diff --git a/piet-gpu-types/src/annotated.rs b/piet-gpu-types/src/annotated.rs
index 681a7ec..04f2111 100644
--- a/piet-gpu-types/src/annotated.rs
+++ b/piet-gpu-types/src/annotated.rs
@@ -18,12 +18,17 @@ piet_gpu! {
             // That's expected to be uncommon, so we could special-case it.
             linewidth: f32,
         }
+        struct AnnoClip {
+            bbox: [f32; 4],
+        }
         enum Annotated {
             Nop,
             Stroke(AnnoStroke),
             Fill(AnnoFill),
             FillMask(AnnoFillMask),
             FillMaskInv(AnnoFillMask),
+            BeginClip(AnnoClip),
+            EndClip(AnnoClip),
         }
     }
 }
diff --git a/piet-gpu-types/src/ptcl.rs b/piet-gpu-types/src/ptcl.rs
index 95dcdc6..d05218b 100644
--- a/piet-gpu-types/src/ptcl.rs
+++ b/piet-gpu-types/src/ptcl.rs
@@ -30,6 +30,14 @@ piet_gpu! {
             backdrop: i32,
             mask: f32,
         }
+        struct CmdBeginClip {
+            tile_ref: u32,
+            backdrop: i32,
+        }
+        struct CmdEndClip {
+            // This will be 1.0 for clips, but we can imagine blend groups.
+            alpha: f32,
+        }
         struct CmdSolid {
             rgba_color: u32,
         }
@@ -46,6 +54,8 @@ piet_gpu! {
             Fill(CmdFill),
             FillMask(CmdFillMask),
             FillMaskInv(CmdFillMask),
+            BeginClip(CmdBeginClip),
+            EndClip(CmdEndClip),
             Stroke(CmdStroke),
             Solid(CmdSolid),
             SolidMask(CmdSolidMask),
diff --git a/piet-gpu-types/src/scene.rs b/piet-gpu-types/src/scene.rs
index 1359c1b..7e2fb43 100644
--- a/piet-gpu-types/src/scene.rs
+++ b/piet-gpu-types/src/scene.rs
@@ -1,6 +1,8 @@
 use piet_gpu_derive::piet_gpu;
 
-pub use self::scene::{CubicSeg, Element, Fill, LineSeg, QuadSeg, SetLineWidth, Stroke, Transform};
+pub use self::scene::{
+    BeginClip, CubicSeg, Element, EndClip, Fill, LineSeg, QuadSeg, SetLineWidth, Stroke, Transform,
+};
 
 piet_gpu! {
     #[rust_encode]
@@ -36,6 +38,15 @@ piet_gpu! {
             mat: [f32; 4],
             translate: [f32; 2],
         }
+        struct BeginClip {
+            bbox: [f32; 4],
+            // TODO: add alpha?
+        }
+        struct EndClip {
+            // The delta between the BeginClip and EndClip element indices.
+            // It is stored as a delta to facilitate binary string concatenation.
+            delta: u32,
+        }
         enum Element {
             Nop,
             // Another approach to encoding would be to use a single
@@ -55,6 +66,8 @@ piet_gpu! {
             Transform(Transform),
             FillMask(FillMask),
             FillMaskInv(FillMask),
+            BeginClip(BeginClip),
+            EndClip(EndClip),
         }
     }
 }
diff --git a/piet-gpu/shader/annotated.h b/piet-gpu/shader/annotated.h
index 847ca06..986eff2 100644
--- a/piet-gpu/shader/annotated.h
+++ b/piet-gpu/shader/annotated.h
@@ -12,6 +12,10 @@ struct AnnoStrokeRef {
     uint offset;
 };
 
+struct AnnoClipRef {
+    uint offset;
+};
+
 struct AnnotatedRef {
     uint offset;
 };
@@ -50,11 +54,23 @@ AnnoStrokeRef AnnoStroke_index(AnnoStrokeRef ref, uint index) {
     return AnnoStrokeRef(ref.offset + index * AnnoStroke_size);
 }
 
