1// Copyright 2016 The SwiftShader Authors. All Rights Reserved.
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7// http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15#include "PixelProgram.hpp"
16
17#include "SamplerCore.hpp"
18#include "Renderer/Primitive.hpp"
19#include "Renderer/Renderer.hpp"
20
21namespace sw
22{
23 extern bool postBlendSRGB;
24 extern bool booleanFaceRegister;
25 extern bool halfIntegerCoordinates; // Pixel centers are not at integer coordinates
26 extern bool fullPixelPositionRegister;
27
28 PixelProgram::PixelProgram(const PixelProcessor::State &state, const PixelShader *shader) :
29 PixelRoutine(state, shader),
30 r(shader->indirectAddressableTemporaries),
31 aL(shader->getLimits().loops),
32 increment(shader->getLimits().loops),
33 iteration(shader->getLimits().loops),
34 callStack(shader->getLimits().stack)
35 {
36 auto limits = shader->getLimits();
37 ifFalseBlock.resize(limits.ifs);
38 loopRepTestBlock.resize(limits.loops);
39 loopRepEndBlock.resize(limits.loops);
40 labelBlock.resize(limits.maxLabel + 1);
41 isConditionalIf.resize(limits.ifs);
42
43 loopDepth = -1;
44 enableStack[0] = Int4(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
45
46 if(shader->containsBreakInstruction())
47 {
48 enableBreak = Int4(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
49 }
50
51 if(shader->containsContinueInstruction())
52 {
53 enableContinue = Int4(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
54 }
55 }
56
57 void PixelProgram::setBuiltins(Int &x, Int &y, Float4(&z)[4], Float4 &w)
58 {
59 if(shader->getShaderModel() >= 0x0300)
60 {
61 if(shader->isVPosDeclared())
62 {
63 if(!halfIntegerCoordinates)
64 {
65 vPos.x = Float4(Float(x)) + Float4(0, 1, 0, 1);
66 vPos.y = Float4(Float(y)) + Float4(0, 0, 1, 1);
67 }
68 else
69 {
70 vPos.x = Float4(Float(x)) + Float4(0.5f, 1.5f, 0.5f, 1.5f);
71 vPos.y = Float4(Float(y)) + Float4(0.5f, 0.5f, 1.5f, 1.5f);
72 }
73
74 if(fullPixelPositionRegister)
75 {
76 vPos.z = z[0]; // FIXME: Centroid?
77 vPos.w = w; // FIXME: Centroid?
78 }
79 }
80
81 if(shader->isVFaceDeclared())
82 {
83 Float4 face = *Pointer<Float>(primitive + OFFSET(Primitive, area));
84
85 if(booleanFaceRegister)
86 {
87 face = As<Float4>(state.frontFaceCCW ? CmpNLT(face, Float4(0.0f)) : CmpLT(face, Float4(0.0f)));
88 }
89
90 vFace.x = face;
91 vFace.y = face;
92 vFace.z = face;
93 vFace.w = face;
94 }
95 }
96 }
97
98 void PixelProgram::applyShader(Int cMask[4])
99 {
100 enableIndex = 0;
101 stackIndex = 0;
102
103 if(shader->containsLeaveInstruction())
104 {
105 enableLeave = Int4(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
106 }
107
108 for(int i = 0; i < RENDERTARGETS; i++)
109 {
110 if(state.targetFormat[i] != FORMAT_NULL)
111 {
112 oC[i] = Vector4f(0.0f, 0.0f, 0.0f, 0.0f);
113 }
114 }
115
116 // Create all call site return blocks up front
117 for(size_t i = 0; i < shader->getLength(); i++)
118 {
119 const Shader::Instruction *instruction = shader->getInstruction(i);
120 Shader::Opcode opcode = instruction->opcode;
121
122 if(opcode == Shader::OPCODE_CALL || opcode == Shader::OPCODE_CALLNZ)
123 {
124 const Dst &dst = instruction->dst;
125
126 ASSERT(callRetBlock[dst.label].size() == dst.callSite);
127 callRetBlock[dst.label].push_back(Nucleus::createBasicBlock());
128 }
129 }
130
131 bool broadcastColor0 = true;
132
133 for(size_t i = 0; i < shader->getLength(); i++)
134 {
135 const Shader::Instruction *instruction = shader->getInstruction(i);
136 Shader::Opcode opcode = instruction->opcode;
137
138 if(opcode == Shader::OPCODE_DCL || opcode == Shader::OPCODE_DEF || opcode == Shader::OPCODE_DEFI || opcode == Shader::OPCODE_DEFB)
139 {
140 continue;
141 }
142
143 const Dst &dst = instruction->dst;
144 const Src &src0 = instruction->src[0];
145 const Src &src1 = instruction->src[1];
146 const Src &src2 = instruction->src[2];
147 const Src &src3 = instruction->src[3];
148 const Src &src4 = instruction->src[4];
149
150 bool predicate = instruction->predicate;
151 Control control = instruction->control;
152 bool pp = dst.partialPrecision;
153 bool project = instruction->project;
154 bool bias = instruction->bias;
155
156 Vector4f d;
157 Vector4f s0;
158 Vector4f s1;
159 Vector4f s2;
160 Vector4f s3;
161 Vector4f s4;
162
163 if(opcode == Shader::OPCODE_TEXKILL) // Takes destination as input
164 {
165 if(dst.type == Shader::PARAMETER_TEXTURE)
166 {
167 d.x = v[2 + dst.index].x;
168 d.y = v[2 + dst.index].y;
169 d.z = v[2 + dst.index].z;
170 d.w = v[2 + dst.index].w;
171 }
172 else
173 {
174 d = r[dst.index];
175 }
176 }
177
178 if(src0.type != Shader::PARAMETER_VOID) s0 = fetchRegister(src0);
179 if(src1.type != Shader::PARAMETER_VOID) s1 = fetchRegister(src1);
180 if(src2.type != Shader::PARAMETER_VOID) s2 = fetchRegister(src2);
181 if(src3.type != Shader::PARAMETER_VOID) s3 = fetchRegister(src3);
182 if(src4.type != Shader::PARAMETER_VOID) s4 = fetchRegister(src4);
183
184 switch(opcode)
185 {
186 case Shader::OPCODE_PS_2_0: break;
187 case Shader::OPCODE_PS_2_x: break;
188 case Shader::OPCODE_PS_3_0: break;
189 case Shader::OPCODE_DEF: break;
190 case Shader::OPCODE_DCL: break;
191 case Shader::OPCODE_NOP: break;
192 case Shader::OPCODE_MOV: mov(d, s0); break;
193 case Shader::OPCODE_NEG: neg(d, s0); break;
194 case Shader::OPCODE_INEG: ineg(d, s0); break;
195 case Shader::OPCODE_F2B: f2b(d, s0); break;
196 case Shader::OPCODE_B2F: b2f(d, s0); break;
197 case Shader::OPCODE_F2I: f2i(d, s0); break;
198 case Shader::OPCODE_I2F: i2f(d, s0); break;
199 case Shader::OPCODE_F2U: f2u(d, s0); break;
200 case Shader::OPCODE_U2F: u2f(d, s0); break;
201 case Shader::OPCODE_I2B: i2b(d, s0); break;
202 case Shader::OPCODE_B2I: b2i(d, s0); break;
203 case Shader::OPCODE_ADD: add(d, s0, s1); break;
204 case Shader::OPCODE_IADD: iadd(d, s0, s1); break;
205 case Shader::OPCODE_SUB: sub(d, s0, s1); break;
206 case Shader::OPCODE_ISUB: isub(d, s0, s1); break;
207 case Shader::OPCODE_MUL: mul(d, s0, s1); break;
208 case Shader::OPCODE_IMUL: imul(d, s0, s1); break;
209 case Shader::OPCODE_MAD: mad(d, s0, s1, s2); break;
210 case Shader::OPCODE_IMAD: imad(d, s0, s1, s2); break;
211 case Shader::OPCODE_DP1: dp1(d, s0, s1); break;
212 case Shader::OPCODE_DP2: dp2(d, s0, s1); break;
213 case Shader::OPCODE_DP2ADD: dp2add(d, s0, s1, s2); break;
214 case Shader::OPCODE_DP3: dp3(d, s0, s1); break;
215 case Shader::OPCODE_DP4: dp4(d, s0, s1); break;
216 case Shader::OPCODE_DET2: det2(d, s0, s1); break;
217 case Shader::OPCODE_DET3: det3(d, s0, s1, s2); break;
218 case Shader::OPCODE_DET4: det4(d, s0, s1, s2, s3); break;
219 case Shader::OPCODE_CMP0: cmp0(d, s0, s1, s2); break;
220 case Shader::OPCODE_ICMP: icmp(d, s0, s1, control); break;
221 case Shader::OPCODE_UCMP: ucmp(d, s0, s1, control); break;
222 case Shader::OPCODE_SELECT: select(d, s0, s1, s2); break;
223 case Shader::OPCODE_EXTRACT: extract(d.x, s0, s1.x); break;
224 case Shader::OPCODE_INSERT: insert(d, s0, s1.x, s2.x); break;
225 case Shader::OPCODE_FRC: frc(d, s0); break;
226 case Shader::OPCODE_TRUNC: trunc(d, s0); break;
227 case Shader::OPCODE_FLOOR: floor(d, s0); break;
228 case Shader::OPCODE_ROUND: round(d, s0); break;
229 case Shader::OPCODE_ROUNDEVEN: roundEven(d, s0); break;
230 case Shader::OPCODE_CEIL: ceil(d, s0); break;
231 case Shader::OPCODE_EXP2X: exp2x(d, s0, pp); break;
232 case Shader::OPCODE_EXP2: exp2(d, s0, pp); break;
233 case Shader::OPCODE_LOG2X: log2x(d, s0, pp); break;
234 case Shader::OPCODE_LOG2: log2(d, s0, pp); break;
235 case Shader::OPCODE_EXP: exp(d, s0, pp); break;
236 case Shader::OPCODE_LOG: log(d, s0, pp); break;
237 case Shader::OPCODE_RCPX: rcpx(d, s0, pp); break;
238 case Shader::OPCODE_DIV: div(d, s0, s1); break;
