ElapsedTimer.h>
@@ -82,11 +84,11 @@ static Vector4<f32x4> to_vec4(u32x4 rgba)
     };
 }
 
-static Gfx::IntRect window_coordinates_to_target_coordinates(Gfx::IntRect const window_rect, Gfx::IntRect const target_rect)
+Gfx::IntRect Device::window_coordinates_to_target_coordinates(Gfx::IntRect const& window_rect)
 {
     return {
         window_rect.x(),
-        target_rect.height() - window_rect.height() - window_rect.y(),
+        m_render_target->rect().height() - window_rect.height() - window_rect.y(),
         window_rect.width(),
         window_rect.height(),
     };
@@ -213,7 +215,7 @@ void Device::rasterize_triangle(const Triangle& triangle)
 
     auto render_bounds = m_render_target->rect();
     if (m_options.scissor_enabled)
-        render_bounds.intersect(window_coordinates_to_target_coordinates(m_options.scissor_box, m_render_target->rect()));
+        render_bounds.intersect(window_coordinates_to_target_coordinates(m_options.scissor_box));
 
     // Obey top-left rule:
     // This sets up "zero" for later pixel coverage tests.
@@ -229,7 +231,7 @@ void Device::rasterize_triangle(const Triangle& triangle)
         zero.set_y(0);
 
     // This function calculates the 3 edge values for the pixel relative to the triangle.
-    auto calculate_edge_values4 = [v0, v1, v2](const Vector2<i32x4>& p) -> Vector3<i32x4> {
+    auto calculate_edge_values4 = [v0, v1, v2](Vector2<i32x4> const& p) -> Vector3<i32x4> {
         return {
             edge_function4(v1, v2, p),
             edge_function4(v2, v0, p),
@@ -238,7 +240,7 @@ void Device::rasterize_triangle(const Triangle& triangle)
     };
 
     // This function tests whether a point as identified by its 3 edge values lies within the triangle
-    auto test_point4 = [zero](const Vector3<i32x4>& edges) -> i32x4 {
+    auto test_point4 = [zero](Vector3<i32x4> const& edges) -> i32x4 {
         return edges.x() >= zero.x()
             && edges.y() >= zero.y()
             && edges.z() >= zero.z();
@@ -257,8 +259,6 @@ void Device::rasterize_triangle(const Triangle& triangle)
     float const vertex1_eye_absz = fabs(vertex1.eye_coordinates.z());
     float const vertex2_eye_absz = fabs(vertex2.eye_coordinates.z());
 
-    // FIXME: implement stencil testing
-
     int const render_bounds_left = render_bounds.x();
     int const render_bounds_right = render_bounds.x() + render_bounds.width();
     int const render_bounds_top = render_bounds.y();
@@ -269,10 +269,46 @@ void Device::rasterize_triangle(const Triangle& triangle)
         expand4(subpixel_factor / 2),
     };
 
+    // Stencil configuration and writing
+    auto const stencil_configuration = m_stencil_configuration[Face::Front];
+    auto const stencil_reference_value = stencil_configuration.reference_value & stencil_configuration.test_mask;
+
+    auto write_to_stencil = [](u8* stencil_ptrs[4], i32x4 stencil_value, StencilOperation op, u8 reference_value, u8 write_mask, i32x4 pixel_mask) {
+        if (write_mask == 0 || op == StencilOperation::Keep)
+            return;
+
+        switch (op) {
+        case StencilOperation::Decrement:
+            stencil_value = (stencil_value & ~write_mask) | (max(stencil_value - 1, expand4(0)) & write_mask);
+            break;
+        case StencilOperation::DecrementWrap:
+            stencil_value = (stencil_value & ~write_mask) | (((stencil_value - 1) & 0xFF) & write_mask);
+            break;
+        case StencilOperation::Increment:
+            stencil_value = (stencil_value & ~write_mask) | (min(stencil_value + 1, expand4(0xFF)) & write_mask);
+            break;
+        case StencilOperation::IncrementWrap:
+            stencil_value = (stencil_value & ~write_mask) | (((stencil_value + 1) & 0xFF) & write_mask);
+            break;
+        case StencilOperation::Invert:
+            stencil_value ^= write_mask;
+            break;
+        case StencilOperation::Replace:
+            stencil_value = (stencil_value & ~write_mask) | (reference_value & write_mask);
+            break;
+        case StencilOperation::Zero:
+            stencil_value &= ~write_mask;
+            break;
+        default:
+            VERIFY_NOT_REACHED();
+        }
+
+        store4_masked(stencil_value, stencil_ptrs[0], stencil_ptrs[1], stencil_ptrs[2], stencil_ptrs[3], pixel_mask);
+    };
+
     // Iterate over all blocks within the bounds of the triangle
     for (int by = by0; by < by1; by += 2) {
         for (int bx = bx0; bx < bx1; bx += 2) {
-
             PixelQuad quad;
 
