src/hb-gpu-fragment.msl - third_party/harfbuzz - Git at Google

 /*
  * Copyright (C) 2026  Behdad Esfahbod
  *
  *  This is part of HarfBuzz, a text shaping library.
  *
  * Permission is hereby granted, without written agreement and without
  * license or royalty fees, to use, copy, modify, and distribute this
  * software and its documentation for any purpose, provided that the
  * above copyright notice and the following two paragraphs appear in
  * all copies of this software.
  *
  * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
  * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
  * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
  * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
  * DAMAGE.
  *
  * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
  * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
  * FITNESS FOR A PARTICULAR PURPOSE.  THE SOFTWARE PROVIDED HEREUNDER IS
  * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
  * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
  */


 /* Shared fragment-shader helpers for the hb-gpu renderers.
  *
  * Requires Metal Shading Language 2.0.
  */


 #ifndef HB_GPU_UNITS_PER_EM
 #define HB_GPU_UNITS_PER_EM 4
 #endif

 #define HB_GPU_INV_UNITS float(1.0 / float(HB_GPU_UNITS_PER_EM))


 int4 hb_gpu_fetch (device const short4* hb_gpu_atlas, int offset)
 {
   return int4 (hb_gpu_atlas[offset]);
 }

 uint _hb_gpu_calc_root_code (float y1, float y2, float y3)
 {
   uint i1 = as_type<uint> (y1) >> 31U;
   uint i2 = as_type<uint> (y2) >> 30U;
   uint i3 = as_type<uint> (y3) >> 29U;

   uint shift = (i2 & 2U) | (i1 & ~2U);
   shift = (i3 & 4U) | (shift & ~4U);

   return (0x2E74U >> shift) & 0x0101U;
 }

 float2 _hb_gpu_solve_horiz_poly (float2 a, float2 b, float2 p1)
 {
   float ra = 1.0 / a.y;
   float rb = 0.5 / b.y;

   float d = sqrt (max (b.y * b.y - a.y * p1.y, 0.0));
   float t1 = (b.y - d) * ra;
   float t2 = (b.y + d) * ra;

   if (a.y == 0.0)
     t1 = t2 = p1.y * rb;

   return float2 ((a.x * t1 - b.x * 2.0) * t1 + p1.x,
 		 (a.x * t2 - b.x * 2.0) * t2 + p1.x);
 }

 float2 _hb_gpu_solve_vert_poly (float2 a, float2 b, float2 p1)
 {
   float ra = 1.0 / a.x;
   float rb = 0.5 / b.x;

   float d = sqrt (max (b.x * b.x - a.x * p1.x, 0.0));
   float t1 = (b.x - d) * ra;
   float t2 = (b.x + d) * ra;

   if (a.x == 0.0)
     t1 = t2 = p1.x * rb;

   return float2 ((a.y * t1 - b.y * 2.0) * t1 + p1.y,
 		 (a.y * t2 - b.y * 2.0) * t2 + p1.y);
 }

 float _hb_gpu_calc_coverage (float xcov, float ycov, float xwgt, float ywgt)
 {
   float coverage = max (abs (xcov * xwgt + ycov * ywgt) /
 			max (xwgt + ywgt, 1.0 / 65536.0),
 			min (abs (xcov), abs (ycov)));

   return clamp (coverage, 0.0, 1.0);
 }

 /* Decoded glyph band info for a pixel position. */
 struct _hb_gpu_glyph_info
 {
   int glyphLoc;
   int bandBase;
   int2 bandIndex;
   int numHBands;
   int numVBands;
   float2 scale;
 };