239 case Shader::OPCODE_IDIV: idiv(d, s0, s1); break;
240 case Shader::OPCODE_UDIV: udiv(d, s0, s1); break;
241 case Shader::OPCODE_MOD: mod(d, s0, s1); break;
242 case Shader::OPCODE_IMOD: imod(d, s0, s1); break;
243 case Shader::OPCODE_UMOD: umod(d, s0, s1); break;
244 case Shader::OPCODE_SHL: shl(d, s0, s1); break;
245 case Shader::OPCODE_ISHR: ishr(d, s0, s1); break;
246 case Shader::OPCODE_USHR: ushr(d, s0, s1); break;
247 case Shader::OPCODE_RSQX: rsqx(d, s0, pp); break;
248 case Shader::OPCODE_SQRT: sqrt(d, s0, pp); break;
249 case Shader::OPCODE_RSQ: rsq(d, s0, pp); break;
250 case Shader::OPCODE_LEN2: len2(d.x, s0, pp); break;
251 case Shader::OPCODE_LEN3: len3(d.x, s0, pp); break;
252 case Shader::OPCODE_LEN4: len4(d.x, s0, pp); break;
253 case Shader::OPCODE_DIST1: dist1(d.x, s0, s1, pp); break;
254 case Shader::OPCODE_DIST2: dist2(d.x, s0, s1, pp); break;
255 case Shader::OPCODE_DIST3: dist3(d.x, s0, s1, pp); break;
256 case Shader::OPCODE_DIST4: dist4(d.x, s0, s1, pp); break;
257 case Shader::OPCODE_MIN: min(d, s0, s1); break;
258 case Shader::OPCODE_IMIN: imin(d, s0, s1); break;
259 case Shader::OPCODE_UMIN: umin(d, s0, s1); break;
260 case Shader::OPCODE_MAX: max(d, s0, s1); break;
261 case Shader::OPCODE_IMAX: imax(d, s0, s1); break;
262 case Shader::OPCODE_UMAX: umax(d, s0, s1); break;
263 case Shader::OPCODE_LRP: lrp(d, s0, s1, s2); break;
264 case Shader::OPCODE_STEP: step(d, s0, s1); break;
265 case Shader::OPCODE_SMOOTH: smooth(d, s0, s1, s2); break;
266 case Shader::OPCODE_ISINF: isinf(d, s0); break;
267 case Shader::OPCODE_ISNAN: isnan(d, s0); break;
268 case Shader::OPCODE_FLOATBITSTOINT:
269 case Shader::OPCODE_FLOATBITSTOUINT:
270 case Shader::OPCODE_INTBITSTOFLOAT:
271 case Shader::OPCODE_UINTBITSTOFLOAT: d = s0; break;
272 case Shader::OPCODE_PACKSNORM2x16: packSnorm2x16(d, s0); break;
273 case Shader::OPCODE_PACKUNORM2x16: packUnorm2x16(d, s0); break;
274 case Shader::OPCODE_PACKHALF2x16: packHalf2x16(d, s0); break;
275 case Shader::OPCODE_UNPACKSNORM2x16: unpackSnorm2x16(d, s0); break;
276 case Shader::OPCODE_UNPACKUNORM2x16: unpackUnorm2x16(d, s0); break;
277 case Shader::OPCODE_UNPACKHALF2x16: unpackHalf2x16(d, s0); break;
278 case Shader::OPCODE_POWX: powx(d, s0, s1, pp); break;
279 case Shader::OPCODE_POW: pow(d, s0, s1, pp); break;
280 case Shader::OPCODE_SGN: sgn(d, s0); break;
281 case Shader::OPCODE_ISGN: isgn(d, s0); break;
282 case Shader::OPCODE_CRS: crs(d, s0, s1); break;
283 case Shader::OPCODE_FORWARD1: forward1(d, s0, s1, s2); break;
284 case Shader::OPCODE_FORWARD2: forward2(d, s0, s1, s2); break;
285 case Shader::OPCODE_FORWARD3: forward3(d, s0, s1, s2); break;
286 case Shader::OPCODE_FORWARD4: forward4(d, s0, s1, s2); break;
287 case Shader::OPCODE_REFLECT1: reflect1(d, s0, s1); break;
288 case Shader::OPCODE_REFLECT2: reflect2(d, s0, s1); break;
289 case Shader::OPCODE_REFLECT3: reflect3(d, s0, s1); break;
290 case Shader::OPCODE_REFLECT4: reflect4(d, s0, s1); break;
291 case Shader::OPCODE_REFRACT1: refract1(d, s0, s1, s2.x); break;
292 case Shader::OPCODE_REFRACT2: refract2(d, s0, s1, s2.x); break;
293 case Shader::OPCODE_REFRACT3: refract3(d, s0, s1, s2.x); break;
294 case Shader::OPCODE_REFRACT4: refract4(d, s0, s1, s2.x); break;
295 case Shader::OPCODE_NRM2: nrm2(d, s0, pp); break;
296 case Shader::OPCODE_NRM3: nrm3(d, s0, pp); break;
297 case Shader::OPCODE_NRM4: nrm4(d, s0, pp); break;
298 case Shader::OPCODE_ABS: abs(d, s0); break;
299 case Shader::OPCODE_IABS: iabs(d, s0); break;
300 case Shader::OPCODE_SINCOS: sincos(d, s0, pp); break;
301 case Shader::OPCODE_COS: cos(d, s0, pp); break;
302 case Shader::OPCODE_SIN: sin(d, s0, pp); break;
303 case Shader::OPCODE_TAN: tan(d, s0, pp); break;
304 case Shader::OPCODE_ACOS: acos(d, s0, pp); break;
305 case Shader::OPCODE_ASIN: asin(d, s0, pp); break;
306 case Shader::OPCODE_ATAN: atan(d, s0, pp); break;
307 case Shader::OPCODE_ATAN2: atan2(d, s0, s1, pp); break;
308 case Shader::OPCODE_COSH: cosh(d, s0, pp); break;
309 case Shader::OPCODE_SINH: sinh(d, s0, pp); break;
310 case Shader::OPCODE_TANH: tanh(d, s0, pp); break;
311 case Shader::OPCODE_ACOSH: acosh(d, s0, pp); break;
312 case Shader::OPCODE_ASINH: asinh(d, s0, pp); break;
313 case Shader::OPCODE_ATANH: atanh(d, s0, pp); break;
314 case Shader::OPCODE_M4X4: M4X4(d, s0, src1); break;
315 case Shader::OPCODE_M4X3: M4X3(d, s0, src1); break;
316 case Shader::OPCODE_M3X4: M3X4(d, s0, src1); break;
317 case Shader::OPCODE_M3X3: M3X3(d, s0, src1); break;
318 case Shader::OPCODE_M3X2: M3X2(d, s0, src1); break;
319 case Shader::OPCODE_TEX: TEX(d, s0, src1, project, bias); break;
320 case Shader::OPCODE_TEXLDD: TEXGRAD(d, s0, src1, s2, s3); break;
321 case Shader::OPCODE_TEXLDL: TEXLOD(d, s0, src1, s0.w); break;
322 case Shader::OPCODE_TEXLOD: TEXLOD(d, s0, src1, s2.x); break;
323 case Shader::OPCODE_TEXSIZE: TEXSIZE(d, s0.x, src1); break;
324 case Shader::OPCODE_TEXKILL: TEXKILL(cMask, d, dst.mask); break;
325 case Shader::OPCODE_TEXOFFSET: TEXOFFSET(d, s0, src1, s2); break;
326 case Shader::OPCODE_TEXLODOFFSET: TEXLODOFFSET(d, s0, src1, s2, s3.x); break;
327 case Shader::OPCODE_TEXELFETCH: TEXELFETCH(d, s0, src1, s2.x); break;
328 case Shader::OPCODE_TEXELFETCHOFFSET: TEXELFETCHOFFSET(d, s0, src1, s2, s3.x); break;
329 case Shader::OPCODE_TEXGRAD: TEXGRAD(d, s0, src1, s2, s3); break;
330 case Shader::OPCODE_TEXGRADOFFSET: TEXGRADOFFSET(d, s0, src1, s2, s3, s4); break;
331 case Shader::OPCODE_TEXBIAS: TEXBIAS(d, s0, src1, s2.x); break;
332 case Shader::OPCODE_TEXOFFSETBIAS: TEXOFFSETBIAS(d, s0, src1, s2, s3.x); break;
333 case Shader::OPCODE_DISCARD: DISCARD(cMask, instruction); break;
334 case Shader::OPCODE_DFDX: DFDX(d, s0); break;
335 case Shader::OPCODE_DFDY: DFDY(d, s0); break;
336 case Shader::OPCODE_FWIDTH: FWIDTH(d, s0); break;
337 case Shader::OPCODE_BREAK: BREAK(); break;
338 case Shader::OPCODE_BREAKC: BREAKC(s0, s1, control); break;
339 case Shader::OPCODE_BREAKP: BREAKP(src0); break;
340 case Shader::OPCODE_CONTINUE: CONTINUE(); break;
341 case Shader::OPCODE_TEST: TEST(); break;
342 case Shader::OPCODE_SCALAR: SCALAR(); break;
343 case Shader::OPCODE_CALL: CALL(dst.label, dst.callSite); break;
344 case Shader::OPCODE_CALLNZ: CALLNZ(dst.label, dst.callSite, src0); break;
345 case Shader::OPCODE_ELSE: ELSE(); break;
346 case Shader::OPCODE_ENDIF: ENDIF(); break;
347 case Shader::OPCODE_ENDLOOP: ENDLOOP(); break;
348 case Shader::OPCODE_ENDREP: ENDREP(); break;
349 case Shader::OPCODE_ENDWHILE: ENDWHILE(); break;
350 case Shader::OPCODE_ENDSWITCH: ENDSWITCH(); break;
351 case Shader::OPCODE_IF: IF(src0); break;
352 case Shader::OPCODE_IFC: IFC(s0, s1, control); break;
353 case Shader::OPCODE_LABEL: LABEL(dst.index); break;
354 case Shader::OPCODE_LOOP: LOOP(src1); break;
355 case Shader::OPCODE_REP: REP(src0); break;
356 case Shader::OPCODE_WHILE: WHILE(src0); break;
357 case Shader::OPCODE_SWITCH: SWITCH(); break;
358 case Shader::OPCODE_RET: RET(); break;
359 case Shader::OPCODE_LEAVE: LEAVE(); break;
360 case Shader::OPCODE_CMP: cmp(d, s0, s1, control); break;
361 case Shader::OPCODE_ALL: all(d.x, s0); break;
362 case Shader::OPCODE_ANY: any(d.x, s0); break;
363 case Shader::OPCODE_NOT: bitwise_not(d, s0); break;
364 case Shader::OPCODE_OR: bitwise_or(d, s0, s1); break;
365 case Shader::OPCODE_XOR: bitwise_xor(d, s0, s1); break;
366 case Shader::OPCODE_AND: bitwise_and(d, s0, s1); break;
367 case Shader::OPCODE_EQ: equal(d, s0, s1); break;
368 case Shader::OPCODE_NE: notEqual(d, s0, s1); break;
369 case Shader::OPCODE_END: break;
370 default:
371 ASSERT(false);
372 }
373
374 if(dst.type != Shader::PARAMETER_VOID && dst.type != Shader::PARAMETER_LABEL && opcode != Shader::OPCODE_TEXKILL && opcode != Shader::OPCODE_NOP)
