• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2011 University of Szeged
3  * Copyright (C) 2011 Zoltan Herczeg
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY UNIVERSITY OF SZEGED ``AS IS'' AND ANY
15  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
17  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL UNIVERSITY OF SZEGED OR
18  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
19  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
20  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
22  * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
24  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 #include "config.h"
28 #include "FELightingNEON.h"
29 
30 #if CPU(ARM_NEON) && COMPILER(GCC)
31 
32 #include <wtf/Vector.h>
33 
34 namespace WebCore {
35 
36 // These constants are copied to the following SIMD registers:
37 //   ALPHAX_Q ALPHAY_Q REMAPX_D REMAPY_D
38 
39 WTF_ALIGNED(short, s_FELightingConstantsForNeon[], 16) = {
40     // Alpha coefficients.
41     -2, 1, 0, -1, 2, 1, 0, -1,
42     0, -1, -2, -1, 0, 1, 2, 1,
43     // Remapping indicies.
44     0x0f0e, 0x0302, 0x0504, 0x0706,
45     0x0b0a, 0x1312, 0x1514, 0x1716,
46 };
47 
feLightingConstantsForNeon()48 short* feLightingConstantsForNeon()
49 {
50     return s_FELightingConstantsForNeon;
51 }
52 
53 #define ASSTRING(str) #str
54 #define TOSTRING(value) ASSTRING(value)
55 
56 #define PIXELS_OFFSET TOSTRING(0)
57 #define WIDTH_OFFSET TOSTRING(4)
58 #define HEIGHT_OFFSET TOSTRING(8)
59 #define FLAGS_OFFSET TOSTRING(12)
60 #define SPECULAR_EXPONENT_OFFSET TOSTRING(16)
61 #define CONE_EXPONENT_OFFSET TOSTRING(20)
62 #define FLOAT_ARGUMENTS_OFFSET TOSTRING(24)
63 #define DRAWING_CONSTANTS_OFFSET TOSTRING(28)
64 #define NL "\n"
65 
66 // Register allocation
67 #define PAINTING_DATA_R       "r11"
68 #define RESET_WIDTH_R         PAINTING_DATA_R
69 #define PIXELS_R              "r4"
70 #define WIDTH_R               "r5"
71 #define HEIGHT_R              "r6"
72 #define FLAGS_R               "r7"
73 #define SPECULAR_EXPONENT_R   "r8"
74 #define CONE_EXPONENT_R       "r10"
75 #define SCANLINE_R            "r12"
76 
77 #define TMP1_Q                "q0"
78 #define TMP1_D0               "d0"
79 #define TMP1_S0               "s0"
80 #define TMP1_S1               "s1"
81 #define TMP1_D1               "d1"
82 #define TMP1_S2               "s2"
83 #define TMP1_S3               "s3"
84 #define TMP2_Q                "q1"
85 #define TMP2_D0               "d2"
86 #define TMP2_S0               "s4"
87 #define TMP2_S1               "s5"
88 #define TMP2_D1               "d3"
89 #define TMP2_S2               "s6"
90 #define TMP2_S3               "s7"
91 #define TMP3_Q                "q2"
92 #define TMP3_D0               "d4"
93 #define TMP3_S0               "s8"
94 #define TMP3_S1               "s9"
95 #define TMP3_D1               "d5"
96 #define TMP3_S2               "s10"
97 #define TMP3_S3               "s11"
98 
99 #define COSINE_OF_ANGLE       "s12"
100 #define POWF_INT_S            "s13"
101 #define POWF_FRAC_S           "s14"
102 #define SPOT_COLOR_Q          "q4"
103 
104 // Because of VMIN and VMAX CONST_ZERO_S and CONST_ONE_S