 _hb_gpu_glyph_info _hb_gpu_decode_glyph (float2 renderCoord, uint glyphLoc_,
 					  device const short4* hb_gpu_atlas)
 {
   _hb_gpu_glyph_info gi;
   gi.glyphLoc = int (glyphLoc_);

   int4 header0 = hb_gpu_fetch (hb_gpu_atlas, gi.glyphLoc);
   int4 header1 = hb_gpu_fetch (hb_gpu_atlas, gi.glyphLoc + 1);
   float4 ext = float4 (header0) * HB_GPU_INV_UNITS;
   gi.numHBands = header1.r;
   gi.numVBands = header1.g;
   gi.scale = float2 (float (header1.b), float (header1.a));

   float2 extSize = ext.zw - ext.xy;
   float2 bandScale = float2 (float (gi.numVBands), float (gi.numHBands)) / max (extSize, float2 (1.0 / 65536.0));
   float2 bandOffset = -ext.xy * bandScale;

   gi.bandIndex = clamp (int2 (renderCoord * bandScale + bandOffset),
 			int2 (0, 0),
 			int2 (gi.numVBands - 1, gi.numHBands - 1));

   gi.bandBase = gi.glyphLoc + 2;
   return gi;
 }

 /* Return pixels per em at this fragment.
  *
  * renderCoord:  em-space sample position
  * glyphLoc:     texel offset of glyph blob in atlas
  */
 float hb_gpu_ppem (float2 renderCoord, uint glyphLoc_,
 		   device const short4* hb_gpu_atlas)
 {
   _hb_gpu_glyph_info gi = _hb_gpu_decode_glyph (renderCoord, glyphLoc_, hb_gpu_atlas);
   float2 emsPerPixel = fwidth (renderCoord);
   return min (gi.scale.x, gi.scale.y) /
 	 max (emsPerPixel.x, emsPerPixel.y);
 }

 /* Return per-pixel curve counts: (horizontal, vertical). */
 int2 _hb_gpu_curve_counts (float2 renderCoord, uint glyphLoc_,
 			   device const short4* hb_gpu_atlas)
 {
   _hb_gpu_glyph_info gi = _hb_gpu_decode_glyph (renderCoord, glyphLoc_, hb_gpu_atlas);
   int hCount = hb_gpu_fetch (hb_gpu_atlas, gi.bandBase + gi.bandIndex.y).r;
   int vCount = hb_gpu_fetch (hb_gpu_atlas, gi.bandBase + gi.numHBands + gi.bandIndex.x).r;
   return int2 (hCount, vCount);
 }

 /* Single-sample coverage in [0, 1]. */
 float _hb_gpu_slug_single (float2 renderCoord, float2 pixelsPerEm, uint glyphLoc_,
 			     device const short4* hb_gpu_atlas)
 {

   _hb_gpu_glyph_info gi = _hb_gpu_decode_glyph (renderCoord, glyphLoc_, hb_gpu_atlas);
   int glyphLoc = gi.glyphLoc;
   int bandBase = gi.bandBase;
   int numHBands = gi.numHBands;

   float xcov = 0.0;
   float xwgt = 0.0;

   int4 hbandData = hb_gpu_fetch (hb_gpu_atlas, bandBase + gi.bandIndex.y);
   int hCurveCount = hbandData.r;
   /* Symmetric: choose rightward (desc) or leftward (asc) sort */
   float hSplit = float (hbandData.a) * HB_GPU_INV_UNITS;
   bool hLeftRay = (renderCoord.x < hSplit);
   int hDataOffset = (hLeftRay ? hbandData.b : hbandData.g) + 32768;

   for (int ci = 0; ci < hCurveCount; ci++)
   {
     int curveOffset = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + hDataOffset + ci).r + 32768;

     int4 raw12 = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + curveOffset);
     int4 raw3 = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + curveOffset + 1);

     float4 q12 = float4 (raw12) * HB_GPU_INV_UNITS;
     float2 q3 = float2 (raw3.rg) * HB_GPU_INV_UNITS;

     float4 p12 = q12 - float4 (renderCoord, renderCoord);
     float2 p3 = q3 - renderCoord;