375 {
376 if(dst.saturate)
377 {
378 if(dst.x) d.x = Max(d.x, Float4(0.0f));
379 if(dst.y) d.y = Max(d.y, Float4(0.0f));
380 if(dst.z) d.z = Max(d.z, Float4(0.0f));
381 if(dst.w) d.w = Max(d.w, Float4(0.0f));
382
383 if(dst.x) d.x = Min(d.x, Float4(1.0f));
384 if(dst.y) d.y = Min(d.y, Float4(1.0f));
385 if(dst.z) d.z = Min(d.z, Float4(1.0f));
386 if(dst.w) d.w = Min(d.w, Float4(1.0f));
387 }
388
389 if(instruction->isPredicated())
390 {
391 Vector4f pDst; // FIXME: Rename
392
393 switch(dst.type)
394 {
395 case Shader::PARAMETER_TEMP:
396 if(dst.rel.type == Shader::PARAMETER_VOID)
397 {
398 if(dst.x) pDst.x = r[dst.index].x;
399 if(dst.y) pDst.y = r[dst.index].y;
400 if(dst.z) pDst.z = r[dst.index].z;
401 if(dst.w) pDst.w = r[dst.index].w;
402 }
403 else if(!dst.rel.dynamic)
404 {
405 Int a = dst.index + relativeAddress(dst.rel);
406
407 if(dst.x) pDst.x = r[a].x;
408 if(dst.y) pDst.y = r[a].y;
409 if(dst.z) pDst.z = r[a].z;
410 if(dst.w) pDst.w = r[a].w;
411 }
412 else
413 {
414 Int4 a = dst.index + dynamicAddress(dst.rel);
415
416 if(dst.x) pDst.x = r[a].x;
417 if(dst.y) pDst.y = r[a].y;
418 if(dst.z) pDst.z = r[a].z;
419 if(dst.w) pDst.w = r[a].w;
420 }
421 break;
422 case Shader::PARAMETER_COLOROUT:
423 if(dst.rel.type == Shader::PARAMETER_VOID)
424 {
425 if(dst.x) pDst.x = oC[dst.index].x;
426 if(dst.y) pDst.y = oC[dst.index].y;
427 if(dst.z) pDst.z = oC[dst.index].z;
428 if(dst.w) pDst.w = oC[dst.index].w;
429 }
430 else if(!dst.rel.dynamic)
431 {
432 Int a = dst.index + relativeAddress(dst.rel);
433
434 if(dst.x) pDst.x = oC[a].x;
435 if(dst.y) pDst.y = oC[a].y;
436 if(dst.z) pDst.z = oC[a].z;
437 if(dst.w) pDst.w = oC[a].w;
438 }
439 else
440 {
441 Int4 a = dst.index + dynamicAddress(dst.rel);
442
443 if(dst.x) pDst.x = oC[a].x;
444 if(dst.y) pDst.y = oC[a].y;
445 if(dst.z) pDst.z = oC[a].z;
446 if(dst.w) pDst.w = oC[a].w;
447 }
448 break;
449 case Shader::PARAMETER_PREDICATE:
450 if(dst.x) pDst.x = p0.x;
451 if(dst.y) pDst.y = p0.y;
452 if(dst.z) pDst.z = p0.z;
453 if(dst.w) pDst.w = p0.w;
454 break;
455 case Shader::PARAMETER_DEPTHOUT:
456 pDst.x = oDepth;
457 break;
458 default:
459 ASSERT(false);
460 }
461
462 Int4 enable = enableMask(instruction);
463
464 Int4 xEnable = enable;
465 Int4 yEnable = enable;
466 Int4 zEnable = enable;
467 Int4 wEnable = enable;
468
469 if(predicate)
470 {
471 unsigned char pSwizzle = instruction->predicateSwizzle;
472
473 Float4 xPredicate = p0[(pSwizzle >> 0) & 0x03];
474 Float4 yPredicate = p0[(pSwizzle >> 2) & 0x03];
475 Float4 zPredicate = p0[(pSwizzle >> 4) & 0x03];
476 Float4 wPredicate = p0[(pSwizzle >> 6) & 0x03];
477
478 if(!instruction->predicateNot)
479 {
480 if(dst.x) xEnable = xEnable & As<Int4>(xPredicate);
481 if(dst.y) yEnable = yEnable & As<Int4>(yPredicate);
482 if(dst.z) zEnable = zEnable & As<Int4>(zPredicate);
483 if(dst.w) wEnable = wEnable & As<Int4>(wPredicate);
484 }
485 else
486 {
487 if(dst.x) xEnable = xEnable & ~As<Int4>(xPredicate);
488 if(dst.y) yEnable = yEnable & ~As<Int4>(yPredicate);
489 if(dst.z) zEnable = zEnable & ~As<Int4>(zPredicate);
490 if(dst.w) wEnable = wEnable & ~As<Int4>(wPredicate);
491 }
492 }
493
494 if(dst.x) d.x = As<Float4>(As<Int4>(d.x) & xEnable);
495 if(dst.y) d.y = As<Float4>(As<Int4>(d.y) & yEnable);
496 if(dst.z) d.z = As<Float4>(As<Int4>(d.z) & zEnable);
497 if(dst.w) d.w = As<Float4>(As<Int4>(d.w) & wEnable);
498
499 if(dst.x) d.x = As<Float4>(As<Int4>(d.x) | (As<Int4>(pDst.x) & ~xEnable));
500 if(dst.y) d.y = As<Float4>(As<Int4>(d.y) | (As<Int4>(pDst.y) & ~yEnable));
501 if(dst.z) d.z = As<Float4>(As<Int4>(d.z) | (As<Int4>(pDst.z) & ~zEnable));
502 if(dst.w) d.w = As<Float4>(As<Int4>(d.w) | (As<Int4>(pDst.w) & ~wEnable));
503 }
504
505 switch(dst.type)
506 {
507 case Shader::PARAMETER_TEMP:
508 if(dst.rel.type == Shader::PARAMETER_VOID)
509 {
510 if(dst.x) r[dst.index].x = d.x;
511 if(dst.y) r[dst.index].y = d.y;
512 if(dst.z) r[dst.index].z = d.z;
513 if(dst.w) r[dst.index].w = d.w;
514 }
515 else if(!dst.rel.dynamic)
516 {
517 Int a = dst.index + relativeAddress(dst.rel);
518
519 if(dst.x) r[a].x = d.x;
520 if(dst.y) r[a].y = d.y;
521 if(dst.z) r[a].z = d.z;
522 if(dst.w) r[a].w = d.w;
523 }
524 else
525 {
526 Int4 a = dst.index + dynamicAddress(dst.rel);
527
528 if(dst.x) r.scatter_x(a, d.x);
529 if(dst.y) r.scatter_y(a, d.y);
530 if(dst.z) r.scatter_z(a, d.z);
531 if(dst.w) r.scatter_w(a, d.w);
532 }
533 break;
534 case Shader::PARAMETER_COLOROUT:
535 if(dst.rel.type == Shader::PARAMETER_VOID)
536 {
537 broadcastColor0 = (dst.index == 0) && broadcastColor0;
538
539 if(dst.x) oC[dst.index].x = d.x;
540 if(dst.y) oC[dst.index].y = d.y;
541 if(dst.z) oC[dst.index].z = d.z;
542 if(dst.w) oC[dst.index].w = d.w;
543 }
544 else if(!dst.rel.dynamic)
545 {
546 broadcastColor0 = false;
547 Int a = dst.index + relativeAddress(dst.rel);
548
549 if(dst.x) oC[a].x = d.x;
550 if(dst.y) oC[a].y = d.y;
551 if(dst.z) oC[a].z = d.z;
552 if(dst.w) oC[a].w = d.w;
553 }
554 else
555 {
556 broadcastColor0 = false;
557 Int4 a = dst.index + dynamicAddress(dst.rel);
558
559 if(dst.x) oC.scatter_x(a, d.x);
560 if(dst.y) oC.scatter_y(a, d.y);
561 if(dst.z) oC.scatter_z(a, d.z);
562 if(dst.w) oC.scatter_w(a, d.w);
563 }
564 break;
565 case Shader::PARAMETER_PREDICATE:
566 if(dst.x) p0.x = d.x;
567 if(dst.y) p0.y = d.y;
568 if(dst.z) p0.z = d.z;
569 if(dst.w) p0.w = d.w;
570 break;
571 case Shader::PARAMETER_DEPTHOUT:
572 oDepth = d.x;
573 break;
574 default:
575 ASSERT(false);
576 }
577 }
578 }
579
580 if(currentLabel != -1)
581 {
582 Nucleus::setInsertBlock(returnBlock);
583 }
584
585 if(broadcastColor0)
586 {
587 for(int i = 0; i < RENDERTARGETS; i++)
588 {
589 c[i] = oC[0];
590 }
591 }
592 else
593 {
594 for(int i = 0; i < RENDERTARGETS; i++)
595 {
596 c[i] = oC[i];
597 }
598 }
599
600 clampColor(c);
601
602 if(state.depthOverride)
603 {
604 oDepth = Min(Max(oDepth, Float4(0.0f)), Float4(1.0f));
605 }
606 }
607
608 Bool PixelProgram::alphaTest(Int cMask[4])
609 {
610 if(!state.alphaTestActive())
611 {
612 return true;
613 }
614
615 Int aMask;
616
617 if(state.transparencyAntialiasing == TRANSPARENCY_NONE)
618 {
619 Short4 alpha = RoundShort4(c[0].w * Float4(0x1000));
620
621 PixelRoutine::alphaTest(aMask, alpha);
622
623 for(unsigned int q = 0; q < state.multiSample; q++)
624 {
625 cMask[q] &= aMask;
626 }
627 }
628 else if(state.transparencyAntialiasing == TRANSPARENCY_ALPHA_TO_COVERAGE)
629 {
630 alphaToCoverage(cMask, c[0].w);
631 }
632 else ASSERT(false);
633
634 Int pass = cMask[0];
635
636 for(unsigned int q = 1; q < state.multiSample; q++)
637 {
638 pass = pass | cMask[q];
639 }
640
641 return pass != 0x0;
642 }
643
644 void PixelProgram::rasterOperation(Float4 &fog, Pointer<Byte> cBuffer[4], Int &x, Int sMask[4], Int zMask[4], Int cMask[4])
645 {
646 for(int index = 0; index < RENDERTARGETS; index++)
647 {
648 if(!state.colorWriteActive(index))
649 {
650 continue;
651 }
652
653 if(!postBlendSRGB && state.writeSRGB && !isSRGB(index))
654 {
655 c[index].x = linearToSRGB(c[index].x);
656 c[index].y = linearToSRGB(c[index].y);
657 c[index].z = linearToSRGB(c[index].z);
658 }
659
660 if(index == 0)
661 {
662 fogBlend(c[index], fog);
663 }
664
665 switch(state.targetFormat[index])
666 {
667 case FORMAT_R5G6B5:
668 case FORMAT_X8R8G8B8:
669 case FORMAT_X8B8G8R8:
670 case FORMAT_A8R8G8B8:
671 case FORMAT_A8B8G8R8:
672 case FORMAT_SRGB8_X8:
673 case FORMAT_SRGB8_A8:
674 case FORMAT_G8R8:
675 case FORMAT_R8:
676 case FORMAT_A8:
677 case FORMAT_G16R16:
678 case FORMAT_A16B16G16R16:
679 for(unsigned int q = 0; q < state.multiSample; q++)
680 {
681 Pointer<Byte> buffer = cBuffer[index] + q * *Pointer<Int>(data + OFFSET(DrawData, colorSliceB[index]));