+struct AnnoClip {
+    vec4 bbox;
+};
+
+#define AnnoClip_size 16
+
+AnnoClipRef AnnoClip_index(AnnoClipRef ref, uint index) {
+    return AnnoClipRef(ref.offset + index * AnnoClip_size);
+}
+
 #define Annotated_Nop 0
 #define Annotated_Stroke 1
 #define Annotated_Fill 2
 #define Annotated_FillMask 3
 #define Annotated_FillMaskInv 4
+#define Annotated_BeginClip 5
+#define Annotated_EndClip 6
 #define Annotated_size 28
 
 AnnotatedRef Annotated_index(AnnotatedRef ref, uint index) {
@@ -130,6 +146,25 @@ void AnnoStroke_write(AnnoStrokeRef ref, AnnoStroke s) {
     annotated[ix + 5] = floatBitsToUint(s.linewidth);
 }
 
+AnnoClip AnnoClip_read(AnnoClipRef ref) {
+    uint ix = ref.offset >> 2;
+    uint raw0 = annotated[ix + 0];
+    uint raw1 = annotated[ix + 1];
+    uint raw2 = annotated[ix + 2];
+    uint raw3 = annotated[ix + 3];
+    AnnoClip s;
+    s.bbox = vec4(uintBitsToFloat(raw0), uintBitsToFloat(raw1), uintBitsToFloat(raw2), uintBitsToFloat(raw3));
+    return s;
+}
+
+void AnnoClip_write(AnnoClipRef ref, AnnoClip s) {
+    uint ix = ref.offset >> 2;
+    annotated[ix + 0] = floatBitsToUint(s.bbox.x);
+    annotated[ix + 1] = floatBitsToUint(s.bbox.y);
+    annotated[ix + 2] = floatBitsToUint(s.bbox.z);
+    annotated[ix + 3] = floatBitsToUint(s.bbox.w);
+}
+
 uint Annotated_tag(AnnotatedRef ref) {
     return annotated[ref.offset >> 2];
 }
@@ -150,6 +185,14 @@ AnnoFillMask Annotated_FillMaskInv_read(AnnotatedRef ref) {
     return AnnoFillMask_read(AnnoFillMaskRef(ref.offset + 4));
 }
 
+AnnoClip Annotated_BeginClip_read(AnnotatedRef ref) {
+    return AnnoClip_read(AnnoClipRef(ref.offset + 4));
+}
+
+AnnoClip Annotated_EndClip_read(AnnotatedRef ref) {
+    return AnnoClip_read(AnnoClipRef(ref.offset + 4));
+}
+
 void Annotated_Nop_write(AnnotatedRef ref) {
     annotated[ref.offset >> 2] = Annotated_Nop;
 }
@@ -174,3 +217,13 @@ void Annotated_FillMaskInv_write(AnnotatedRef ref, AnnoFillMask s) {
     AnnoFillMask_write(AnnoFillMaskRef(ref.offset + 4), s);
 }
 
+void Annotated_BeginClip_write(AnnotatedRef ref, AnnoClip s) {
+    annotated[ref.offset >> 2] = Annotated_BeginClip;
+    AnnoClip_write(AnnoClipRef(ref.offset + 4), s);
+}
+
+void Annotated_EndClip_write(AnnotatedRef ref, AnnoClip s) {
+    annotated[ref.offset >> 2] = Annotated_EndClip;
+    AnnoClip_write(AnnoClipRef(ref.offset + 4), s);
+}
+
diff --git a/piet-gpu/shader/backdrop.spv b/piet-gpu/shader/backdrop.spv
index e2093d9..7f0852d 100644
Binary files a/piet-gpu/shader/backdrop.spv and b/piet-gpu/shader/backdrop.spv differ
diff --git a/piet-gpu/shader/binning.spv b/piet-gpu/shader/binning.spv
index 50070a1..8f44c89 100644
Binary files a/piet-gpu/shader/binning.spv and b/piet-gpu/shader/binning.spv differ
diff --git a/piet-gpu/shader/coarse.spv b/piet-gpu/shader/coarse.spv
index 7758b96..17bae64 100644
Binary files a/piet-gpu/shader/coarse.spv and b/piet-gpu/shader/coarse.spv differ
diff --git a/piet-gpu/shader/kernel4.comp b/piet-gpu/shader/kernel4.comp
index d21aa87..919d120 100644
--- a/piet-gpu/shader/kernel4.comp
+++ b/piet-gpu/shader/kernel4.comp
@@ -27,6 +27,8 @@ layout(rgba8, set = 0, binding = 2) uniform writeonly image2D image;
 #include "ptcl.h"
 #include "tile.h"
 