     if (hLeftRay) {
       if (min (min (p12.x, p12.z), p3.x) * pixelsPerEm.x > 0.5) break;
     } else {
       if (max (max (p12.x, p12.z), p3.x) * pixelsPerEm.x < -0.5) break;
     }

     uint code = _hb_gpu_calc_root_code (p12.y, p12.w, p3.y);
     if (code != 0U)
     {
       float2 a = q12.xy - q12.zw * 2.0 + q3;
       float2 b = q12.xy - q12.zw;
       float2 r = _hb_gpu_solve_horiz_poly (a, b, p12.xy) * pixelsPerEm.x;
       /* For leftward ray: saturate(0.5 - r) counts coverage from the left */
       float2 cov = hLeftRay ? clamp (float2 (0.5) - r, 0.0, 1.0)
 			    : clamp (r + float2 (0.5), 0.0, 1.0);

       if ((code & 1U) != 0U)
       {
 	xcov += cov.x;
 	xwgt = max (xwgt, clamp (1.0 - abs (r.x) * 2.0, 0.0, 1.0));
       }

       if (code > 1U)
       {
 	xcov -= cov.y;
 	xwgt = max (xwgt, clamp (1.0 - abs (r.y) * 2.0, 0.0, 1.0));
       }
     }
   }

   float ycov = 0.0;
   float ywgt = 0.0;

   int4 vbandData = hb_gpu_fetch (hb_gpu_atlas, bandBase + numHBands + gi.bandIndex.x);
   int vCurveCount = vbandData.r;
   float vSplit = float (vbandData.a) * HB_GPU_INV_UNITS;
   bool vLeftRay = (renderCoord.y < vSplit);
   int vDataOffset = (vLeftRay ? vbandData.b : vbandData.g) + 32768;

   for (int ci = 0; ci < vCurveCount; ci++)
   {
     int curveOffset = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + vDataOffset + ci).r + 32768;

     int4 raw12 = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + curveOffset);
     int4 raw3 = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + curveOffset + 1);

     float4 q12 = float4 (raw12) * HB_GPU_INV_UNITS;
     float2 q3 = float2 (raw3.rg) * HB_GPU_INV_UNITS;

     float4 p12 = q12 - float4 (renderCoord, renderCoord);
     float2 p3 = q3 - renderCoord;

     if (vLeftRay) {
       if (min (min (p12.y, p12.w), p3.y) * pixelsPerEm.y > 0.5) break;
     } else {
       if (max (max (p12.y, p12.w), p3.y) * pixelsPerEm.y < -0.5) break;
     }

     uint code = _hb_gpu_calc_root_code (p12.x, p12.z, p3.x);
     if (code != 0U)
     {
       float2 a = q12.xy - q12.zw * 2.0 + q3;
       float2 b = q12.xy - q12.zw;
       float2 r = _hb_gpu_solve_vert_poly (a, b, p12.xy) * pixelsPerEm.y;
       float2 cov = vLeftRay ? clamp (float2 (0.5) - r, 0.0, 1.0)
 			    : clamp (r + float2 (0.5), 0.0, 1.0);

       if ((code & 1U) != 0U)
       {
 	ycov -= cov.x;
 	ywgt = max (ywgt, clamp (1.0 - abs (r.x) * 2.0, 0.0, 1.0));
       }

       if (code > 1U)
       {
 	ycov += cov.y;
 	ywgt = max (ywgt, clamp (1.0 - abs (r.y) * 2.0, 0.0, 1.0));
       }
     }
   }

   return _hb_gpu_calc_coverage (xcov, ycov, xwgt, ywgt);
 }

 /* Return coverage in [0, 1].
  *
  * renderCoord:    em-space sample position
  * glyphLoc:       texel offset of glyph blob in atlas
  * hb_gpu_atlas:   device pointer to the atlas buffer
  */
 /* The MSAA-aware implementation.  Caller supplies pixelsPerEm so
  * this function can be invoked from non-uniform control flow (for
  * example from a paint op-stream branch where a recomputed fwidth
  * would be rejected by strict derivative-uniformity rules). */
 float _hb_gpu_slug (float2 renderCoord, float2 pixelsPerEm, uint glyphLoc_,
 			 device const short4* hb_gpu_atlas)
 {
   float c = _hb_gpu_slug_single (renderCoord, pixelsPerEm, glyphLoc_, hb_gpu_atlas);