682 Vector4s color;
683
684 if(state.targetFormat[index] == FORMAT_R5G6B5)
685 {
686 color.x = UShort4(c[index].x * Float4(0xFBFF), false);
687 color.y = UShort4(c[index].y * Float4(0xFDFF), false);
688 color.z = UShort4(c[index].z * Float4(0xFBFF), false);
689 color.w = UShort4(c[index].w * Float4(0xFFFF), false);
690 }
691 else
692 {
693 color.x = convertFixed16(c[index].x, false);
694 color.y = convertFixed16(c[index].y, false);
695 color.z = convertFixed16(c[index].z, false);
696 color.w = convertFixed16(c[index].w, false);
697 }
698
699 if(state.multiSampleMask & (1 << q))
700 {
701 alphaBlend(index, buffer, color, x);
702 logicOperation(index, buffer, color, x);
703 writeColor(index, buffer, x, color, sMask[q], zMask[q], cMask[q]);
704 }
705 }
706 break;
707 case FORMAT_R32F:
708 case FORMAT_G32R32F:
709 case FORMAT_X32B32G32R32F:
710 case FORMAT_A32B32G32R32F:
711 case FORMAT_X32B32G32R32F_UNSIGNED:
712 case FORMAT_R32I:
713 case FORMAT_G32R32I:
714 case FORMAT_A32B32G32R32I:
715 case FORMAT_R32UI:
716 case FORMAT_G32R32UI:
717 case FORMAT_A32B32G32R32UI:
718 case FORMAT_R16I:
719 case FORMAT_G16R16I:
720 case FORMAT_A16B16G16R16I:
721 case FORMAT_R16UI:
722 case FORMAT_G16R16UI:
723 case FORMAT_A16B16G16R16UI:
724 case FORMAT_R8I:
725 case FORMAT_G8R8I:
726 case FORMAT_A8B8G8R8I:
727 case FORMAT_R8UI:
728 case FORMAT_G8R8UI:
729 case FORMAT_A8B8G8R8UI:
730 for(unsigned int q = 0; q < state.multiSample; q++)
731 {
732 Pointer<Byte> buffer = cBuffer[index] + q * *Pointer<Int>(data + OFFSET(DrawData, colorSliceB[index]));
733 Vector4f color = c[index];
734
735 if(state.multiSampleMask & (1 << q))
736 {
737 alphaBlend(index, buffer, color, x);
738 writeColor(index, buffer, x, color, sMask[q], zMask[q], cMask[q]);
739 }
740 }
741 break;
742 default:
743 ASSERT(false);
744 }
745 }
746 }
747
748 Vector4f PixelProgram::sampleTexture(const Src &sampler, Vector4f &uvwq, Float4 &bias, Vector4f &dsx, Vector4f &dsy, Vector4f &offset, SamplerFunction function)
749 {
750 Vector4f tmp;
751
752 if(sampler.type == Shader::PARAMETER_SAMPLER && sampler.rel.type == Shader::PARAMETER_VOID)
753 {
754 tmp = sampleTexture(sampler.index, uvwq, bias, dsx, dsy, offset, function);
755 }
756 else
757 {
758 Int index = As<Int>(Float(fetchRegister(sampler).x.x));
759
760 for(int i = 0; i < TEXTURE_IMAGE_UNITS; i++)
761 {
762 if(shader->usesSampler(i))
763 {
764 If(index == i)
765 {
766 tmp = sampleTexture(i, uvwq, bias, dsx, dsy, offset, function);
767 // FIXME: When the sampler states are the same, we could use one sampler and just index the texture
768 }
769 }
770 }
771 }
772
773 Vector4f c;
774 c.x = tmp[(sampler.swizzle >> 0) & 0x3];
775 c.y = tmp[(sampler.swizzle >> 2) & 0x3];
776 c.z = tmp[(sampler.swizzle >> 4) & 0x3];
777 c.w = tmp[(sampler.swizzle >> 6) & 0x3];
778
779 return c;
780 }
781
782 Vector4f PixelProgram::sampleTexture(int samplerIndex, Vector4f &uvwq, Float4 &bias, Vector4f &dsx, Vector4f &dsy, Vector4f &offset, SamplerFunction function)
783 {
784 #if PERF_PROFILE
785 Long texTime = Ticks();
786 #endif
787
788 Pointer<Byte> texture = data + OFFSET(DrawData, mipmap) + samplerIndex * sizeof(Texture);
789 Vector4f c = SamplerCore(constants, state.sampler[samplerIndex]).sampleTexture(texture, uvwq.x, uvwq.y, uvwq.z, uvwq.w, bias, dsx, dsy, offset, function);
790
791 #if PERF_PROFILE
792 cycles[PERF_TEX] += Ticks() - texTime;
793 #endif
794
795 return c;
796 }
797
798 void PixelProgram::clampColor(Vector4f oC[RENDERTARGETS])
799 {
800 for(int index = 0; index < RENDERTARGETS; index++)
801 {
802 if(!state.colorWriteActive(index) && !(index == 0 && state.alphaTestActive()))
803 {
804 continue;
805 }
806
807 switch(state.targetFormat[index])
808 {
809 case FORMAT_NULL:
810 break;
811 case FORMAT_R5G6B5:
812 case FORMAT_A8R8G8B8:
813 case FORMAT_A8B8G8R8:
814 case FORMAT_X8R8G8B8:
815 case FORMAT_X8B8G8R8:
816 case FORMAT_SRGB8_X8:
817 case FORMAT_SRGB8_A8:
818 case FORMAT_G8R8:
819 case FORMAT_R8:
820 case FORMAT_A8:
821 case FORMAT_G16R16:
822 case FORMAT_A16B16G16R16:
823 oC[index].x = Max(oC[index].x, Float4(0.0f)); oC[index].x = Min(oC[index].x, Float4(1.0f));
824 oC[index].y = Max(oC[index].y, Float4(0.0f)); oC[index].y = Min(oC[index].y, Float4(1.0f));
825 oC[index].z = Max(oC[index].z, Float4(0.0f)); oC[index].z = Min(oC[index].z, Float4(1.0f));
826 oC[index].w = Max(oC[index].w, Float4(0.0f)); oC[index].w = Min(oC[index].w, Float4(1.0f));
827 break;
828 case FORMAT_R32F:
829 case FORMAT_G32R32F:
830 case FORMAT_X32B32G32R32F:
831 case FORMAT_A32B32G32R32F:
832 case FORMAT_R32I:
833 case FORMAT_G32R32I:
834 case FORMAT_A32B32G32R32I:
835 case FORMAT_R32UI:
836 case FORMAT_G32R32UI:
837 case FORMAT_A32B32G32R32UI:
838 case FORMAT_R16I:
839 case FORMAT_G16R16I:
840 case FORMAT_A16B16G16R16I:
841 case FORMAT_R16UI:
842 case FORMAT_G16R16UI:
843 case FORMAT_A16B16G16R16UI:
844 case FORMAT_R8I:
845 case FORMAT_G8R8I:
846 case FORMAT_A8B8G8R8I:
847 case FORMAT_R8UI:
848 case FORMAT_G8R8UI:
849 case FORMAT_A8B8G8R8UI:
850 break;
851 case FORMAT_X32B32G32R32F_UNSIGNED:
852 oC[index].x = Max(oC[index].x, Float4(0.0f));
853 oC[index].y = Max(oC[index].y, Float4(0.0f));
854 oC[index].z = Max(oC[index].z, Float4(0.0f));
855 oC[index].w = Max(oC[index].w, Float4(0.0f));
856 break;
857 default:
858 ASSERT(false);
859 }
860 }
861 }
862
863 Int4 PixelProgram::enableMask(const Shader::Instruction *instruction)
864 {
865 if(scalar)
866 {
867 return Int4(0xFFFFFFFF);
868 }
869
870 Int4 enable = instruction->analysisBranch ? Int4(enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))]) : Int4(0xFFFFFFFF);
871
872 if(shader->containsBreakInstruction() && instruction->analysisBreak)
873 {
874 enable &= enableBreak;
875 }
876
877 if(shader->containsContinueInstruction() && instruction->analysisContinue)
878 {
879 enable &= enableContinue;
880 }
881
882 if(shader->containsLeaveInstruction() && instruction->analysisLeave)
883 {
884 enable &= enableLeave;
885 }
886
887 return enable;
888 }
889
890 Vector4f PixelProgram::fetchRegister(const Src &src, unsigned int offset)
891 {
892 Vector4f reg;
893 unsigned int i = src.index + offset;
894
895 switch(src.type)
896 {
897 case Shader::PARAMETER_TEMP:
898 if(src.rel.type == Shader::PARAMETER_VOID)
899 {
900 reg = r[i];
901 }
902 else if(!src.rel.dynamic)
903 {
904 reg = r[i + relativeAddress(src.rel, src.bufferIndex)];
905 }
906 else
907 {
908 reg = r[i + dynamicAddress(src.rel)];
909 }
910 break;
911 case Shader::PARAMETER_INPUT:
912 if(src.rel.type == Shader::PARAMETER_VOID) // Not relative
913 {
914 reg = v[i];
915 }
916 else if(!src.rel.dynamic)
917 {
918 reg = v[i + relativeAddress(src.rel, src.bufferIndex)];
919 }
920 else
921 {
922 reg = v[i + dynamicAddress(src.rel)];
923 }
924 break;
925 case Shader::PARAMETER_CONST:
926 reg = readConstant(src, offset);
927 break;
928 case Shader::PARAMETER_TEXTURE:
929 reg = v[2 + i];
930 break;
931 case Shader::PARAMETER_MISCTYPE:
932 if(src.index == Shader::VPosIndex) reg = vPos;
933 if(src.index == Shader::VFaceIndex) reg = vFace;
934 break;
935 case Shader::PARAMETER_SAMPLER:
936 if(src.rel.type == Shader::PARAMETER_VOID)
937 {
938 reg.x = As<Float4>(Int4(i));
939 }
940 else if(src.rel.type == Shader::PARAMETER_TEMP)
941 {
942 reg.x = As<Float4>(Int4(i) + As<Int4>(r[src.rel.index].x));
943 }
944 return reg;
945 case Shader::PARAMETER_PREDICATE: return reg; // Dummy
946 case Shader::PARAMETER_VOID: return reg; // Dummy
947 case Shader::PARAMETER_FLOAT4LITERAL:
948 // This is used for all literal types, and since Reactor doesn't guarantee
949 // preserving the bit pattern of float constants, we must construct them
950 // as integer constants and bitcast.
