Skip to content
This repository was archived by the owner on Feb 25, 2025. It is now read-only.

Commit ae022fa

Browse files
amaioranoamaiorano
committed
Add support for configurable subpixel precision
Bug: b/141676114 Change-Id: I47e7d90e14b44533e64d352ecc6440495c0b7d3f Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36597 Presubmit-Ready: Antonio Maiorano <amaiorano@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by: Alexis Hétu <sugoi@google.com> Tested-by: Antonio Maiorano <amaiorano@google.com>
1 parent 53096e4 commit ae022fa

File tree

9 files changed

+110
-65
lines changed

9 files changed

+110
-65
lines changed

‎src/Device/Renderer.cpp‎

Lines changed: 13 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -330,16 +330,17 @@ namespace sw
330330
float N = viewport.minDepth;
331331
float F = viewport.maxDepth;
332332
float Z = F - N;
333+
constexpr float subPixF = vk::SUBPIXEL_PRECISION_FACTOR;
333334

334335
if(context->isDrawTriangle(false))
335336
{
336337
N += context->depthBias;
337338
}
338339

339-
data->Wx16 = replicate(W * 16);
340-
data->Hx16 = replicate(H * 16);
341-
data->X0x16 = replicate(X0 * 16 - 8);
342-
data->Y0x16 = replicate(Y0 * 16 - 8);
340+
data->WxF = replicate(W * subPixF);
341+
data->HxF = replicate(H * subPixF);
342+
data->X0xF = replicate(X0 * subPixF - subPixF / 2);
343+
data->Y0xF = replicate(Y0 * subPixF - subPixF / 2);
343344
data->halfPixelX = replicate(0.5f / W);
344345
data->halfPixelY = replicate(0.5f / H);
345346
data->viewportHeight = abs(viewport.height);
@@ -799,8 +800,10 @@ namespace sw
799800
return false;
800801
}
801802

802-
const float W = data.Wx16[0] * (1.0f / 16.0f);
803-
const float H = data.Hx16[0] * (1.0f / 16.0f);
803+
constexpr float subPixF = vk::SUBPIXEL_PRECISION_FACTOR;
804+
805+
const float W = data.WxF[0] * (1.0f / subPixF);
806+
const float H = data.HxF[0] * (1.0f / subPixF);
804807

805808
float dx = W * (P1.x / P1.w - P0.x / P0.w);
806809
float dy = H * (P1.y / P1.w - P0.y / P0.w);
@@ -1029,8 +1032,10 @@ namespace sw
10291032
triangle.v1 = triangle.v0;
10301033
triangle.v2 = triangle.v0;
10311034

1032-
triangle.v1.projected.x += iround(16 * 0.5f * pSize);
1033-
triangle.v2.projected.y -= iround(16 * 0.5f * pSize) * (data.Hx16[0] > 0.0f ? 1 : -1); // Both Direct3D and OpenGL expect (0, 0) in the top-left corner
1035+
constexpr float subPixF = vk::SUBPIXEL_PRECISION_FACTOR;
1036+
1037+
triangle.v1.projected.x += iround(subPixF * 0.5f * pSize);
1038+
triangle.v2.projected.y -= iround(subPixF * 0.5f * pSize) * (data.HxF[0] > 0.0f ? 1 : -1); // Both Direct3D and OpenGL expect (0, 0) in the top-left corner
10341039
return setupRoutine(&primitive, &triangle, &polygon, &data);
10351040
}
10361041

‎src/Device/Renderer.hpp‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -79,10 +79,10 @@ namespace sw
7979
PixelProcessor::Factor factor;
8080
unsigned int occlusion[MaxClusterCount]; // Number of pixels passing depth test
8181

82-
float4 Wx16;
83-
float4 Hx16;
84-
float4 X0x16;
85-
float4 Y0x16;
82+
float4 WxF;
83+
float4 HxF;
84+
float4 X0xF;
85+
float4 Y0xF;
8686
float4 halfPixelX;
8787
float4 halfPixelY;
8888
float viewportHeight;

‎src/Pipeline/Constants.cpp‎

Lines changed: 13 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -290,29 +290,18 @@ namespace sw
290290
sRGBtoLinear12_16[i] = (unsigned short)(clamp(sw::sRGBtoLinear((float)i / 0x0FFF) * 0xFFFF + 0.5f, 0.0f, (float)0xFFFF));
291291
}
292292