 #ifndef HB_GPU_NO_MSAA
   float ppem = hb_gpu_ppem (renderCoord, glyphLoc_, hb_gpu_atlas);

   if (ppem < 16.0)
   {
     float2 emsPerPixel = 1.0 / pixelsPerEm;
     float2 d = emsPerPixel * (1.0 / 3.0);
     float msaa = 0.25 *
       (_hb_gpu_slug_single (renderCoord + float2 (-d.x, -d.y), pixelsPerEm, glyphLoc_, hb_gpu_atlas) +
        _hb_gpu_slug_single (renderCoord + float2 ( d.x, -d.y), pixelsPerEm, glyphLoc_, hb_gpu_atlas) +
        _hb_gpu_slug_single (renderCoord + float2 (-d.x,  d.y), pixelsPerEm, glyphLoc_, hb_gpu_atlas) +
        _hb_gpu_slug_single (renderCoord + float2 ( d.x,  d.y), pixelsPerEm, glyphLoc_, hb_gpu_atlas));

     c = mix (c, msaa, smoothstep (16.0, 8.0, ppem));
   }
 #endif

   return c;
 }

 /* Stem darkening for small sizes.
  *
  * coverage:    output of hb_gpu_draw
  * brightness:  foreground brightness in [0, 1]
  * ppem:        pixels per em at this fragment
  */
 float hb_gpu_stem_darken (float coverage, float brightness, float ppem)
 {
   return pow (coverage,
 	      mix (pow (2.0, brightness - 0.5), 1.0,
 		   smoothstep (8.0, 48.0, ppem)));
 }
	/*
	* Copyright (C) 2026 Behdad Esfahbod
	*
	* This is part of HarfBuzz, a text shaping library.
	*
	* Permission is hereby granted, without written agreement and without
	* license or royalty fees, to use, copy, modify, and distribute this
	* software and its documentation for any purpose, provided that the
	* above copyright notice and the following two paragraphs appear in
	* all copies of this software.
	*
	* IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
	* DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
	* ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
	* IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
	* DAMAGE.
	*
	* THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
	* BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
	* FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS
	* ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
	* PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
	*/


	/* Shared fragment-shader helpers for the hb-gpu renderers.
	*
	* Requires Metal Shading Language 2.0.
	*/


	#ifndef HB_GPU_UNITS_PER_EM
	#define HB_GPU_UNITS_PER_EM 4
	#endif

	#define HB_GPU_INV_UNITS float(1.0 / float(HB_GPU_UNITS_PER_EM))


	int4 hb_gpu_fetch (device const short4* hb_gpu_atlas, int offset)
	{
	return int4 (hb_gpu_atlas[offset]);
	}

	uint _hb_gpu_calc_root_code (float y1, float y2, float y3)
	{
	uint i1 = as_type<uint> (y1) >> 31U;
	uint i2 = as_type<uint> (y2) >> 30U;
	uint i3 = as_type<uint> (y3) >> 29U;

	uint shift = (i2 & 2U) \| (i1 & ~2U);
	shift = (i3 & 4U) \| (shift & ~4U);

	return (0x2E74U >> shift) & 0x0101U;
	}

	float2 _hb_gpu_solve_horiz_poly (float2 a, float2 b, float2 p1)
	{
	float ra = 1.0 / a.y;
	float rb = 0.5 / b.y;