951 reg.x = As<Float4>(Int4(src.integer[0]));
952 reg.y = As<Float4>(Int4(src.integer[1]));
953 reg.z = As<Float4>(Int4(src.integer[2]));
954 reg.w = As<Float4>(Int4(src.integer[3]));
955 break;
956 case Shader::PARAMETER_CONSTINT: return reg; // Dummy
957 case Shader::PARAMETER_CONSTBOOL: return reg; // Dummy
958 case Shader::PARAMETER_LOOP: return reg; // Dummy
959 case Shader::PARAMETER_COLOROUT:
960 if(src.rel.type == Shader::PARAMETER_VOID) // Not relative
961 {
962 reg = oC[i];
963 }
964 else if(!src.rel.dynamic)
965 {
966 reg = oC[i + relativeAddress(src.rel, src.bufferIndex)];
967 }
968 else
969 {
970 reg = oC[i + dynamicAddress(src.rel)];
971 }
972 break;
973 case Shader::PARAMETER_DEPTHOUT:
974 reg.x = oDepth;
975 break;
976 default:
977 ASSERT(false);
978 }
979
980 const Float4 &x = reg[(src.swizzle >> 0) & 0x3];
981 const Float4 &y = reg[(src.swizzle >> 2) & 0x3];
982 const Float4 &z = reg[(src.swizzle >> 4) & 0x3];
983 const Float4 &w = reg[(src.swizzle >> 6) & 0x3];
984
985 Vector4f mod;
986
987 switch(src.modifier)
988 {
989 case Shader::MODIFIER_NONE:
990 mod.x = x;
991 mod.y = y;
992 mod.z = z;
993 mod.w = w;
994 break;
995 case Shader::MODIFIER_NEGATE:
996 mod.x = -x;
997 mod.y = -y;
998 mod.z = -z;
999 mod.w = -w;
1000 break;
1001 case Shader::MODIFIER_ABS:
1002 mod.x = Abs(x);
1003 mod.y = Abs(y);
1004 mod.z = Abs(z);
1005 mod.w = Abs(w);
1006 break;
1007 case Shader::MODIFIER_ABS_NEGATE:
1008 mod.x = -Abs(x);
1009 mod.y = -Abs(y);
1010 mod.z = -Abs(z);
1011 mod.w = -Abs(w);
1012 break;
1013 case Shader::MODIFIER_NOT:
1014 mod.x = As<Float4>(As<Int4>(x) ^ Int4(0xFFFFFFFF));
1015 mod.y = As<Float4>(As<Int4>(y) ^ Int4(0xFFFFFFFF));
1016 mod.z = As<Float4>(As<Int4>(z) ^ Int4(0xFFFFFFFF));
1017 mod.w = As<Float4>(As<Int4>(w) ^ Int4(0xFFFFFFFF));
1018 break;
1019 default:
1020 ASSERT(false);
1021 }
1022
1023 return mod;
1024 }
1025
1026 RValue<Pointer<Byte>> PixelProgram::uniformAddress(int bufferIndex, unsigned int index)
1027 {
1028 if(bufferIndex == -1)
1029 {
1030 return data + OFFSET(DrawData, ps.c[index]);
1031 }
1032 else
1033 {
1034 return *Pointer<Pointer<Byte>>(data + OFFSET(DrawData, ps.u[bufferIndex])) + index;
1035 }
1036 }
1037
1038 RValue<Pointer<Byte>> PixelProgram::uniformAddress(int bufferIndex, unsigned int index, Int& offset)
1039 {
1040 return uniformAddress(bufferIndex, index) + offset * sizeof(float4);
1041 }
1042
1043 Vector4f PixelProgram::readConstant(const Src &src, unsigned int offset)
1044 {
1045 Vector4f c;
1046 unsigned int i = src.index + offset;
1047
1048 if(src.rel.type == Shader::PARAMETER_VOID) // Not relative
1049 {
1050 c.x = c.y = c.z = c.w = *Pointer<Float4>(uniformAddress(src.bufferIndex, i));
1051
1052 c.x = c.x.xxxx;
1053 c.y = c.y.yyyy;
1054 c.z = c.z.zzzz;
1055 c.w = c.w.wwww;
1056
1057 if(shader->containsDefineInstruction()) // Constant may be known at compile time
1058 {
1059 for(size_t j = 0; j < shader->getLength(); j++)
1060 {
1061 const Shader::Instruction &instruction = *shader->getInstruction(j);
1062
1063 if(instruction.opcode == Shader::OPCODE_DEF)
1064 {
1065 if(instruction.dst.index == i)
1066 {
1067 c.x = Float4(instruction.src[0].value[0]);
1068 c.y = Float4(instruction.src[0].value[1]);
1069 c.z = Float4(instruction.src[0].value[2]);
1070 c.w = Float4(instruction.src[0].value[3]);
1071
1072 break;
1073 }
1074 }
1075 }
1076 }
1077 }
1078 else if(!src.rel.dynamic || src.rel.type == Shader::PARAMETER_LOOP)
1079 {
1080 Int a = relativeAddress(src.rel, src.bufferIndex);
1081
1082 c.x = c.y = c.z = c.w = *Pointer<Float4>(uniformAddress(src.bufferIndex, i, a));
1083
1084 c.x = c.x.xxxx;
1085 c.y = c.y.yyyy;
1086 c.z = c.z.zzzz;
1087 c.w = c.w.wwww;
1088 }
1089 else
1090 {
1091 int component = src.rel.swizzle & 0x03;
1092 Float4 a;
1093
1094 switch(src.rel.type)
1095 {
1096 case Shader::PARAMETER_TEMP: a = r[src.rel.index][component]; break;
1097 case Shader::PARAMETER_INPUT: a = v[src.rel.index][component]; break;
1098 case Shader::PARAMETER_OUTPUT: a = oC[src.rel.index][component]; break;
1099 case Shader::PARAMETER_CONST: a = *Pointer<Float>(uniformAddress(src.bufferIndex, src.rel.index) + component * sizeof(float)); break;
1100 case Shader::PARAMETER_MISCTYPE:
1101 switch(src.rel.index)
1102 {
1103 case Shader::VPosIndex: a = vPos.x; break;
1104 case Shader::VFaceIndex: a = vFace.x; break;
1105 default: ASSERT(false);
1106 }
1107 break;
1108 default: ASSERT(false);
1109 }
1110
1111 Int4 index = Int4(i) + As<Int4>(a) * Int4(src.rel.scale);
1112
1113 index = Min(As<UInt4>(index), UInt4(VERTEX_UNIFORM_VECTORS)); // Clamp to constant register range, c[VERTEX_UNIFORM_VECTORS] = {0, 0, 0, 0}
1114
1115 Int index0 = Extract(index, 0);
1116 Int index1 = Extract(index, 1);
1117 Int index2 = Extract(index, 2);
1118 Int index3 = Extract(index, 3);
1119
1120 c.x = *Pointer<Float4>(uniformAddress(src.bufferIndex, 0, index0), 16);
1121 c.y = *Pointer<Float4>(uniformAddress(src.bufferIndex, 0, index1), 16);
1122 c.z = *Pointer<Float4>(uniformAddress(src.bufferIndex, 0, index2), 16);
1123 c.w = *Pointer<Float4>(uniformAddress(src.bufferIndex, 0, index3), 16);
1124
1125 transpose4x4(c.x, c.y, c.z, c.w);
1126 }
1127
1128 return c;
1129 }
1130
1131 Int PixelProgram::relativeAddress(const Shader::Relative &rel, int bufferIndex)
1132 {
1133 ASSERT(!rel.dynamic);
1134
1135 if(rel.type == Shader::PARAMETER_TEMP)
1136 {
1137 return As<Int>(Extract(r[rel.index].x, 0)) * rel.scale;
1138 }
1139 else if(rel.type == Shader::PARAMETER_INPUT)
1140 {
1141 return As<Int>(Extract(v[rel.index].x, 0)) * rel.scale;
1142 }
1143 else if(rel.type == Shader::PARAMETER_OUTPUT)
1144 {
1145 return As<Int>(Extract(oC[rel.index].x, 0)) * rel.scale;
1146 }
1147 else if(rel.type == Shader::PARAMETER_CONST)
1148 {
1149 return *Pointer<Int>(uniformAddress(bufferIndex, rel.index)) * rel.scale;
1150 }
1151 else if(rel.type == Shader::PARAMETER_LOOP)
1152 {
1153 return aL[loopDepth];
1154 }
1155 else ASSERT(false);
1156
1157 return 0;
1158 }
1159
1160 Int4 PixelProgram::dynamicAddress(const Shader::Relative &rel)
1161 {
1162 int component = rel.swizzle & 0x03;
1163 Float4 a;
1164
1165 switch(rel.type)
1166 {
1167 case Shader::PARAMETER_TEMP: a = r[rel.index][component]; break;
1168 case Shader::PARAMETER_INPUT: a = v[rel.index][component]; break;