             quad.screen_coordinates = {
@@ -306,14 +342,70 @@ void Device::rasterize_triangle(const Triangle& triangle)
 
             int coverage_bits = maskbits(quad.mask);
 
+            // Stencil testing
+            u8* stencil_ptrs[4];
+            i32x4 stencil_value;
+            if (m_options.enable_stencil_test) {
+                stencil_ptrs[0] = coverage_bits & 1 ? &m_stencil_buffer->scanline(by)[bx] : nullptr;
+                stencil_ptrs[1] = coverage_bits & 2 ? &m_stencil_buffer->scanline(by)[bx + 1] : nullptr;
+                stencil_ptrs[2] = coverage_bits & 4 ? &m_stencil_buffer->scanline(by + 1)[bx] : nullptr;
+                stencil_ptrs[3] = coverage_bits & 8 ? &m_stencil_buffer->scanline(by + 1)[bx + 1] : nullptr;
+
+                stencil_value = load4_masked(stencil_ptrs[0], stencil_ptrs[1], stencil_ptrs[2], stencil_ptrs[3], quad.mask);
+                stencil_value &= stencil_configuration.test_mask;
+
+                i32x4 stencil_test_passed;
+                switch (stencil_configuration.test_function) {
+                case StencilTestFunction::Always:
+                    stencil_test_passed = expand4(~0);
+                    break;
+                case StencilTestFunction::Equal:
+                    stencil_test_passed = stencil_value == stencil_reference_value;
+                    break;
+                case StencilTestFunction::Greater:
+                    stencil_test_passed = stencil_value > stencil_reference_value;
+                    break;
+                case StencilTestFunction::GreaterOrEqual:
+                    stencil_test_passed = stencil_value >= stencil_reference_value;
+                    break;
+                case StencilTestFunction::Less:
+                    stencil_test_passed = stencil_value < stencil_reference_value;
+                    break;
+                case StencilTestFunction::LessOrEqual:
+                    stencil_test_passed = stencil_value <= stencil_reference_value;
+                    break;
+                case StencilTestFunction::Never:
+                    stencil_test_passed = expand4(0);
+                    break;
+                case StencilTestFunction::NotEqual:
+                    stencil_test_passed = stencil_value != stencil_reference_value;
+                    break;
+                default:
+                    VERIFY_NOT_REACHED();
+                }
+
+                // Update stencil buffer for pixels that failed the stencil test
+                write_to_stencil(
+                    stencil_ptrs,
+                    stencil_value,
+                    stencil_configuration.on_stencil_test_fail,
+                    stencil_reference_value,
+                    stencil_configuration.write_mask,
+                    quad.mask & ~stencil_test_passed);
+
+                // Update coverage mask + early quad rejection
+                quad.mask &= stencil_test_passed;
+                if (none(quad.mask))
+                    continue;
+            }
+
+            // Depth testing
             float* depth_ptrs[4] = {
                 coverage_bits & 1 ? &m_depth_buffer->scanline(by)[bx] : nullptr,
                 coverage_bits & 2 ? &m_depth_buffer->scanline(by)[bx + 1] : nullptr,
                 coverage_bits & 4 ? &m_depth_buffer->scanline(by + 1)[bx] : nullptr,
                 coverage_bits & 8 ? &m_depth_buffer->scanline(by + 1)[bx + 1] : nullptr,
             };
-
-            // AND the depth mask onto the coverage mask
             if (m_options.enable_depth_test) {
                 auto depth = load4_masked(depth_ptrs[0], depth_ptrs[1], depth_ptrs[2], depth_ptrs[3], quad.mask);
 