	float d = sqrt (max (b.y * b.y - a.y * p1.y, 0.0));
	float t1 = (b.y - d) * ra;
	float t2 = (b.y + d) * ra;

	if (a.y == 0.0)
	t1 = t2 = p1.y * rb;

	return float2 ((a.x * t1 - b.x * 2.0) * t1 + p1.x,
	(a.x * t2 - b.x * 2.0) * t2 + p1.x);
	}

	float2 _hb_gpu_solve_vert_poly (float2 a, float2 b, float2 p1)
	{
	float ra = 1.0 / a.x;
	float rb = 0.5 / b.x;

	float d = sqrt (max (b.x * b.x - a.x * p1.x, 0.0));
	float t1 = (b.x - d) * ra;
	float t2 = (b.x + d) * ra;

	if (a.x == 0.0)
	t1 = t2 = p1.x * rb;

	return float2 ((a.y * t1 - b.y * 2.0) * t1 + p1.y,
	(a.y * t2 - b.y * 2.0) * t2 + p1.y);
	}

	float _hb_gpu_calc_coverage (float xcov, float ycov, float xwgt, float ywgt)
	{
	float coverage = max (abs (xcov * xwgt + ycov * ywgt) /
	max (xwgt + ywgt, 1.0 / 65536.0),
	min (abs (xcov), abs (ycov)));

	return clamp (coverage, 0.0, 1.0);
	}

	/* Decoded glyph band info for a pixel position. */
	struct _hb_gpu_glyph_info
	{
	int glyphLoc;
	int bandBase;
	int2 bandIndex;
	int numHBands;
	int numVBands;
	float2 scale;
	};

	_hb_gpu_glyph_info _hb_gpu_decode_glyph (float2 renderCoord, uint glyphLoc_,
	device const short4* hb_gpu_atlas)
	{
	_hb_gpu_glyph_info gi;
	gi.glyphLoc = int (glyphLoc_);

	int4 header0 = hb_gpu_fetch (hb_gpu_atlas, gi.glyphLoc);
	int4 header1 = hb_gpu_fetch (hb_gpu_atlas, gi.glyphLoc + 1);
	float4 ext = float4 (header0) * HB_GPU_INV_UNITS;
	gi.numHBands = header1.r;
	gi.numVBands = header1.g;
	gi.scale = float2 (float (header1.b), float (header1.a));

	float2 extSize = ext.zw - ext.xy;
	float2 bandScale = float2 (float (gi.numVBands), float (gi.numHBands)) / max (extSize, float2 (1.0 / 65536.0));
	float2 bandOffset = -ext.xy * bandScale;

	gi.bandIndex = clamp (int2 (renderCoord * bandScale + bandOffset),
	int2 (0, 0),
	int2 (gi.numVBands - 1, gi.numHBands - 1));

	gi.bandBase = gi.glyphLoc + 2;
	return gi;
	}

	/* Return pixels per em at this fragment.
	*
	* renderCoord: em-space sample position
	* glyphLoc: texel offset of glyph blob in atlas
	*/
	float hb_gpu_ppem (float2 renderCoord, uint glyphLoc_,
	device const short4* hb_gpu_atlas)
	{
	_hb_gpu_glyph_info gi = _hb_gpu_decode_glyph (renderCoord, glyphLoc_, hb_gpu_atlas);
	float2 emsPerPixel = fwidth (renderCoord);
	return min (gi.scale.x, gi.scale.y) /
	max (emsPerPixel.x, emsPerPixel.y);
	}

	/* Return per-pixel curve counts: (horizontal, vertical). */
	int2 _hb_gpu_curve_counts (float2 renderCoord, uint glyphLoc_,
	device const short4* hb_gpu_atlas)
	{
	_hb_gpu_glyph_info gi = _hb_gpu_decode_glyph (renderCoord, glyphLoc_, hb_gpu_atlas);
	int hCount = hb_gpu_fetch (hb_gpu_atlas, gi.bandBase + gi.bandIndex.y).r;
	int vCount = hb_gpu_fetch (hb_gpu_atlas, gi.bandBase + gi.numHBands + gi.bandIndex.x).r;
	return int2 (hCount, vCount);
	}