1169 case Shader::PARAMETER_OUTPUT: a = oC[rel.index][component]; break;
1170 case Shader::PARAMETER_MISCTYPE:
1171 switch(rel.index)
1172 {
1173 case Shader::VPosIndex: a = vPos.x; break;
1174 case Shader::VFaceIndex: a = vFace.x; break;
1175 default: ASSERT(false);
1176 }
1177 break;
1178 default: ASSERT(false);
1179 }
1180
1181 return As<Int4>(a) * Int4(rel.scale);
1182 }
1183
1184 Float4 PixelProgram::linearToSRGB(const Float4 &x) // Approximates x^(1.0/2.2)
1185 {
1186 Float4 sqrtx = Rcp_pp(RcpSqrt_pp(x));
1187 Float4 sRGB = sqrtx * Float4(1.14f) - x * Float4(0.14f);
1188
1189 return Min(Max(sRGB, Float4(0.0f)), Float4(1.0f));
1190 }
1191
1192 void PixelProgram::M3X2(Vector4f &dst, Vector4f &src0, const Src &src1)
1193 {
1194 Vector4f row0 = fetchRegister(src1, 0);
1195 Vector4f row1 = fetchRegister(src1, 1);
1196
1197 dst.x = dot3(src0, row0);
1198 dst.y = dot3(src0, row1);
1199 }
1200
1201 void PixelProgram::M3X3(Vector4f &dst, Vector4f &src0, const Src &src1)
1202 {
1203 Vector4f row0 = fetchRegister(src1, 0);
1204 Vector4f row1 = fetchRegister(src1, 1);
1205 Vector4f row2 = fetchRegister(src1, 2);
1206
1207 dst.x = dot3(src0, row0);
1208 dst.y = dot3(src0, row1);
1209 dst.z = dot3(src0, row2);
1210 }
1211
1212 void PixelProgram::M3X4(Vector4f &dst, Vector4f &src0, const Src &src1)
1213 {
1214 Vector4f row0 = fetchRegister(src1, 0);
1215 Vector4f row1 = fetchRegister(src1, 1);
1216 Vector4f row2 = fetchRegister(src1, 2);
1217 Vector4f row3 = fetchRegister(src1, 3);
1218
1219 dst.x = dot3(src0, row0);
1220 dst.y = dot3(src0, row1);
1221 dst.z = dot3(src0, row2);
1222 dst.w = dot3(src0, row3);
1223 }
1224
1225 void PixelProgram::M4X3(Vector4f &dst, Vector4f &src0, const Src &src1)
1226 {
1227 Vector4f row0 = fetchRegister(src1, 0);
1228 Vector4f row1 = fetchRegister(src1, 1);
1229 Vector4f row2 = fetchRegister(src1, 2);
1230
1231 dst.x = dot4(src0, row0);
1232 dst.y = dot4(src0, row1);
1233 dst.z = dot4(src0, row2);
1234 }
1235
1236 void PixelProgram::M4X4(Vector4f &dst, Vector4f &src0, const Src &src1)
1237 {
1238 Vector4f row0 = fetchRegister(src1, 0);
1239 Vector4f row1 = fetchRegister(src1, 1);
1240 Vector4f row2 = fetchRegister(src1, 2);
1241 Vector4f row3 = fetchRegister(src1, 3);
1242
1243 dst.x = dot4(src0, row0);
1244 dst.y = dot4(src0, row1);
1245 dst.z = dot4(src0, row2);
1246 dst.w = dot4(src0, row3);
1247 }
1248
1249 void PixelProgram::TEX(Vector4f &dst, Vector4f &src0, const Src &src1, bool project, bool bias)
1250 {
1251 if(project)
1252 {
1253 Vector4f proj;
1254 Float4 rw = reciprocal(src0.w);
1255 proj.x = src0.x * rw;
1256 proj.y = src0.y * rw;
1257 proj.z = src0.z * rw;
1258
1259 dst = sampleTexture(src1, proj, src0.x, (src0), (src0), (src0), Implicit);
1260 }
1261 else
1262 {
1263 dst = sampleTexture(src1, src0, src0.x, (src0), (src0), (src0), bias ? Bias : Implicit);
1264 }
1265 }
1266
1267 void PixelProgram::TEXOFFSET(Vector4f &dst, Vector4f &src0, const Src &src1, Vector4f &offset)
1268 {
1269 dst = sampleTexture(src1, src0, (src0.x), (src0), (src0), offset, {Implicit, Offset});
1270 }
1271
1272 void PixelProgram::TEXLODOFFSET(Vector4f &dst, Vector4f &src0, const Src &src1, Vector4f &offset, Float4 &lod)
1273 {
1274 dst = sampleTexture(src1, src0, lod, (src0), (src0), offset, {Lod, Offset});
1275 }
1276
1277 void PixelProgram::TEXBIAS(Vector4f &dst, Vector4f &src0, const Src &src1, Float4 &bias)
1278 {
1279 dst = sampleTexture(src1, src0, bias, (src0), (src0), (src0), Bias);
1280 }
1281
1282 void PixelProgram::TEXOFFSETBIAS(Vector4f &dst, Vector4f &src0, const Src &src1, Vector4f &offset, Float4 &bias)
1283 {
1284 dst = sampleTexture(src1, src0, bias, (src0), (src0), offset, {Bias, Offset});
1285 }
1286
1287 void PixelProgram::TEXELFETCH(Vector4f &dst, Vector4f &src0, const Src& src1, Float4 &lod)
1288 {
1289 dst = sampleTexture(src1, src0, lod, (src0), (src0), (src0), Fetch);
1290 }
1291
1292 void PixelProgram::TEXELFETCHOFFSET(Vector4f &dst, Vector4f &src0, const Src& src1, Vector4f &offset, Float4 &lod)
1293 {
1294 dst = sampleTexture(src1, src0, lod, (src0), (src0), offset, {Fetch, Offset});
1295 }
1296
1297 void PixelProgram::TEXGRAD(Vector4f &dst, Vector4f &src0, const Src& src1, Vector4f &dsx, Vector4f &dsy)
1298 {
1299 dst = sampleTexture(src1, src0, (src0.x), dsx, dsy, (src0), Grad);
1300 }
1301
1302 void PixelProgram::TEXGRADOFFSET(Vector4f &dst, Vector4f &src0, const Src& src1, Vector4f &dsx, Vector4f &dsy, Vector4f &offset)
1303 {
1304 dst = sampleTexture(src1, src0, (src0.x), dsx, dsy, offset, {Grad, Offset});
1305 }
1306
1307 void PixelProgram::TEXLOD(Vector4f &dst, Vector4f &src0, const Src &src1, Float4 &lod)
1308 {
1309 dst = sampleTexture(src1, src0, lod, (src0), (src0), (src0), Lod);
1310 }
1311
1312 void PixelProgram::TEXSIZE(Vector4f &dst, Float4 &lod, const Src &src1)
1313 {
1314 bool uniformSampler = (src1.type == Shader::PARAMETER_SAMPLER && src1.rel.type == Shader::PARAMETER_VOID);
1315 Int offset = uniformSampler ? src1.index * sizeof(Texture) : As<Int>(Float(fetchRegister(src1).x.x)) * sizeof(Texture);
1316 Pointer<Byte> texture = data + OFFSET(DrawData, mipmap) + offset;
1317
1318 dst = SamplerCore::textureSize(texture, lod);
1319 }
1320
1321 void PixelProgram::TEXKILL(Int cMask[4], Vector4f &src, unsigned char mask)
1322 {
1323 Int kill = -1;
1324
1325 if(mask & 0x1) kill &= SignMask(CmpNLT(src.x, Float4(0.0f)));
1326 if(mask & 0x2) kill &= SignMask(CmpNLT(src.y, Float4(0.0f)));
1327 if(mask & 0x4) kill &= SignMask(CmpNLT(src.z, Float4(0.0f)));
1328 if(mask & 0x8) kill &= SignMask(CmpNLT(src.w, Float4(0.0f)));
1329
1330 // FIXME: Dynamic branching affects TEXKILL?
1331 // if(shader->containsDynamicBranching())
1332 // {
1333 // kill = ~SignMask(enableMask());
1334 // }
1335
1336 for(unsigned int q = 0; q < state.multiSample; q++)
1337 {
1338 cMask[q] &= kill;
1339 }
1340
1341 // FIXME: Branch to end of shader if all killed?
1342 }
1343
1344 void PixelProgram::DISCARD(Int cMask[4], const Shader::Instruction *instruction)
1345 {
1346 Int kill = 0;
1347
1348 if(shader->containsDynamicBranching())
1349 {
1350 kill = ~SignMask(enableMask(instruction));
1351 }
1352
1353 for(unsigned int q = 0; q < state.multiSample; q++)
1354 {
1355 cMask[q] &= kill;
1356 }
1357
1358 // FIXME: Branch to end of shader if all killed?