+#define BLEND_STACK_SIZE 4
+
 // Calculate coverage based on backdrop + coverage of each line segment
 float[CHUNK] computeArea(vec2 xy, int backdrop, uint tile_ref) {
     // Probably better to store as float, but conversion is no doubt cheap.
@@ -69,6 +71,8 @@ void main() {
     vec2 xy = vec2(xy_uint);
     vec3 rgb[CHUNK];
     float mask[CHUNK];
+    uint blend_stack[BLEND_STACK_SIZE][CHUNK];
+    uint blend_sp = 0;
     for (uint i = 0; i < CHUNK; i++) {
         rgb[i] = vec3(0.5);
         mask[i] = 1.0;
@@ -137,6 +141,22 @@ void main() {
                 mask[k] = mix(mask[k], fill_mask.mask, 1.0 - area[k]);
             }
             break;
+        case Cmd_BeginClip:
+            CmdBeginClip begin_clip = Cmd_BeginClip_read(cmd_ref);
+            area = computeArea(xy, begin_clip.backdrop, begin_clip.tile_ref);
+            for (uint k = 0; k < CHUNK; k++) {
+                blend_stack[blend_sp][k] = packUnorm4x8(vec4(rgb[k], clamp(abs(area[k]), 0.0, 1.0)));
+            }
+            blend_sp++;
+            break;
+        case Cmd_EndClip:
+            CmdEndClip end_clip = Cmd_EndClip_read(cmd_ref);
+            blend_sp--;
+            for (uint k = 0; k < CHUNK; k++) {
+                vec4 rgba = unpackUnorm4x8(blend_stack[blend_sp][k]);
+                rgb[k] = mix(rgb[k], rgba.rgb, end_clip.alpha * rgba.a);
+            }
+            break;
         case Cmd_Solid:
             CmdSolid solid = Cmd_Solid_read(cmd_ref);
             fg_rgba = unpackUnorm4x8(solid.rgba_color).wzyx;
diff --git a/piet-gpu/shader/kernel4.spv b/piet-gpu/shader/kernel4.spv
index d159fe5..0d38581 100644
Binary files a/piet-gpu/shader/kernel4.spv and b/piet-gpu/shader/kernel4.spv differ
diff --git a/piet-gpu/shader/path_coarse.spv b/piet-gpu/shader/path_coarse.spv
index db5bc57..f82a031 100644
Binary files a/piet-gpu/shader/path_coarse.spv and b/piet-gpu/shader/path_coarse.spv differ
diff --git a/piet-gpu/shader/ptcl.h b/piet-gpu/shader/ptcl.h
index d1cc83d..33ab664 100644
--- a/piet-gpu/shader/ptcl.h
+++ b/piet-gpu/shader/ptcl.h
@@ -20,6 +20,14 @@ struct CmdFillMaskRef {
     uint offset;
 };
 
+struct CmdBeginClipRef {
+    uint offset;
+};
+
+struct CmdEndClipRef {
+    uint offset;
+};
+
 struct CmdSolidRef {
     uint offset;
 };
@@ -103,6 +111,27 @@ CmdFillMaskRef CmdFillMask_index(CmdFillMaskRef ref, uint index) {
     return CmdFillMaskRef(ref.offset + index * CmdFillMask_size);
 }
 
+struct CmdBeginClip {
+    uint tile_ref;
+    int backdrop;
+};
+
+#define CmdBeginClip_size 8
+
+CmdBeginClipRef CmdBeginClip_index(CmdBeginClipRef ref, uint index) {
+    return CmdBeginClipRef(ref.offset + index * CmdBeginClip_size);
+}
+
+struct CmdEndClip {
+    float alpha;
+};
+
+#define CmdEndClip_size 4
+
+CmdEndClipRef CmdEndClip_index(CmdEndClipRef ref, uint index) {
+    return CmdEndClipRef(ref.offset + index * CmdEndClip_size);
+}
+
 struct CmdSolid {
     uint rgba_color;
 };
@@ -139,10 +168,12 @@ CmdJumpRef CmdJump_index(CmdJumpRef ref, uint index) {
 #define Cmd_Fill 3
 #define Cmd_FillMask 4
 #define Cmd_FillMaskInv 5
-#define Cmd_Stroke 6
-#define Cmd_Solid 7
-#define Cmd_SolidMask 8
-#define Cmd_Jump 9
+#define Cmd_BeginClip 6
+#define Cmd_EndClip 7
+#define Cmd_Stroke 8
+#define Cmd_Solid 9
+#define Cmd_SolidMask 10
+#define Cmd_Jump 11
 #define Cmd_size 20
 