@@ -321,31 +413,35 @@ void Device::rasterize_triangle(const Triangle& triangle)
                 // FIXME: Also apply depth_offset_factor which depends on the depth gradient
                 quad.depth += m_options.depth_offset_constant * NumericLimits<float>::epsilon();
 
+                i32x4 depth_test_passed;
                 switch (m_options.depth_func) {
                 case DepthTestFunction::Always:
+                    depth_test_passed = expand4(~0);
                     break;
                 case DepthTestFunction::Never:
-                    quad.mask ^= quad.mask;
+                    depth_test_passed = expand4(0);
                     break;
                 case DepthTestFunction::Greater:
-                    quad.mask &= quad.depth > depth;
+                    depth_test_passed = quad.depth > depth;
                     break;
                 case DepthTestFunction::GreaterOrEqual:
-                    quad.mask &= quad.depth >= depth;
+                    depth_test_passed = quad.depth >= depth;
                     break;
                 case DepthTestFunction::NotEqual:
 #ifdef __SSE__
-                    quad.mask &= quad.depth != depth;
+                    depth_test_passed = quad.depth != depth;
 #else
-                    quad.mask[0] = bit_cast<u32>(quad.depth[0]) != bit_cast<u32>(depth[0]) ? -1 : 0;
-                    quad.mask[1] = bit_cast<u32>(quad.depth[1]) != bit_cast<u32>(depth[1]) ? -1 : 0;
-                    quad.mask[2] = bit_cast<u32>(quad.depth[2]) != bit_cast<u32>(depth[2]) ? -1 : 0;
-                    quad.mask[3] = bit_cast<u32>(quad.depth[3]) != bit_cast<u32>(depth[3]) ? -1 : 0;
+                    depth_test_passed = i32x4 {
+                        bit_cast<u32>(quad.depth[0]) != bit_cast<u32>(depth[0]) ? -1 : 0,
+                        bit_cast<u32>(quad.depth[1]) != bit_cast<u32>(depth[1]) ? -1 : 0,
+                        bit_cast<u32>(quad.depth[2]) != bit_cast<u32>(depth[2]) ? -1 : 0,
+                        bit_cast<u32>(quad.depth[3]) != bit_cast<u32>(depth[3]) ? -1 : 0,
+                    };
 #endif
                     break;
                 case DepthTestFunction::Equal:
 #ifdef __SSE__
-                    quad.mask &= quad.depth == depth;
+                    depth_test_passed = quad.depth == depth;
 #else
                     //
                     // This is an interesting quirk that occurs due to us using the x87 FPU when Serenity is
@@ -358,25 +454,52 @@ void Device::rasterize_triangle(const Triangle& triangle)
                     // the first 32-bits of this depth value is "good enough" that if we get a hit on it being
                     // equal, we can pretty much guarantee that it's actually equal.
                     //
-                    quad.mask[0] = bit_cast<u32>(quad.depth[0]) == bit_cast<u32>(depth[0]) ? -1 : 0;
-                    quad.mask[1] = bit_cast<u32>(quad.depth[1]) == bit_cast<u32>(depth[1]) ? -1 : 0;
-                    quad.mask[2] = bit_cast<u32>(quad.depth[2]) == bit_cast<u32>(depth[2]) ? -1 : 0;
-                    quad.mask[3] = bit_cast<u32>(quad.depth[3]) == bit_cast<u32>(depth[3]) ? -1 : 0;
+                    depth_test_passed = i32x4 {
+                        bit_cast<u32>(quad.depth[0]) == bit_cast<u32>(depth[0]) ? -1 : 0,
+                        bit_cast<u32>(quad.depth[1]) == bit_cast<u32>(depth[1]) ? -1 : 0,
+                        bit_cast<u32>(quad.depth[2]) == bit_cast<u32>(depth[2]) ? -1 : 0,
+                        bit_cast<u32>(quad.depth[3]) == bit_cast<u32>(depth[3]) ? -1 : 0,
+                    };
 #endif
                     break;
                 case DepthTestFunction::LessOrEqual:
-                    quad.mask &= quad.depth <= depth;
+                    depth_test_passed = quad.depth <= depth;
                     break;
                 case DepthTestFunction::Less:
-                    quad.mask &= quad.depth < depth;
+                    depth_test_passed = quad.depth < depth;
                     break;
+                default:
+                    VERIFY_NOT_REACHED();
                 }
 