1359 }
1360
1361 void PixelProgram::DFDX(Vector4f &dst, Vector4f &src)
1362 {
1363 dst.x = src.x.yyww - src.x.xxzz;
1364 dst.y = src.y.yyww - src.y.xxzz;
1365 dst.z = src.z.yyww - src.z.xxzz;
1366 dst.w = src.w.yyww - src.w.xxzz;
1367 }
1368
1369 void PixelProgram::DFDY(Vector4f &dst, Vector4f &src)
1370 {
1371 dst.x = src.x.zwzw - src.x.xyxy;
1372 dst.y = src.y.zwzw - src.y.xyxy;
1373 dst.z = src.z.zwzw - src.z.xyxy;
1374 dst.w = src.w.zwzw - src.w.xyxy;
1375 }
1376
1377 void PixelProgram::FWIDTH(Vector4f &dst, Vector4f &src)
1378 {
1379 // abs(dFdx(src)) + abs(dFdy(src));
1380 dst.x = Abs(src.x.yyww - src.x.xxzz) + Abs(src.x.zwzw - src.x.xyxy);
1381 dst.y = Abs(src.y.yyww - src.y.xxzz) + Abs(src.y.zwzw - src.y.xyxy);
1382 dst.z = Abs(src.z.yyww - src.z.xxzz) + Abs(src.z.zwzw - src.z.xyxy);
1383 dst.w = Abs(src.w.yyww - src.w.xxzz) + Abs(src.w.zwzw - src.w.xyxy);
1384 }
1385
1386 void PixelProgram::BREAK()
1387 {
1388 enableBreak = enableBreak & ~enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))];
1389 }
1390
1391 void PixelProgram::BREAKC(Vector4f &src0, Vector4f &src1, Control control)
1392 {
1393 Int4 condition;
1394
1395 switch(control)
1396 {
1397 case Shader::CONTROL_GT: condition = CmpNLE(src0.x, src1.x); break;
1398 case Shader::CONTROL_EQ: condition = CmpEQ(src0.x, src1.x); break;
1399 case Shader::CONTROL_GE: condition = CmpNLT(src0.x, src1.x); break;
1400 case Shader::CONTROL_LT: condition = CmpLT(src0.x, src1.x); break;
1401 case Shader::CONTROL_NE: condition = CmpNEQ(src0.x, src1.x); break;
1402 case Shader::CONTROL_LE: condition = CmpLE(src0.x, src1.x); break;
1403 default:
1404 ASSERT(false);
1405 }
1406
1407 BREAK(condition);
1408 }
1409
1410 void PixelProgram::BREAKP(const Src &predicateRegister) // FIXME: Factor out parts common with BREAKC
1411 {
1412 Int4 condition = As<Int4>(p0[predicateRegister.swizzle & 0x3]);
1413
1414 if(predicateRegister.modifier == Shader::MODIFIER_NOT)
1415 {
1416 condition = ~condition;
1417 }
1418
1419 BREAK(condition);
1420 }
1421
1422 void PixelProgram::BREAK(Int4 &condition)
1423 {
1424 condition &= enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))];
1425
1426 enableBreak = enableBreak & ~condition;
1427 }
1428
1429 void PixelProgram::CONTINUE()
1430 {
1431 enableContinue = enableContinue & ~enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))];
1432 }
1433
1434 void PixelProgram::TEST()
1435 {
1436 enableContinue = restoreContinue.back();
1437 restoreContinue.pop_back();
1438 }
1439
1440 void PixelProgram::SCALAR()
1441 {
1442 scalar = true;
1443 }
1444
1445 void PixelProgram::CALL(int labelIndex, int callSiteIndex)
1446 {
1447 if(!labelBlock[labelIndex])
1448 {
1449 labelBlock[labelIndex] = Nucleus::createBasicBlock();
1450 }
1451
1452 if(callRetBlock[labelIndex].size() > 1)
1453 {
1454 callStack[stackIndex++] = UInt(callSiteIndex);
1455 }
1456
1457 Int4 restoreLeave = enableLeave;
1458
1459 Nucleus::createBr(labelBlock[labelIndex]);
1460 Nucleus::setInsertBlock(callRetBlock[labelIndex][callSiteIndex]);
1461
1462 enableLeave = restoreLeave;
1463 }
1464
1465 void PixelProgram::CALLNZ(int labelIndex, int callSiteIndex, const Src &src)
1466 {
1467 if(src.type == Shader::PARAMETER_CONSTBOOL)
1468 {
1469 CALLNZb(labelIndex, callSiteIndex, src);
1470 }
1471 else if(src.type == Shader::PARAMETER_PREDICATE)
1472 {
1473 CALLNZp(labelIndex, callSiteIndex, src);
1474 }
1475 else ASSERT(false);
1476 }
1477
1478 void PixelProgram::CALLNZb(int labelIndex, int callSiteIndex, const Src &boolRegister)
1479 {
1480 Bool condition = (*Pointer<Byte>(data + OFFSET(DrawData, ps.b[boolRegister.index])) != Byte(0)); // FIXME
1481
1482 if(boolRegister.modifier == Shader::MODIFIER_NOT)
1483 {
1484 condition = !condition;
1485 }
1486
1487 if(!labelBlock[labelIndex])
1488 {
1489 labelBlock[labelIndex] = Nucleus::createBasicBlock();
1490 }
1491
1492 if(callRetBlock[labelIndex].size() > 1)
1493 {
1494 callStack[stackIndex++] = UInt(callSiteIndex);
1495 }
1496
1497 Int4 restoreLeave = enableLeave;
1498
1499 branch(condition, labelBlock[labelIndex], callRetBlock[labelIndex][callSiteIndex]);
1500 Nucleus::setInsertBlock(callRetBlock[labelIndex][callSiteIndex]);
1501
1502 enableLeave = restoreLeave;
1503 }
1504
1505 void PixelProgram::CALLNZp(int labelIndex, int callSiteIndex, const Src &predicateRegister)
1506 {
1507 Int4 condition = As<Int4>(p0[predicateRegister.swizzle & 0x3]);
1508
1509 if(predicateRegister.modifier == Shader::MODIFIER_NOT)
1510 {
1511 condition = ~condition;
1512 }
1513
1514 condition &= enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))];
1515
1516 if(!labelBlock[labelIndex])
1517 {
1518 labelBlock[labelIndex] = Nucleus::createBasicBlock();
1519 }
1520
1521 if(callRetBlock[labelIndex].size() > 1)
1522 {
1523 callStack[stackIndex++] = UInt(callSiteIndex);
1524 }
1525
1526 enableIndex++;
1527 enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))] = condition;
1528 Int4 restoreLeave = enableLeave;
1529
1530 Bool notAllFalse = SignMask(condition) != 0;
1531 branch(notAllFalse, labelBlock[labelIndex], callRetBlock[labelIndex][callSiteIndex]);
1532 Nucleus::setInsertBlock(callRetBlock[labelIndex][callSiteIndex]);
1533
1534 enableIndex--;
1535 enableLeave = restoreLeave;
1536 }
1537
1538 void PixelProgram::ELSE()
1539 {
1540 ifDepth--;
1541
1542 BasicBlock *falseBlock = ifFalseBlock[ifDepth];
1543 BasicBlock *endBlock = Nucleus::createBasicBlock();
1544
1545 if(isConditionalIf[ifDepth])
1546 {
1547 Int4 condition = ~enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))] & enableStack[Min(enableIndex - 1, Int(MAX_SHADER_ENABLE_STACK_SIZE))];
1548 Bool notAllFalse = SignMask(condition) != 0;
1549
1550 branch(notAllFalse, falseBlock, endBlock);
1551
1552 enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))] = ~enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))] & enableStack[Min(enableIndex - 1, Int(MAX_SHADER_ENABLE_STACK_SIZE))];
1553 }
1554 else
1555 {
1556 Nucleus::createBr(endBlock);
1557 Nucleus::setInsertBlock(falseBlock);
1558 }
1559
1560 ifFalseBlock[ifDepth] = endBlock;
1561
1562 ifDepth++;
1563 }
1564
1565 void PixelProgram::ENDIF()
1566 {
1567 ifDepth--;
1568
1569 BasicBlock *endBlock = ifFalseBlock[ifDepth];
1570
1571 Nucleus::createBr(endBlock);
1572 Nucleus::setInsertBlock(endBlock);
1573
1574 if(isConditionalIf[ifDepth])
1575 {
1576 enableIndex--;
1577 }
1578 }
1579
1580 void PixelProgram::ENDLOOP()
1581 {
1582 loopRepDepth--;
1583
1584 aL[loopDepth] = aL[loopDepth] + increment[loopDepth]; // FIXME: +=
1585
1586 BasicBlock *testBlock = loopRepTestBlock[loopRepDepth];
1587 BasicBlock *endBlock = loopRepEndBlock[loopRepDepth];
1588
1589 Nucleus::createBr(testBlock);
1590 Nucleus::setInsertBlock(endBlock);
1591
1592 loopDepth--;
1593 enableBreak = Int4(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
1594 }
1595
1596 void PixelProgram::ENDREP()
1597 {
1598 loopRepDepth--;
1599
1600 BasicBlock *testBlock = loopRepTestBlock[loopRepDepth];
1601 BasicBlock *endBlock = loopRepEndBlock[loopRepDepth];
1602
1603 Nucleus::createBr(testBlock);
1604 Nucleus::setInsertBlock(endBlock);
1605
1606 loopDepth--;
1607 enableBreak = Int4(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
1608 }
1609
1610 void PixelProgram::ENDWHILE()
1611 {
1612 loopRepDepth--;
1613
1614 BasicBlock *testBlock = loopRepTestBlock[loopRepDepth];
1615 BasicBlock *endBlock = loopRepEndBlock[loopRepDepth];
1616
1617 Nucleus::createBr(testBlock);
1618 Nucleus::setInsertBlock(endBlock);
1619
1620 enableIndex--;
1621 scalar = false;
1622 }
1623
1624 void PixelProgram::ENDSWITCH()
1625 {
1626 loopRepDepth--;
1627
1628 BasicBlock *endBlock = loopRepEndBlock[loopRepDepth];
1629
1630 Nucleus::createBr(endBlock);
1631 Nucleus::setInsertBlock(endBlock);
1632 }
1633