 CmdRef Cmd_index(CmdRef ref, uint index) {
@@ -271,6 +302,35 @@ void CmdFillMask_write(CmdFillMaskRef ref, CmdFillMask s) {
     ptcl[ix + 2] = floatBitsToUint(s.mask);
 }
 
+CmdBeginClip CmdBeginClip_read(CmdBeginClipRef ref) {
+    uint ix = ref.offset >> 2;
+    uint raw0 = ptcl[ix + 0];
+    uint raw1 = ptcl[ix + 1];
+    CmdBeginClip s;
+    s.tile_ref = raw0;
+    s.backdrop = int(raw1);
+    return s;
+}
+
+void CmdBeginClip_write(CmdBeginClipRef ref, CmdBeginClip s) {
+    uint ix = ref.offset >> 2;
+    ptcl[ix + 0] = s.tile_ref;
+    ptcl[ix + 1] = uint(s.backdrop);
+}
+
+CmdEndClip CmdEndClip_read(CmdEndClipRef ref) {
+    uint ix = ref.offset >> 2;
+    uint raw0 = ptcl[ix + 0];
+    CmdEndClip s;
+    s.alpha = uintBitsToFloat(raw0);
+    return s;
+}
+
+void CmdEndClip_write(CmdEndClipRef ref, CmdEndClip s) {
+    uint ix = ref.offset >> 2;
+    ptcl[ix + 0] = floatBitsToUint(s.alpha);
+}
+
 CmdSolid CmdSolid_read(CmdSolidRef ref) {
     uint ix = ref.offset >> 2;
     uint raw0 = ptcl[ix + 0];
@@ -334,6 +394,14 @@ CmdFillMask Cmd_FillMaskInv_read(CmdRef ref) {
     return CmdFillMask_read(CmdFillMaskRef(ref.offset + 4));
 }
 
+CmdBeginClip Cmd_BeginClip_read(CmdRef ref) {
+    return CmdBeginClip_read(CmdBeginClipRef(ref.offset + 4));
+}
+
+CmdEndClip Cmd_EndClip_read(CmdRef ref) {
+    return CmdEndClip_read(CmdEndClipRef(ref.offset + 4));
+}
+
 CmdStroke Cmd_Stroke_read(CmdRef ref) {
     return CmdStroke_read(CmdStrokeRef(ref.offset + 4));
 }
@@ -379,6 +447,16 @@ void Cmd_FillMaskInv_write(CmdRef ref, CmdFillMask s) {
     CmdFillMask_write(CmdFillMaskRef(ref.offset + 4), s);
 }
 
+void Cmd_BeginClip_write(CmdRef ref, CmdBeginClip s) {
+    ptcl[ref.offset >> 2] = Cmd_BeginClip;
+    CmdBeginClip_write(CmdBeginClipRef(ref.offset + 4), s);
+}
+
+void Cmd_EndClip_write(CmdRef ref, CmdEndClip s) {
+    ptcl[ref.offset >> 2] = Cmd_EndClip;
+    CmdEndClip_write(CmdEndClipRef(ref.offset + 4), s);
+}
+
 void Cmd_Stroke_write(CmdRef ref, CmdStroke s) {
     ptcl[ref.offset >> 2] = Cmd_Stroke;
     CmdStroke_write(CmdStrokeRef(ref.offset + 4), s);
diff --git a/piet-gpu/shader/scene.h b/piet-gpu/shader/scene.h
index 6823fe6..0a4a2ce 100644
--- a/piet-gpu/shader/scene.h
+++ b/piet-gpu/shader/scene.h
@@ -32,6 +32,14 @@ struct TransformRef {
     uint offset;
 };
 