	/* Single-sample coverage in [0, 1]. */
	float _hb_gpu_slug_single (float2 renderCoord, float2 pixelsPerEm, uint glyphLoc_,
	device const short4* hb_gpu_atlas)
	{

	_hb_gpu_glyph_info gi = _hb_gpu_decode_glyph (renderCoord, glyphLoc_, hb_gpu_atlas);
	int glyphLoc = gi.glyphLoc;
	int bandBase = gi.bandBase;
	int numHBands = gi.numHBands;

	float xcov = 0.0;
	float xwgt = 0.0;

	int4 hbandData = hb_gpu_fetch (hb_gpu_atlas, bandBase + gi.bandIndex.y);
	int hCurveCount = hbandData.r;
	/* Symmetric: choose rightward (desc) or leftward (asc) sort */
	float hSplit = float (hbandData.a) * HB_GPU_INV_UNITS;
	bool hLeftRay = (renderCoord.x < hSplit);
	int hDataOffset = (hLeftRay ? hbandData.b : hbandData.g) + 32768;

	for (int ci = 0; ci < hCurveCount; ci++)
	{
	int curveOffset = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + hDataOffset + ci).r + 32768;

	int4 raw12 = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + curveOffset);
	int4 raw3 = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + curveOffset + 1);

	float4 q12 = float4 (raw12) * HB_GPU_INV_UNITS;
	float2 q3 = float2 (raw3.rg) * HB_GPU_INV_UNITS;

	float4 p12 = q12 - float4 (renderCoord, renderCoord);
	float2 p3 = q3 - renderCoord;

	if (hLeftRay) {
	if (min (min (p12.x, p12.z), p3.x) * pixelsPerEm.x > 0.5) break;
	} else {
	if (max (max (p12.x, p12.z), p3.x) * pixelsPerEm.x < -0.5) break;
	}

	uint code = _hb_gpu_calc_root_code (p12.y, p12.w, p3.y);
	if (code != 0U)
	{
	float2 a = q12.xy - q12.zw * 2.0 + q3;
	float2 b = q12.xy - q12.zw;
	float2 r = _hb_gpu_solve_horiz_poly (a, b, p12.xy) * pixelsPerEm.x;
	/* For leftward ray: saturate(0.5 - r) counts coverage from the left */
	float2 cov = hLeftRay ? clamp (float2 (0.5) - r, 0.0, 1.0)
	: clamp (r + float2 (0.5), 0.0, 1.0);

	if ((code & 1U) != 0U)
	{
	xcov += cov.x;
	xwgt = max (xwgt, clamp (1.0 - abs (r.x) * 2.0, 0.0, 1.0));
	}

	if (code > 1U)
	{
	xcov -= cov.y;
	xwgt = max (xwgt, clamp (1.0 - abs (r.y) * 2.0, 0.0, 1.0));
	}
	}
	}

	float ycov = 0.0;
	float ywgt = 0.0;

	int4 vbandData = hb_gpu_fetch (hb_gpu_atlas, bandBase + numHBands + gi.bandIndex.x);
	int vCurveCount = vbandData.r;
	float vSplit = float (vbandData.a) * HB_GPU_INV_UNITS;
	bool vLeftRay = (renderCoord.y < vSplit);
	int vDataOffset = (vLeftRay ? vbandData.b : vbandData.g) + 32768;

	for (int ci = 0; ci < vCurveCount; ci++)
	{
	int curveOffset = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + vDataOffset + ci).r + 32768;

	int4 raw12 = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + curveOffset);
	int4 raw3 = hb_gpu_fetch (hb_gpu_atlas, glyphLoc + curveOffset + 1);