-                // Nice, no pixels passed the depth test -> block rejected by early z
+                // Update stencil buffer for pixels that failed the depth test
+                if (m_options.enable_stencil_test) {
+                    write_to_stencil(
+                        stencil_ptrs,
+                        stencil_value,
+                        stencil_configuration.on_depth_test_fail,
+                        stencil_reference_value,
+                        stencil_configuration.write_mask,
+                        quad.mask & ~depth_test_passed);
+                }
+
+                // Update coverage mask + early quad rejection
+                quad.mask &= depth_test_passed;
                 if (none(quad.mask))
                     continue;
             }
 
+            // Update stencil buffer for passed pixels
+            if (m_options.enable_stencil_test) {
+                write_to_stencil(
+                    stencil_ptrs,
+                    stencil_value,
+                    stencil_configuration.on_pass,
+                    stencil_reference_value,
+                    stencil_configuration.write_mask,
+                    quad.mask);
+            }
+
             INCREASE_STATISTICS_COUNTER(g_num_pixels_shaded, maskcount(quad.mask));
 
             // Draw the pixels according to the previously generated mask
@@ -415,9 +538,8 @@ void Device::rasterize_triangle(const Triangle& triangle)
             }
 
             // Write to depth buffer
-            if (m_options.enable_depth_test && m_options.enable_depth_write) {
+            if (m_options.enable_depth_test && m_options.enable_depth_write)
                 store4_masked(quad.depth, depth_ptrs[0], depth_ptrs[1], depth_ptrs[2], depth_ptrs[3], quad.mask);
-            }
 
             // We will not update the color buffer at all
             if (!m_options.color_mask || !m_options.enable_color_write)
@@ -465,8 +587,9 @@ void Device::rasterize_triangle(const Triangle& triangle)
 }
 
 Device::Device(const Gfx::IntSize& size)
-    : m_render_target { Gfx::Bitmap::try_create(Gfx::BitmapFormat::BGRA8888, size).release_value_but_fixme_should_propagate_errors() }
-    , m_depth_buffer { adopt_own(*new DepthBuffer(size)) }
+    : m_render_target(Gfx::Bitmap::try_create(Gfx::BitmapFormat::BGRA8888, size).release_value_but_fixme_should_propagate_errors())
+    , m_depth_buffer(make<DepthBuffer>(size))
+    , m_stencil_buffer(MUST(StencilBuffer::try_create(size)))
 {
     m_options.scissor_box = m_render_target->rect();
     m_options.viewport = m_render_target->rect();
@@ -478,7 +601,8 @@ DeviceInfo Device::info() const
         .vendor_name = "SerenityOS",
         .device_name = "SoftGPU",
         .num_texture_units = NUM_SAMPLERS,
-        .num_lights = NUM_LIGHTS
+        .num_lights = NUM_LIGHTS,
+        .stencil_bits = sizeof(u8) * 8,
     };
 }
 