+struct BeginClipRef {
+    uint offset;
+};
+
+struct EndClipRef {
+    uint offset;
+};
+
 struct ElementRef {
     uint offset;
 };
@@ -123,6 +131,26 @@ TransformRef Transform_index(TransformRef ref, uint index) {
     return TransformRef(ref.offset + index * Transform_size);
 }
 
+struct BeginClip {
+    vec4 bbox;
+};
+
+#define BeginClip_size 16
+
+BeginClipRef BeginClip_index(BeginClipRef ref, uint index) {
+    return BeginClipRef(ref.offset + index * BeginClip_size);
+}
+
+struct EndClip {
+    uint clip_size;
+};
+
+#define EndClip_size 4
+
+EndClipRef EndClip_index(EndClipRef ref, uint index) {
+    return EndClipRef(ref.offset + index * EndClip_size);
+}
+
 #define Element_Nop 0
 #define Element_StrokeLine 1
 #define Element_FillLine 2
@@ -136,6 +164,8 @@ TransformRef Transform_index(TransformRef ref, uint index) {
 #define Element_Transform 10
 #define Element_FillMask 11
 #define Element_FillMaskInv 12
+#define Element_BeginClip 13
+#define Element_EndClip 14
 #define Element_size 36
 
 ElementRef Element_index(ElementRef ref, uint index) {
@@ -233,6 +263,25 @@ Transform Transform_read(TransformRef ref) {
     return s;
 }
 
+BeginClip BeginClip_read(BeginClipRef ref) {
+    uint ix = ref.offset >> 2;
+    uint raw0 = scene[ix + 0];
+    uint raw1 = scene[ix + 1];
+    uint raw2 = scene[ix + 2];
+    uint raw3 = scene[ix + 3];
+    BeginClip s;
+    s.bbox = vec4(uintBitsToFloat(raw0), uintBitsToFloat(raw1), uintBitsToFloat(raw2), uintBitsToFloat(raw3));
+    return s;
+}
+
+EndClip EndClip_read(EndClipRef ref) {
+    uint ix = ref.offset >> 2;
+    uint raw0 = scene[ix + 0];
+    EndClip s;
+    s.clip_size = raw0;
+    return s;
+}
+
 uint Element_tag(ElementRef ref) {
     return scene[ref.offset >> 2];
 }
@@ -285,3 +334,11 @@ FillMask Element_FillMaskInv_read(ElementRef ref) {
     return FillMask_read(FillMaskRef(ref.offset + 4));
 }
 
+BeginClip Element_BeginClip_read(ElementRef ref) {
+    return BeginClip_read(BeginClipRef(ref.offset + 4));
+}
+
+EndClip Element_EndClip_read(ElementRef ref) {
+    return EndClip_read(EndClipRef(ref.offset + 4));
+}
+
diff --git a/piet-gpu/shader/tile_alloc.spv b/piet-gpu/shader/tile_alloc.spv
index af52665..449f4d6 100644
Binary files a/piet-gpu/shader/tile_alloc.spv and b/piet-gpu/shader/tile_alloc.spv differ
diff --git a/piet-gpu/src/render_ctx.rs b/piet-gpu/src/render_ctx.rs
index 9593704..b3aa895 100644
--- a/piet-gpu/src/render_ctx.rs
+++ b/piet-gpu/src/render_ctx.rs
@@ -1,12 +1,13 @@
-use std::{borrow::Cow, ops::RangeBounds};
+use std::{borrow::Cow, convert::TryInto, ops::RangeBounds};
 
 use piet_gpu_types::encoder::{Encode, Encoder};
 
-use piet_gpu_types::scene::{CubicSeg, Element, Fill, LineSeg, QuadSeg, SetLineWidth, Stroke};
+use piet_gpu_types::scene::{
+    BeginClip, CubicSeg, Element, EndClip, Fill, LineSeg, QuadSeg, SetLineWidth, Stroke, Transform,
+};
 
 use piet::{
-    kurbo::Size,
-    kurbo::{Affine, PathEl, Point, Rect, Shape},
+    kurbo::{Affine, Insets, PathEl, Point, Rect, Shape, Size},
     HitTestPosition, TextAttribute, TextStorage,
 };
 