293-
// VK_SAMPLE_COUNT_4_BIT
294-
// https://www.khronos.org/registry/vulkan/specs/1.1/html/vkspec.html#primsrast-multisampling
295-
constexpr float sampleLocations4[][2] = {
296-
{0.375, 0.125},
297-
{0.875, 0.375},
298-
{0.125, 0.625},
299-
{0.625, 0.875},
300-
};
301-
302-
// Vulkan spec sample positions are relative to 0,0 in top left corner, with Y+ going down.
303-
// Convert to our space, with 0,0 in center, and Y+ going up.
304293
constexpr float4 X[4] = {
305-
sw::replicate(sampleLocations4[0][0] - 0.5f), // -0.125
306-
sw::replicate(sampleLocations4[1][0] - 0.5f), // +0.375
307-
sw::replicate(sampleLocations4[2][0] - 0.5f), // -0.375
308-
sw::replicate(sampleLocations4[3][0] - 0.5f), // +0.125
294+
sw::replicate(SampleLocationsX[0]),
295+
sw::replicate(SampleLocationsX[1]),
296+
sw::replicate(SampleLocationsX[2]),
297+
sw::replicate(SampleLocationsX[3]),
309298
};
310299

311300
constexpr float4 Y[4] = {
312-
sw::replicate(-(sampleLocations4[0][1] - 0.5f)), // +0.375
313-
sw::replicate(-(sampleLocations4[1][1] - 0.5f)), // +0.125
314-
sw::replicate(-(sampleLocations4[2][1] - 0.5f)), // -0.125
315-
sw::replicate(-(sampleLocations4[3][1] - 0.5f)), // -0.375
301+
sw::replicate(SampleLocationsY[0]),
302+
sw::replicate(SampleLocationsY[1]),
303+
sw::replicate(SampleLocationsY[2]),
304+
sw::replicate(SampleLocationsY[3]),
316305
};
317306

318307
for(int q = 0; q < 4; q++)
@@ -332,8 +321,11 @@ namespace sw
332321
}
333322
}
334323

335-
const int Xf[4] = {-5, +5, +2, -2}; // Fragment offsets
336-
const int Yf[4] = {-2, +2, -5, +5}; // Fragment offsets
324+
constexpr auto subPixB = vk::SUBPIXEL_PRECISION_BITS;
325+
326+
// Reorder sample points for fragment offset computation
327+
const int Xf[4] = { toFixedPoint(X[2][0], subPixB), toFixedPoint(X[1][0], subPixB), toFixedPoint(X[3][0], subPixB), toFixedPoint(X[0][0], subPixB) };
328+
const int Yf[4] = { toFixedPoint(Y[2][0], subPixB), toFixedPoint(Y[1][0], subPixB), toFixedPoint(Y[3][0], subPixB), toFixedPoint(Y[0][0], subPixB) };
337329

338330
memcpy(&this->Xf, &Xf, sizeof(Xf));
339331
memcpy(&this->Yf, &Yf, sizeof(Yf));

‎src/Pipeline/Constants.hpp‎

Lines changed: 32 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -16,6 +16,8 @@
1616
#define sw_Constants_hpp
1717

1818
#include "System/Types.hpp"
19+
#include "System/Math.hpp"
20+
#include "Vulkan/VkConfig.h"
1921

2022
namespace sw
2123
{
@@ -86,6 +88,36 @@ namespace sw
8688
float4 X[4];
8789
float4 Y[4];
8890