@@ -626,7 +750,7 @@ void Device::draw_primitives(PrimitiveType primitive_type, FloatMatrix4x4 const&
     }
 
     // Now let's transform each triangle and send that to the GPU
-    auto const viewport = window_coordinates_to_target_coordinates(m_options.viewport, m_render_target->rect());
+    auto const viewport = window_coordinates_to_target_coordinates(m_options.viewport);
     auto const viewport_half_width = viewport.width() / 2.0f;
     auto const viewport_half_height = viewport.height() / 2.0f;
     auto const viewport_center_x = viewport.x() + viewport_half_width;
@@ -956,7 +1080,7 @@ void Device::clear_color(const FloatVector4& color)
 
     if (m_options.scissor_enabled) {
         auto fill_rect = m_render_target->rect();
-        fill_rect.intersect(window_coordinates_to_target_coordinates(m_options.scissor_box, fill_rect));
+        fill_rect.intersect(window_coordinates_to_target_coordinates(m_options.scissor_box));
         Gfx::Painter painter { *m_render_target };
         painter.fill_rect(fill_rect, fill_color);
         return;
@@ -970,13 +1094,23 @@ void Device::clear_depth(float depth)
     wait_for_all_threads();
 
     if (m_options.scissor_enabled) {
-        m_depth_buffer->clear(window_coordinates_to_target_coordinates(m_options.scissor_box, m_render_target->rect()), depth);
+        m_depth_buffer->clear(window_coordinates_to_target_coordinates(m_options.scissor_box), depth);
         return;
     }
 
     m_depth_buffer->clear(depth);
 }
 
+void Device::clear_stencil(u8 value)
+{
+    Gfx::IntRect clear_rect = m_stencil_buffer->rect();
+
+    if (m_options.scissor_enabled)
+        clear_rect.intersect(window_coordinates_to_target_coordinates(m_options.scissor_box));
+
+    m_stencil_buffer->clear(clear_rect, value);
+}
+
 void Device::blit_to_color_buffer_at_raster_position(Gfx::Bitmap const& source)
 {
     if (!m_raster_position.valid)
@@ -1148,6 +1282,11 @@ void Device::set_material_state(Face face, Material const& material)
     m_materials[face] = material;
 }
 
+void Device::set_stencil_configuration(Face face, StencilConfiguration const& stencil_configuration)
+{
+    m_stencil_configuration[face] = stencil_configuration;
+}
+
 void Device::set_raster_position(RasterPosition const& raster_position)
 {
     m_raster_position = raster_position;
@@ -1192,7 +1331,7 @@ Gfx::IntRect Device::raster_rect_in_target_coordinates(Gfx::IntSize size)
         size.width(),
         size.height(),
     };
-    return window_coordinates_to_target_coordinates(raster_rect, m_render_target->rect());
+    return window_coordinates_to_target_coordinates(raster_rect);
 }
 
 }
diff --git a/Userland/Libraries/LibSoftGPU/Device.h b/Userland/Libraries/LibSoftGPU/Device.h
index fc38e2cebb..8d761a49f5 100644
--- a/Userland/Libraries/LibSoftGPU/Device.h
+++ b/Userland/Libraries/LibSoftGPU/Device.h
@@ -10,6 +10,7 @@
 #include <AK/Array.h>
 #include <AK/NonnullRefPtr.h>
 #include <AK/OwnPtr.h>
+#include <AK/Vector.h>
 #include <LibGfx/Bitmap.h>
 #include <LibGfx/Matrix3x3.h>
 #include <LibGfx/Matrix4x4.h>
@@ -26,6 +27,7 @@
 #include <LibSoftGPU/Light/Light.h>
 #include <LibSoftGPU/Light/Material.h>
 #include <LibSoftGPU/Sampler.h>
+#include <LibSoftGPU/StencilBuffer.h>
 #include <LibSoftGPU/Triangle.h>
 #include <LibSoftGPU/Vertex.h>
 
