diff --git a/webrender/res/brush.glsl b/webrender/res/brush.glsl
index baeebd45dc..f828e6cae1 100644
--- a/webrender/res/brush.glsl
+++ b/webrender/res/brush.glsl
@@ -91,6 +91,7 @@ void main(void) {
 #ifdef WR_FEATURE_ALPHA_PASS
     write_clip(
         vi.world_pos,
+        vi.snap_offset,
         clip_area
     );
 #endif
diff --git a/webrender/res/prim_shared.glsl b/webrender/res/prim_shared.glsl
index 42ce495d35..7dce4dde27 100644
--- a/webrender/res/prim_shared.glsl
+++ b/webrender/res/prim_shared.glsl
@@ -27,6 +27,7 @@ vec2 clamp_rect(vec2 pt, RectWithSize rect) {
 
 // TODO: convert back to RectWithEndPoint if driver issues are resolved, if ever.
 flat varying vec4 vClipMaskUvBounds;
+flat varying vec4 vClipMaskUvSampleBounds;
 // XY and W are homogeneous coordinates, Z is the layer index
 varying vec4 vClipMaskUv;
 
@@ -222,14 +223,21 @@ VertexInfo write_transform_vertex(RectWithSize local_segment_rect,
     return vi;
 }
 
-void write_clip(vec4 world_pos, ClipArea area) {
+void write_clip(vec4 world_pos, vec2 snap_offset, ClipArea area) {
     vec2 uv = world_pos.xy * uDevicePixelRatio +
-        world_pos.w * (area.common_data.task_rect.p0 - area.screen_origin);
+        world_pos.w * (snap_offset + area.common_data.task_rect.p0 - area.screen_origin);
     vClipMaskUvBounds = vec4(
         area.common_data.task_rect.p0,
         area.common_data.task_rect.p0 + area.common_data.task_rect.size
     );
+    vClipMaskUvSampleBounds.xy = vClipMaskUvBounds.xy + vec2(0.5);
+    vClipMaskUvSampleBounds.zw = vClipMaskUvBounds.zw - vec2(0.5);
     vClipMaskUv = vec4(uv, area.common_data.texture_layer_index, world_pos.w);
+
+    vec2 texture_size = vec2(textureSize(sCacheA8, 0).xy);
+    vClipMaskUv.xy /= texture_size;
+    vClipMaskUvBounds /= texture_size.xyxy;
+    vClipMaskUvSampleBounds /= texture_size.xyxy;
 }
 #endif //WR_VERTEX_SHADER
 
@@ -250,9 +258,14 @@ float do_clip() {
     if (!all(inside)) {
         return 0.0;
     }
+
     // finally, the slow path - fetch the mask value from an image
-    ivec3 tc = ivec3(mask_uv, vClipMaskUv.z);
-    return texelFetch(sCacheA8, tc, 0).r;
+
+    // TODO(gw): texelFetch here fails on some nVidia hardware in
+    //           some cases. For now, just use texture()
+    //           unconditionally.
+    mask_uv = clamp(mask_uv, vClipMaskUvSampleBounds.xy, vClipMaskUvSampleBounds.zw);
+    return texture(sCacheA8, vec3(mask_uv, vClipMaskUv.z)).r;
 }
 
 #ifdef WR_FEATURE_DITHERING
diff --git a/webrender/res/ps_split_composite.glsl b/webrender/res/ps_split_composite.glsl
index dae9677d8b..7896b66433 100644
--- a/webrender/res/ps_split_composite.glsl
+++ b/webrender/res/ps_split_composite.glsl
@@ -80,6 +80,7 @@ void main(void) {
 
     write_clip(
         world_pos,
+        vec2(0.0),
         clip_area
     );
 
diff --git a/webrender/res/ps_text_run.glsl b/webrender/res/ps_text_run.glsl
index c523a91021..468663eeed 100644
--- a/webrender/res/ps_text_run.glsl
+++ b/webrender/res/ps_text_run.glsl
@@ -229,7 +229,7 @@ void main(void) {
     vec2 f = (vi.local_pos - glyph_rect.p0) / glyph_rect.size;
 #endif
 
-    write_clip(vi.world_pos, clip_area);
+    write_clip(vi.world_pos, vi.snap_offset, clip_area);
 
     switch (color_mode) {
         case COLOR_MODE_ALPHA:
diff --git a/wrench/reftests/clip/clip-45-degree-rotation-ref.png b/wrench/reftests/clip/clip-45-degree-rotation-ref.png
index d554807aff..d6fdfa99a4 100644
Binary files a/wrench/reftests/clip/clip-45-degree-rotation-ref.png and b/wrench/reftests/clip/clip-45-degree-rotation-ref.png differ
diff --git a/wrench/reftests/gradient/linear-aligned-border-radius.png b/wrench/reftests/gradient/linear-aligned-border-radius.png
index 55518031cd..892a9606ad 100644
Binary files a/wrench/reftests/gradient/linear-aligned-border-radius.png and b/wrench/reftests/gradient/linear-aligned-border-radius.png differ
diff --git a/wrench/reftests/gradient/repeat-border-radius.png b/wrench/reftests/gradient/repeat-border-radius.png
index 10dd14b48a..47bdae159f 100644
Binary files a/wrench/reftests/gradient/repeat-border-radius.png and b/wrench/reftests/gradient/repeat-border-radius.png differ
diff --git a/wrench/reftests/transforms/rotated-clip-large.png b/wrench/reftests/transforms/rotated-clip-large.png
index 476febd29d..40a094ba73 100644
Binary files a/wrench/reftests/transforms/rotated-clip-large.png and b/wrench/reftests/transforms/rotated-clip-large.png differ
diff --git a/wrench/reftests/transforms/rotated-clip.png b/wrench/reftests/transforms/rotated-clip.png
index 4fcfcc7329..2e9c6f0c49 100644
Binary files a/wrench/reftests/transforms/rotated-clip.png and b/wrench/reftests/transforms/rotated-clip.png differ