91+
// VK_SAMPLE_COUNT_4_BIT
92+
// https://www.khronos.org/registry/vulkan/specs/1.1/html/vkspec.html#primsrast-multisampling
93+
static constexpr float VkSampleLocations4[][2] = {
94+
{0.375, 0.125},
95+
{0.875, 0.375},
96+
{0.125, 0.625},
97+
{0.625, 0.875},
98+
};
99+
100+
// Vulkan spec sample positions are relative to 0,0 in top left corner, with Y+ going down.
101+
// Convert to our space, with 0,0 in center, and Y+ going up.
102+
static constexpr float SampleLocationsX[4] = {
103+
VkSampleLocations4[0][0] - 0.5f,
104+
VkSampleLocations4[1][0] - 0.5f,
105+
VkSampleLocations4[2][0] - 0.5f,
106+
VkSampleLocations4[3][0] - 0.5f,
107+
};
108+
109+
static constexpr float SampleLocationsY[4] = {
110+
-(VkSampleLocations4[0][1] - 0.5f),
111+
-(VkSampleLocations4[1][1] - 0.5f),
112+
-(VkSampleLocations4[2][1] - 0.5f),
113+
-(VkSampleLocations4[3][1] - 0.5f),
114+
};
115+
116+
// Compute the yMin and yMax multisample offsets so that they are just
117+
// large enough (+/- max range - epsilon) to include sample points
118+
static constexpr int yMinMultiSampleOffset = sw::toFixedPoint(1, vk::SUBPIXEL_PRECISION_BITS) - sw::toFixedPoint(sw::max(SampleLocationsY[0], SampleLocationsY[1], SampleLocationsY[2], SampleLocationsY[3]), vk::SUBPIXEL_PRECISION_BITS) - 1;
119+
static constexpr int yMaxMultiSampleOffset = sw::toFixedPoint(1, vk::SUBPIXEL_PRECISION_BITS) + sw::toFixedPoint(sw::max(SampleLocationsY[0], SampleLocationsY[1], SampleLocationsY[2], SampleLocationsY[3]), vk::SUBPIXEL_PRECISION_BITS) - 1;
120+
89121
dword maxX[16];
90122
dword maxY[16];
91123
dword maxZ[16];

‎src/Pipeline/SetupRoutine.cpp‎

Lines changed: 30 additions & 23 deletions
Original file line numberDiff line numberDiff line change
@@ -139,8 +139,8 @@ namespace sw
139139
Float w = v.w;
140140
Float rhw = IfThenElse(w != 0.0f, 1.0f / w, Float(1.0f));
141141

142-
X[i] = RoundInt(*Pointer<Float>(data + OFFSET(DrawData,X0x16)) + v.x * rhw * *Pointer<Float>(data + OFFSET(DrawData,Wx16)));
143-
Y[i] = RoundInt(*Pointer<Float>(data + OFFSET(DrawData,Y0x16)) + v.y * rhw * *Pointer<Float>(data + OFFSET(DrawData,Hx16)));
142+
X[i] = RoundInt(*Pointer<Float>(data + OFFSET(DrawData,X0xF)) + v.x * rhw * *Pointer<Float>(data + OFFSET(DrawData,WxF)));
143+
Y[i] = RoundInt(*Pointer<Float>(data + OFFSET(DrawData,Y0xF)) + v.y * rhw * *Pointer<Float>(data + OFFSET(DrawData,HxF)));
144144

145145
i++;
146146
}
@@ -162,15 +162,19 @@ namespace sw
162162
}
163163
Until(i >= n)
164164

165+
constexpr int subPixB = vk::SUBPIXEL_PRECISION_BITS;
166+
constexpr int subPixM = vk::SUBPIXEL_PRECISION_MASK;
167+
constexpr float subPixF = vk::SUBPIXEL_PRECISION_FACTOR;
168+
165169
if(state.multiSample > 1)
166170
{
167-
yMin = (yMin + 0x0A) >> 4;
168-
yMax = (yMax + 0x14) >> 4;
171+
yMin = (yMin + Constants::yMinMultiSampleOffset) >> subPixB;
172+
yMax = (yMax + Constants::yMaxMultiSampleOffset) >> subPixB;
169173
}
170174
else
171175
{
172-
yMin = (yMin + 0x0F) >> 4;
173-
yMax = (yMax + 0x0F) >> 4;
176+
yMin = (yMin + subPixM) >> subPixB;
177+
yMax = (yMax + subPixM) >> subPixB;
174178
}
175179

176180
yMin = Max(yMin, *Pointer<Int>(data + OFFSET(DrawData,scissorY0)));
@@ -213,7 +217,7 @@ namespace sw
213217
{
214218
Int xMin = *Pointer<Int>(data + OFFSET(DrawData, scissorX0));
215219
Int xMax = *Pointer<Int>(data + OFFSET(DrawData, scissorX1));
216-
Short x = Short(Clamp((X[0] + 0xF) >> 4, xMin, xMax));
220+
Short x = Short(Clamp((X[0] + subPixM) >> subPixB, xMin, xMax));
217221