@@ -38,6 +40,7 @@ struct TexCoordGenerationConfig {
 
 struct RasterizerOptions {
     bool shade_smooth { true };
+    bool enable_stencil_test { false };
     bool enable_depth_test { false };
     bool enable_depth_write { true };
     bool enable_alpha_test { false };
@@ -94,6 +97,17 @@ struct RasterPosition {
     FloatVector4 texture_coordinates { 0.0f, 0.0f, 0.0f, 1.0f };
 };
 
+struct StencilConfiguration {
+    StencilTestFunction test_function;
+    u8 reference_value;
+    u8 test_mask;
+
+    StencilOperation on_stencil_test_fail;
+    StencilOperation on_depth_test_fail;
+    StencilOperation on_pass;
+    u8 write_mask;
+};
+
 class Device final {
 public:
     Device(const Gfx::IntSize& min_size);
@@ -104,6 +118,7 @@ public:
     void resize(const Gfx::IntSize& min_size);
     void clear_color(const FloatVector4&);
     void clear_depth(float);
+    void clear_stencil(u8);
     void blit_to(Gfx::Bitmap&);
     void blit_to_color_buffer_at_raster_position(Gfx::Bitmap const&);
     void blit_to_depth_buffer_at_raster_position(Vector<float> const&, size_t, size_t);
@@ -120,6 +135,7 @@ public:
     void set_sampler_config(unsigned, SamplerConfig const&);
     void set_light_state(unsigned, Light const&);
     void set_material_state(Face, Material const&);
+    void set_stencil_configuration(Face, StencilConfiguration const&);
 
     RasterPosition raster_position() const { return m_raster_position; }
     void set_raster_position(RasterPosition const& raster_position);
@@ -128,15 +144,16 @@ public:
 private:
     void draw_statistics_overlay(Gfx::Bitmap&);
     Gfx::IntRect raster_rect_in_target_coordinates(Gfx::IntSize size);
+    Gfx::IntRect window_coordinates_to_target_coordinates(Gfx::IntRect const&);
 
     void rasterize_triangle(const Triangle& triangle);
     void setup_blend_factors();
     void shade_fragments(PixelQuad&);
     bool test_alpha(PixelQuad&);
 
-private:
     RefPtr<Gfx::Bitmap> m_render_target;
-    OwnPtr<DepthBuffer> m_depth_buffer;
+    NonnullOwnPtr<DepthBuffer> m_depth_buffer;
+    NonnullOwnPtr<StencilBuffer> m_stencil_buffer;
     RasterizerOptions m_options;
     LightModelParameters m_lighting_model;
     Clipper m_clipper;
@@ -149,6 +166,7 @@ private:
     Array<Light, NUM_LIGHTS> m_lights;
     Array<Material, 2u> m_materials;
     RasterPosition m_raster_position;
+    Array<StencilConfiguration, 2u> m_stencil_configuration;
 };
 
 }
diff --git a/Userland/Libraries/LibSoftGPU/DeviceInfo.h b/Userland/Libraries/LibSoftGPU/DeviceInfo.h
index 41038aa2a7..840d87536b 100644
--- a/Userland/Libraries/LibSoftGPU/DeviceInfo.h
+++ b/Userland/Libraries/LibSoftGPU/DeviceInfo.h
@@ -15,6 +15,7 @@ struct DeviceInfo final {
     String device_name;
     unsigned num_texture_units;
     unsigned num_lights;
+    u8 stencil_bits;
 };
 
 }
diff --git a/Userland/Libraries/LibSoftGPU/Enums.h b/Userland/Libraries/LibSoftGPU/Enums.h
index 38f5fa08fe..79c31d2ac7 100644
--- a/Userland/Libraries/LibSoftGPU/Enums.h
+++ b/Userland/Libraries/LibSoftGPU/Enums.h
@@ -87,6 +87,28 @@ enum class PrimitiveType {
     Quads,
 };
 