	float4 q12 = float4 (raw12) * HB_GPU_INV_UNITS;
	float2 q3 = float2 (raw3.rg) * HB_GPU_INV_UNITS;

	float4 p12 = q12 - float4 (renderCoord, renderCoord);
	float2 p3 = q3 - renderCoord;

	if (vLeftRay) {
	if (min (min (p12.y, p12.w), p3.y) * pixelsPerEm.y > 0.5) break;
	} else {
	if (max (max (p12.y, p12.w), p3.y) * pixelsPerEm.y < -0.5) break;
	}

	uint code = _hb_gpu_calc_root_code (p12.x, p12.z, p3.x);
	if (code != 0U)
	{
	float2 a = q12.xy - q12.zw * 2.0 + q3;
	float2 b = q12.xy - q12.zw;
	float2 r = _hb_gpu_solve_vert_poly (a, b, p12.xy) * pixelsPerEm.y;
	float2 cov = vLeftRay ? clamp (float2 (0.5) - r, 0.0, 1.0)
	: clamp (r + float2 (0.5), 0.0, 1.0);

	if ((code & 1U) != 0U)
	{
	ycov -= cov.x;
	ywgt = max (ywgt, clamp (1.0 - abs (r.x) * 2.0, 0.0, 1.0));
	}

	if (code > 1U)
	{
	ycov += cov.y;
	ywgt = max (ywgt, clamp (1.0 - abs (r.y) * 2.0, 0.0, 1.0));
	}
	}
	}

	return _hb_gpu_calc_coverage (xcov, ycov, xwgt, ywgt);
	}

	/* Return coverage in [0, 1].
	*
	* renderCoord: em-space sample position
	* glyphLoc: texel offset of glyph blob in atlas
	* hb_gpu_atlas: device pointer to the atlas buffer
	*/
	/* The MSAA-aware implementation. Caller supplies pixelsPerEm so
	* this function can be invoked from non-uniform control flow (for
	* example from a paint op-stream branch where a recomputed fwidth
	* would be rejected by strict derivative-uniformity rules). */
	float _hb_gpu_slug (float2 renderCoord, float2 pixelsPerEm, uint glyphLoc_,
	device const short4* hb_gpu_atlas)
	{
	float c = _hb_gpu_slug_single (renderCoord, pixelsPerEm, glyphLoc_, hb_gpu_atlas);

	#ifndef HB_GPU_NO_MSAA
	float ppem = hb_gpu_ppem (renderCoord, glyphLoc_, hb_gpu_atlas);

	if (ppem < 16.0)
	{
	float2 emsPerPixel = 1.0 / pixelsPerEm;
	float2 d = emsPerPixel * (1.0 / 3.0);
	float msaa = 0.25 *
	(_hb_gpu_slug_single (renderCoord + float2 (-d.x, -d.y), pixelsPerEm, glyphLoc_, hb_gpu_atlas) +
	_hb_gpu_slug_single (renderCoord + float2 ( d.x, -d.y), pixelsPerEm, glyphLoc_, hb_gpu_atlas) +
	_hb_gpu_slug_single (renderCoord + float2 (-d.x, d.y), pixelsPerEm, glyphLoc_, hb_gpu_atlas) +
	_hb_gpu_slug_single (renderCoord + float2 ( d.x, d.y), pixelsPerEm, glyphLoc_, hb_gpu_atlas));

	c = mix (c, msaa, smoothstep (16.0, 8.0, ppem));
	}
	#endif

	return c;
	}

	/* Stem darkening for small sizes.
	*
	* coverage: output of hb_gpu_draw
	* brightness: foreground brightness in [0, 1]
	* ppem: pixels per em at this fragment
	*/
	float hb_gpu_stem_darken (float coverage, float brightness, float ppem)
	{
	return pow (coverage,
	mix (pow (2.0, brightness - 0.5), 1.0,
	smoothstep (8.0, 48.0, ppem)));
	}