@@ -33,8 +34,14 @@ pub struct PietGpuRenderContext {
     stroke_width: f32,
     // We're tallying these cpu-side for expedience, but will probably
     // move this to some kind of readback from element processing.
+    /// The count of elements that make it through to coarse rasterization.
     path_count: usize,
+    /// The count of path segment elements.
     pathseg_count: usize,
+
+    cur_transform: Affine,
+    state_stack: Vec<State>,
+    clip_stack: Vec<ClipElement>,
 }
 
 #[derive(Clone)]
@@ -43,6 +50,21 @@ pub enum PietGpuBrush {
     Gradient,
 }
 
+#[derive(Default)]
+struct State {
+    /// The transform relative to the parent state.
+    transform: Affine,
+    n_clip: usize,
+}
+
+struct ClipElement {
+    /// Index of BeginClip element in element vec, for bbox fixup.
+    begin_ix: usize,
+    bbox: Option<Rect>,
+    /// The transform relative to the next clip element on the stack.
+    transform: Affine,
+}
+
 const TOLERANCE: f64 = 0.25;
 
 impl PietGpuRenderContext {
@@ -58,6 +80,9 @@ impl PietGpuRenderContext {
             stroke_width,
             path_count: 0,
             pathseg_count: 0,
+            cur_transform: Affine::default(),
+            state_stack: Vec::new(),
+            clip_stack: Vec::new(),
         }
     }
 
@@ -96,17 +121,19 @@ impl RenderContext for PietGpuRenderContext {
     fn clear(&mut self, _color: Color) {}
 
     fn stroke(&mut self, shape: impl Shape, brush: &impl IntoBrush<Self>, width: f64) {
-        let width = width as f32;
-        if self.stroke_width != width {
+        let width_f32 = width as f32;
+        if self.stroke_width != width_f32 {
             self.elements
-                .push(Element::SetLineWidth(SetLineWidth { width }));
-            self.stroke_width = width;
+                .push(Element::SetLineWidth(SetLineWidth { width: width_f32 }));
+            self.stroke_width = width_f32;
         }
         let brush = brush.make_brush(self, || shape.bounding_box()).into_owned();
-        let path = shape.path_elements(TOLERANCE);
-        self.encode_path(path, false);
         match brush {
             PietGpuBrush::Solid(rgba_color) => {
+                // Note: the bbox contribution of stroke becomes more complicated with miter joins.
+                self.accumulate_bbox(|| shape.bounding_box() + Insets::uniform(width * 0.5));
+                let path = shape.path_elements(TOLERANCE);
+                self.encode_path(path, false);
                 let stroke = Stroke { rgba_color };
                 self.elements.push(Element::Stroke(stroke));
                 self.path_count += 1;
@@ -126,21 +153,34 @@ impl RenderContext for PietGpuRenderContext {
 
     fn fill(&mut self, shape: impl Shape, brush: &impl IntoBrush<Self>) {
         let brush = brush.make_brush(self, || shape.bounding_box()).into_owned();
-        let path = shape.path_elements(TOLERANCE);
-        self.encode_path(path, true);
-        match brush {
-            PietGpuBrush::Solid(rgba_color) => {
-                let fill = Fill { rgba_color };
-                self.elements.push(Element::Fill(fill));
-                self.path_count += 1;
-            }
-            _ => (),
+        if let PietGpuBrush::Solid(rgba_color) = brush {
+            // Note: we might get a good speedup from using an approximate bounding box.
+            // Perhaps that should be added to kurbo.
+            self.accumulate_bbox(|| shape.bounding_box());
+            let path = shape.path_elements(TOLERANCE);
+            self.encode_path(path, true);
+            let fill = Fill { rgba_color };
+            self.elements.push(Element::Fill(fill));
+            self.path_count += 1;
         }
     }
 
     fn fill_even_odd(&mut self, _shape: impl Shape, _brush: &impl IntoBrush<Self>) {}
 