218222
For(Int y = yMin - 1, y < yMax + 1, y++)
219223
{
@@ -323,20 +327,20 @@ namespace sw
323327
Y2 = Y1 + X0 - X1;
324328
}
325329

326-
Float dx = Float(X0) * (1.0f / 16.0f);
327-
Float dy = Float(Y0) * (1.0f / 16.0f);
330+
Float dx = Float(X0) * (1.0f / subPixF);
331+
Float dy = Float(Y0) * (1.0f / subPixF);
328332

329333
X1 -= X0;
330334
Y1 -= Y0;
331335

332336
X2 -= X0;
333337
Y2 -= Y0;
334338

335-
Float x1 = w1 * (1.0f / 16.0f) * Float(X1);
336-
Float y1 = w1 * (1.0f / 16.0f) * Float(Y1);
339+
Float x1 = w1 * (1.0f / subPixF) * Float(X1);
340+
Float y1 = w1 * (1.0f / subPixF) * Float(Y1);
337341

338-
Float x2 = w2 * (1.0f / 16.0f) * Float(X2);
339-
Float y2 = w2 * (1.0f / 16.0f) * Float(Y2);
342+
Float x2 = w2 * (1.0f / subPixF) * Float(X2);
343+
Float y2 = w2 * (1.0f / subPixF) * Float(Y2);
340344

341345
Float a = x1 * y2 - x2 * y1;
342346

@@ -403,10 +407,10 @@ namespace sw
403407

404408
if(!point)
405409
{
406-
Float x1 = Float(X1) * (1.0f / 16.0f);
407-
Float y1 = Float(Y1) * (1.0f / 16.0f);
408-
Float x2 = Float(X2) * (1.0f / 16.0f);
409-
Float y2 = Float(Y2) * (1.0f / 16.0f);
410+
Float x1 = Float(X1) * (1.0f / subPixF);
411+
Float y1 = Float(Y1) * (1.0f / subPixF);
412+
Float x2 = Float(X2) * (1.0f / subPixF);
413+
Float y2 = Float(Y2) * (1.0f / subPixF);
410414

411415
Float D = *Pointer<Float>(data + OFFSET(DrawData,depthRange)) / (x1 * y2 - x2 * y1);
412416

@@ -509,8 +513,11 @@ namespace sw
509513
Int Y1 = IfThenElse(swap, Yb, Ya);
510514
Int Y2 = IfThenElse(swap, Ya, Yb);
511515

512-
Int y1 = Max((Y1 + 0x0000000F) >> 4, *Pointer<Int>(data + OFFSET(DrawData,scissorY0)));
513-
Int y2 = Min((Y2 + 0x0000000F) >> 4, *Pointer<Int>(data + OFFSET(DrawData,scissorY1)));
516+
constexpr int subPixB = vk::SUBPIXEL_PRECISION_BITS;
517+
constexpr int subPixM = vk::SUBPIXEL_PRECISION_MASK;
518+
519+
Int y1 = Max((Y1 + subPixM) >> subPixB, *Pointer<Int>(data + OFFSET(DrawData,scissorY0)));
520+
Int y2 = Min((Y2 + subPixM) >> subPixB, *Pointer<Int>(data + OFFSET(DrawData,scissorY1)));
514521

515522
If(y1 < y2)
516523
{
@@ -525,11 +532,11 @@ namespace sw
525532
Int DX12 = X2 - X1;
526533
Int DY12 = Y2 - Y1;
527534

528-
Int FDX12 = DX12 << 4;
529-
Int FDY12 = DY12 << 4;
535+
Int FDX12 = DX12 << subPixB;
536+
Int FDY12 = DY12 << subPixB;
530537

531-
Int X = DX12 * ((y1 << 4) - Y1) + (X1 & 0x0000000F) * DY12;
532-
Int x = (X1 >> 4) + X / FDY12; // Edge
538+
Int X = DX12 * ((y1 << subPixB) - Y1) + (X1 & subPixM) * DY12;
539+
Int x = (X1 >> subPixB) + X / FDY12; // Edge
533540
Int d = X % FDY12; // Error-term
534541
Int ceil = -d >> 31; // Ceiling division: remainder <= 0
535542
x -= ceil;