1634 void PixelProgram::IF(const Src &src)
1635 {
1636 if(src.type == Shader::PARAMETER_CONSTBOOL)
1637 {
1638 IFb(src);
1639 }
1640 else if(src.type == Shader::PARAMETER_PREDICATE)
1641 {
1642 IFp(src);
1643 }
1644 else
1645 {
1646 Int4 condition = As<Int4>(fetchRegister(src).x);
1647 IF(condition);
1648 }
1649 }
1650
1651 void PixelProgram::IFb(const Src &boolRegister)
1652 {
1653 ASSERT(ifDepth < 24 + 4);
1654
1655 Bool condition = (*Pointer<Byte>(data + OFFSET(DrawData, ps.b[boolRegister.index])) != Byte(0)); // FIXME
1656
1657 if(boolRegister.modifier == Shader::MODIFIER_NOT)
1658 {
1659 condition = !condition;
1660 }
1661
1662 BasicBlock *trueBlock = Nucleus::createBasicBlock();
1663 BasicBlock *falseBlock = Nucleus::createBasicBlock();
1664
1665 branch(condition, trueBlock, falseBlock);
1666
1667 isConditionalIf[ifDepth] = false;
1668 ifFalseBlock[ifDepth] = falseBlock;
1669
1670 ifDepth++;
1671 }
1672
1673 void PixelProgram::IFp(const Src &predicateRegister)
1674 {
1675 Int4 condition = As<Int4>(p0[predicateRegister.swizzle & 0x3]);
1676
1677 if(predicateRegister.modifier == Shader::MODIFIER_NOT)
1678 {
1679 condition = ~condition;
1680 }
1681
1682 IF(condition);
1683 }
1684
1685 void PixelProgram::IFC(Vector4f &src0, Vector4f &src1, Control control)
1686 {
1687 Int4 condition;
1688
1689 switch(control)
1690 {
1691 case Shader::CONTROL_GT: condition = CmpNLE(src0.x, src1.x); break;
1692 case Shader::CONTROL_EQ: condition = CmpEQ(src0.x, src1.x); break;
1693 case Shader::CONTROL_GE: condition = CmpNLT(src0.x, src1.x); break;
1694 case Shader::CONTROL_LT: condition = CmpLT(src0.x, src1.x); break;
1695 case Shader::CONTROL_NE: condition = CmpNEQ(src0.x, src1.x); break;
1696 case Shader::CONTROL_LE: condition = CmpLE(src0.x, src1.x); break;
1697 default:
1698 ASSERT(false);
1699 }
1700
1701 IF(condition);
1702 }
1703
1704 void PixelProgram::IF(Int4 &condition)
1705 {
1706 condition &= enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))];
1707
1708 enableIndex++;
1709 enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))] = condition;
1710
1711 BasicBlock *trueBlock = Nucleus::createBasicBlock();
1712 BasicBlock *falseBlock = Nucleus::createBasicBlock();
1713
1714 Bool notAllFalse = SignMask(condition) != 0;
1715
1716 branch(notAllFalse, trueBlock, falseBlock);
1717
1718 isConditionalIf[ifDepth] = true;
1719 ifFalseBlock[ifDepth] = falseBlock;
1720
1721 ifDepth++;
1722 }
1723
1724 void PixelProgram::LABEL(int labelIndex)
1725 {
1726 if(!labelBlock[labelIndex])
1727 {
1728 labelBlock[labelIndex] = Nucleus::createBasicBlock();
1729 }
1730
1731 Nucleus::setInsertBlock(labelBlock[labelIndex]);
1732 currentLabel = labelIndex;
1733 }
1734
1735 void PixelProgram::LOOP(const Src &integerRegister)
1736 {
1737 loopDepth++;
1738
1739 iteration[loopDepth] = *Pointer<Int>(data + OFFSET(DrawData, ps.i[integerRegister.index][0]));
1740 aL[loopDepth] = *Pointer<Int>(data + OFFSET(DrawData, ps.i[integerRegister.index][1]));
1741 increment[loopDepth] = *Pointer<Int>(data + OFFSET(DrawData, ps.i[integerRegister.index][2]));
1742
1743 // If(increment[loopDepth] == 0)
1744 // {
1745 // increment[loopDepth] = 1;
1746 // }
1747
1748 BasicBlock *loopBlock = Nucleus::createBasicBlock();
1749 BasicBlock *testBlock = Nucleus::createBasicBlock();
1750 BasicBlock *endBlock = Nucleus::createBasicBlock();
1751
1752 loopRepTestBlock[loopRepDepth] = testBlock;
1753 loopRepEndBlock[loopRepDepth] = endBlock;
1754
1755 // FIXME: jump(testBlock)
1756 Nucleus::createBr(testBlock);
1757 Nucleus::setInsertBlock(testBlock);
1758
1759 branch(iteration[loopDepth] > 0, loopBlock, endBlock);
1760 Nucleus::setInsertBlock(loopBlock);
1761
1762 iteration[loopDepth] = iteration[loopDepth] - 1; // FIXME: --
1763
1764 loopRepDepth++;
1765 }
1766
1767 void PixelProgram::REP(const Src &integerRegister)
1768 {
1769 loopDepth++;
1770
1771 iteration[loopDepth] = *Pointer<Int>(data + OFFSET(DrawData, ps.i[integerRegister.index][0]));
1772 aL[loopDepth] = aL[loopDepth - 1];
1773
1774 BasicBlock *loopBlock = Nucleus::createBasicBlock();
1775 BasicBlock *testBlock = Nucleus::createBasicBlock();
1776 BasicBlock *endBlock = Nucleus::createBasicBlock();
1777
1778 loopRepTestBlock[loopRepDepth] = testBlock;
1779 loopRepEndBlock[loopRepDepth] = endBlock;
1780
1781 // FIXME: jump(testBlock)
1782 Nucleus::createBr(testBlock);
1783 Nucleus::setInsertBlock(testBlock);
1784
1785 branch(iteration[loopDepth] > 0, loopBlock, endBlock);
1786 Nucleus::setInsertBlock(loopBlock);
1787
1788 iteration[loopDepth] = iteration[loopDepth] - 1; // FIXME: --
1789
1790 loopRepDepth++;
1791 }
1792
1793 void PixelProgram::WHILE(const Src &temporaryRegister)
1794 {
1795 enableIndex++;
1796
1797 BasicBlock *loopBlock = Nucleus::createBasicBlock();
1798 BasicBlock *testBlock = Nucleus::createBasicBlock();
1799 BasicBlock *endBlock = Nucleus::createBasicBlock();
1800
1801 loopRepTestBlock[loopRepDepth] = testBlock;
1802 loopRepEndBlock[loopRepDepth] = endBlock;
1803
1804 Int4 restoreBreak = enableBreak;
1805 restoreContinue.push_back(enableContinue);
1806
1807 // TODO: jump(testBlock)
1808 Nucleus::createBr(testBlock);
1809 Nucleus::setInsertBlock(testBlock);
1810
1811 const Vector4f &src = fetchRegister(temporaryRegister);
1812 Int4 condition = As<Int4>(src.x);
1813 condition &= enableStack[Min(enableIndex - 1, Int(MAX_SHADER_ENABLE_STACK_SIZE))];
1814 if(shader->containsLeaveInstruction()) condition &= enableLeave;
1815 if(shader->containsBreakInstruction()) condition &= enableBreak;
1816 enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))] = condition;
1817
1818 Bool notAllFalse = SignMask(condition) != 0;
1819 branch(notAllFalse, loopBlock, endBlock);
1820
1821 Nucleus::setInsertBlock(endBlock);
1822 enableBreak = restoreBreak;
1823
1824 Nucleus::setInsertBlock(loopBlock);
1825
1826 loopRepDepth++;
1827 scalar = false;
1828 }
1829
1830 void PixelProgram::SWITCH()
1831 {
1832 BasicBlock *endBlock = Nucleus::createBasicBlock();
1833
1834 loopRepTestBlock[loopRepDepth] = nullptr;
1835 loopRepEndBlock[loopRepDepth] = endBlock;
1836
1837 Int4 restoreBreak = enableBreak;
1838
1839 BasicBlock *currentBlock = Nucleus::getInsertBlock();
1840
1841 Nucleus::setInsertBlock(endBlock);
1842 enableBreak = restoreBreak;
1843
1844 Nucleus::setInsertBlock(currentBlock);
1845
1846 loopRepDepth++;
1847 }
1848
1849 void PixelProgram::RET()
1850 {
1851 if(currentLabel == -1)
1852 {
1853 returnBlock = Nucleus::createBasicBlock();
1854 Nucleus::createBr(returnBlock);
1855 }
1856 else
1857 {
1858 BasicBlock *unreachableBlock = Nucleus::createBasicBlock();
1859
1860 if(callRetBlock[currentLabel].size() > 1) // Pop the return destination from the call stack
1861 {
1862 // FIXME: Encapsulate
1863 UInt index = callStack[--stackIndex];
1864
1865 Value *value = index.loadValue();
1866 SwitchCases *switchCases = Nucleus::createSwitch(value, unreachableBlock, (int)callRetBlock[currentLabel].size());
1867
1868 for(unsigned int i = 0; i < callRetBlock[currentLabel].size(); i++)
1869 {
1870 Nucleus::addSwitchCase(switchCases, i, callRetBlock[currentLabel][i]);
1871 }
1872 }
1873 else if(callRetBlock[currentLabel].size() == 1) // Jump directly to the unique return destination
1874 {
1875 Nucleus::createBr(callRetBlock[currentLabel][0]);
1876 }
1877 else // Function isn't called
1878 {
1879 Nucleus::createBr(unreachableBlock);
1880 }
1881
1882 Nucleus::setInsertBlock(unreachableBlock);
1883 Nucleus::createUnreachable();
1884 }
1885 }
1886
1887 void PixelProgram::LEAVE()
1888 {
1889 enableLeave = enableLeave & ~enableStack[Min(enableIndex, Int(MAX_SHADER_ENABLE_STACK_SIZE))];
1890
1891 // FIXME: Return from function if all instances left
1892 // FIXME: Use enableLeave in other control-flow constructs
1893 }
1894}
1895