+enum StencilOperation {
+    Decrement,
+    DecrementWrap,
+    Increment,
+    IncrementWrap,
+    Invert,
+    Keep,
+    Replace,
+    Zero,
+};
+
+enum StencilTestFunction {
+    Always,
+    Equal,
+    Greater,
+    GreaterOrEqual,
+    Less,
+    LessOrEqual,
+    Never,
+    NotEqual,
+};
+
 enum TexCoordGenerationCoordinate {
     None = 0x0,
     S = 0x1,
diff --git a/Userland/Libraries/LibSoftGPU/StencilBuffer.cpp b/Userland/Libraries/LibSoftGPU/StencilBuffer.cpp
new file mode 100644
index 0000000000..b9aad2e626
--- /dev/null
+++ b/Userland/Libraries/LibSoftGPU/StencilBuffer.cpp
@@ -0,0 +1,41 @@
+/*
+ * Copyright (c) 2022, Jelle Raaijmakers <jelle@gmta.nl>
+ *
+ * SPDX-License-Identifier: BSD-2-Clause
+ */
+
+#include <LibSoftGPU/StencilBuffer.h>
+
+namespace SoftGPU {
+
+ErrorOr<NonnullOwnPtr<StencilBuffer>> StencilBuffer::try_create(Gfx::IntSize const& size)
+{
+    auto rect = Gfx::IntRect { 0, 0, size.width(), size.height() };
+    auto data = TRY(FixedArray<u8>::try_create(size.area()));
+    return adopt_own(*new StencilBuffer(rect, move(data)));
+}
+
+StencilBuffer::StencilBuffer(Gfx::IntRect const& rect, FixedArray<u8> data)
+    : m_data(move(data))
+    , m_rect(rect)
+{
+}
+
+void StencilBuffer::clear(Gfx::IntRect rect, u8 value)
+{
+    rect.intersect(m_rect);
+
+    for (int y = rect.top(); y <= rect.bottom(); ++y) {
+        auto* line = scanline(y);
+        for (int x = rect.left(); x <= rect.right(); ++x)
+            line[x] = value;
+    }
+}
+
+u8* StencilBuffer::scanline(int y)
+{
+    VERIFY(m_rect.contains_vertically(y));
+    return &m_data[y * m_rect.width()];
+}
+
+}
diff --git a/Userland/Libraries/LibSoftGPU/StencilBuffer.h b/Userland/Libraries/LibSoftGPU/StencilBuffer.h
new file mode 100644
index 0000000000..b79340557a
--- /dev/null
+++ b/Userland/Libraries/LibSoftGPU/StencilBuffer.h
@@ -0,0 +1,33 @@
+/*
+ * Copyright (c) 2021, Jelle Raaijmakers <jelle@gmta.nl>
+ *
+ * SPDX-License-Identifier: BSD-2-Clause
+ */
+
+#pragma once
+
+#include <AK/Error.h>
+#include <AK/FixedArray.h>
+#include <AK/NonnullOwnPtr.h>
+#include <AK/Try.h>
+#include <LibGfx/Rect.h>
+#include <LibGfx/Size.h>
+
+namespace SoftGPU {
+
+class StencilBuffer final {
+public:
+    static ErrorOr<NonnullOwnPtr<StencilBuffer>> try_create(Gfx::IntSize const& size);
+
+    void clear(Gfx::IntRect rect, u8 value);
+    Gfx::IntRect const& rect() const { return m_rect; }
+    u8* scanline(int y);
+
+private:
+    StencilBuffer(Gfx::IntRect const& rect, FixedArray<u8> data);
+
+    FixedArray<u8> m_data;
+    Gfx::IntRect m_rect;
+};
+
+}