‎src/Pipeline/VertexRoutine.cpp‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -543,8 +543,8 @@ namespace sw
543543
Float4 rhw = Float4(1.0f) / w;
544544

545545
Vector4f proj;
546-
proj.x = As<Float4>(RoundInt(*Pointer<Float4>(data + OFFSET(DrawData,X0x16)) + pos.x * rhw * *Pointer<Float4>(data + OFFSET(DrawData,Wx16))));
547-
proj.y = As<Float4>(RoundInt(*Pointer<Float4>(data + OFFSET(DrawData,Y0x16)) + pos.y * rhw * *Pointer<Float4>(data + OFFSET(DrawData,Hx16))));
546+
proj.x = As<Float4>(RoundInt(*Pointer<Float4>(data + OFFSET(DrawData,X0xF)) + pos.x * rhw * *Pointer<Float4>(data + OFFSET(DrawData,WxF))));
547+
proj.y = As<Float4>(RoundInt(*Pointer<Float4>(data + OFFSET(DrawData,Y0xF)) + pos.y * rhw * *Pointer<Float4>(data + OFFSET(DrawData,HxF))));
548548
proj.z = pos.z * rhw;
549549
proj.w = rhw;
550550

‎src/System/Math.hpp‎

Lines changed: 11 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -32,37 +32,37 @@ namespace sw
3232
#undef max
3333

3434
template<class T>
35-
inline T max(T a, T b)
35+
inline T constexpr max(T a, T b)
3636
{
3737
return a > b ? a : b;
3838
}
3939

4040
template<class T>
41-
inline T min(T a, T b)
41+
inline constexpr T min(T a, T b)
4242
{
4343
return a < b ? a : b;
4444
}
4545

4646
template<class T>
47-
inline T max(T a, T b, T c)
47+
inline constexpr T max(T a, T b, T c)
4848
{
4949
return max(max(a, b), c);
5050
}
5151

5252
template<class T>
53-
inline T min(T a, T b, T c)
53+
inline constexpr T min(T a, T b, T c)
5454
{
5555
return min(min(a, b), c);
5656
}
5757

5858
template<class T>
59-
inline T max(T a, T b, T c, T d)
59+
inline constexpr T max(T a, T b, T c, T d)
6060
{
6161
return max(max(a, b), max(c, d));
6262
}
6363

6464
template<class T>
65-
inline T min(T a, T b, T c, T d)
65+
inline constexpr T min(T a, T b, T c, T d)
6666
{
6767
return min(min(a, b), min(c, d));
6868
}
@@ -372,6 +372,11 @@ namespace sw
372372
{
373373
return static_cast<int>(min(x, 0x7FFFFFFFu));
374374
}
375+
376+
// Convert floating value v to fixed point with p digits after the decimal point
377+
constexpr int toFixedPoint(float v, int p) {
378+
return static_cast<int>(v * (1 << p));
379+
}
375380
}
376381

377382
#endif // sw_Math_hpp

‎src/Vulkan/VkConfig.h‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -78,6 +78,10 @@ enum
7878
MAX_POINT_SIZE = 1, // Large points are not supported. If/when we turn this on, must be >= 64.
7979
};
8080

81+
constexpr int SUBPIXEL_PRECISION_BITS = 4;
82+
constexpr float SUBPIXEL_PRECISION_FACTOR = static_cast<float>(1 << SUBPIXEL_PRECISION_BITS);
83+
constexpr int SUBPIXEL_PRECISION_MASK = 0xFFFFFFFF >> (32 - SUBPIXEL_PRECISION_BITS);
84+
8185
}
8286

8387
#endif // VK_CONFIG_HPP_

‎src/Vulkan/VkPhysicalDevice.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -200,7 +200,7 @@ const VkPhysicalDeviceLimits& PhysicalDevice::getLimits() const
200200
{ 65535, 65535, 65535 }, // maxComputeWorkGroupCount[3]
201201
128, // maxComputeWorkGroupInvocations
202202
{ 128, 128, 64, }, // maxComputeWorkGroupSize[3]
203-
4, // subPixelPrecisionBits
203+
vk::SUBPIXEL_PRECISION_BITS, // subPixelPrecisionBits
204204
4, // subTexelPrecisionBits
205205
4, // mipmapPrecisionBits
206206
UINT32_MAX, // maxDrawIndexedIndexValue

0 commit comments

Comments
 (0)