-    fn clip(&mut self, _shape: impl Shape) {}
+    fn clip(&mut self, shape: impl Shape) {
+        let begin_ix = self.elements.len();
+        let path = shape.path_elements(TOLERANCE);
+        self.encode_path(path, true);
+        self.elements.push(Element::BeginClip(BeginClip {
+            bbox: Default::default(),
+        }));
+        self.clip_stack.push(ClipElement {
+            bbox: None,
+            begin_ix,
+            transform: Affine::default(),
+        });
+        self.path_count += 1;
+    }
 
     fn text(&mut self) -> &mut Self::Text {
         &mut self.inner_text
@@ -149,15 +189,42 @@ impl RenderContext for PietGpuRenderContext {
     fn draw_text(&mut self, _layout: &Self::TextLayout, _pos: impl Into<Point>) {}
 
     fn save(&mut self) -> Result<(), Error> {
+        self.state_stack.push(Default::default());
         Ok(())
     }
+
     fn restore(&mut self) -> Result<(), Error> {
-        Ok(())
+        if let Some(state) = self.state_stack.pop() {
+            if state.transform != Affine::default() {
+                let a_inv = state.transform.inverse();
+                self.elements
+                    .push(Element::Transform(to_scene_transform(a_inv)));
+                self.cur_transform *= a_inv;
+            }
+            for _ in 0..state.n_clip {
+                self.pop_clip();
+            }
+            Ok(())
+        } else {
+            Err(Error::StackUnbalance)
+        }
     }
+
     fn finish(&mut self) -> Result<(), Error> {
         Ok(())
     }
-    fn transform(&mut self, _transform: Affine) {}
+
+    fn transform(&mut self, transform: Affine) {
+        self.elements
+            .push(Element::Transform(to_scene_transform(transform)));
+        if let Some(tos) = self.state_stack.last_mut() {
+            tos.transform *= transform;
+        }
+        if let Some(tos) = self.clip_stack.last_mut() {
+            tos.transform *= transform;
+        }
+        self.cur_transform *= transform;
+    }
 
     fn make_image(
         &mut self,
@@ -189,7 +256,13 @@ impl RenderContext for PietGpuRenderContext {
     fn blurred_rect(&mut self, _rect: Rect, _blur_radius: f64, _brush: &impl IntoBrush<Self>) {}
 
     fn current_transform(&self) -> Affine {
-        Default::default()
+        self.cur_transform
+    }
+
+    fn with_save(&mut self, f: impl FnOnce(&mut Self) -> Result<(), Error>) -> Result<(), Error> {
+        self.save()?;
+        // Always try to restore the stack, even if `f` errored.
+        f(self).and(self.restore())
     }
 }
 
@@ -316,6 +389,33 @@ impl PietGpuRenderContext {
             }
         }
     }
+
+    fn pop_clip(&mut self) {
+        let tos = self.clip_stack.pop().unwrap();
+        let delta = (self.elements.len() - tos.begin_ix).try_into().unwrap();
+        self.elements.push(Element::EndClip(EndClip { delta }));
+        self.path_count += 1;
+        if let Some(bbox) = tos.bbox {
+            if let Element::BeginClip(begin_clip) = &mut self.elements[tos.begin_ix] {
+                begin_clip.bbox = rect_to_f32_4(bbox);
+            } else {
+                unreachable!("expected BeginClip, not found");
+            }
+            self.accumulate_bbox(|| bbox);
+        }
+    }
+
+    fn accumulate_bbox(&mut self, f: impl FnOnce() -> Rect) {
+        if let Some(tos) = self.clip_stack.last_mut() {
+            let bbox = f();
+            let bbox = tos.transform.transform_rect_bbox(bbox);
+            tos.bbox = if let Some(old_bbox) = tos.bbox {
+                Some(old_bbox.union(bbox))
+            } else {
+                Some(bbox)
+            };
+        }
+    }
 }
 
 impl Text for PietGpuText {
@@ -410,3 +510,15 @@ impl IntoBrush<PietGpuRenderContext> for PietGpuBrush {
 fn to_f32_2(point: Point) -> [f32; 2] {
     [point.x as f32, point.y as f32]
 }
+
+fn rect_to_f32_4(rect: Rect) -> [f32; 4] {
+    [rect.x0 as f32, rect.y0 as f32, rect.x1 as f32, rect.y1 as f32]
+}
+
+fn to_scene_transform(transform: Affine) -> Transform {
+    let c = transform.as_coeffs();
+    Transform {
+        mat: [c[0] as f32, c[1] as f32, c[2] as f32, c[3] as f32],
+        translate: [c[4] as f32, c[5] as f32],
+    }
+}