105 // must be placed on the same side of the double vector
106 
107 // Current pixel position
108 #define POSITION_Q            "q5"
109 #define POSITION_X_S          "s20"
110 #define POSITION_Y_S          "s21"
111 #define POSITION_Z_S          "s22"
112 #define CONST_ZERO_HI_D       "d11"
113 #define CONST_ZERO_S          "s23"
114 
115 // -------------------------------
116 //     Variable arguments
117 // Misc arguments
118 #define READ1_RANGE           "d12-d15"
119 #define READ2_RANGE           "d16-d19"
120 #define READ3_RANGE           "d20-d21"
121 
122 #define SCALE_S               "s24"
123 #define SCALE_DIV4_S          "s25"
124 #define DIFFUSE_CONST_S       "s26"
125 
126 // Light source position
127 #define CONE_CUT_OFF_S        "s28"
128 #define CONE_FULL_LIGHT_S     "s29"
129 #define CONE_CUT_OFF_RANGE_S  "s30"
130 #define CONST_ONE_HI_D        "d15"
131 #define CONST_ONE_S           "s31"
132 
133 #define LIGHT_Q               "q8"
134 #define DIRECTION_Q           "q9"
135 #define COLOR_Q               "q10"
136 // -------------------------------
137 //    Constant coefficients
138 #define READ4_RANGE           "d22-d25"
139 #define READ5_RANGE           "d26-d27"
140 
141 #define ALPHAX_Q              "q11"
142 #define ALPHAY_Q              "q12"
143 #define REMAPX_D              "d26"
144 #define REMAPY_D              "d27"
145 // -------------------------------
146 
147 #define ALL_ROWS_D            "{d28,d29,d30}"
148 #define TOP_ROW_D             "d28"
149 #define MIDDLE_ROW_D          "d29"
150 #define BOTTOM_ROW_D          "d30"
151 
152 #define GET_LENGTH(source, temp) \
153     "vmul.f32 " temp##_Q ", " source##_Q ", " source##_Q NL \
154     "vadd.f32 " source##_S3 ", " temp##_S0 ", " temp##_S1 NL \
155     "vadd.f32 " source##_S3 ", " source##_S3 ", " temp##_S2 NL \
156     "vsqrt.f32 " source##_S3 ", " source##_S3 NL
157 
158 // destination##_S3 can contain the multiply of length.
159 #define DOT_PRODUCT(destination, source1, source2) \
160     "vmul.f32 " destination##_Q ", " source1##_Q ", " source2##_Q NL \
161     "vadd.f32 " destination##_S0 ", " destination##_S0 ", " destination##_S1 NL \
162     "vadd.f32 " destination##_S0 ", " destination##_S0 ", " destination##_S2 NL
163 
164 #define MULTIPLY_BY_DIFFUSE_CONST(normalVectorLength, dotProductLength) \
165     "tst " FLAGS_R ", #" TOSTRING(FLAG_DIFFUSE_CONST_IS_1) NL \
166     "vmuleq.f32 " TMP2_S1 ", " DIFFUSE_CONST_S ", " normalVectorLength NL \
167     "vdiveq.f32 " TMP2_S1 ", " TMP2_S1 ", " dotProductLength NL \
168     "vdivne.f32 " TMP2_S1 ", " normalVectorLength ", " dotProductLength NL
169 
170 #define POWF_SQR(value, exponent, current, remaining) \
171     "tst " exponent ", #" ASSTRING(current) NL \
172     "vmulne.f32 " value ", " value ", " POWF_INT_S NL \
173     "tst " exponent ", #" ASSTRING(remaining) NL \
174     "vmulne.f32 " POWF_INT_S ", " POWF_INT_S ", " POWF_INT_S NL
175 
176 #define POWF_SQRT(value, exponent, current, remaining) \
177     "tst " exponent ", #" ASSTRING(remaining) NL \
178     "vsqrtne.f32 " POWF_FRAC_S ", " POWF_FRAC_S NL \
179     "tst " exponent ", #" ASSTRING(current) NL \
180     "vmulne.f32 " value ", " value ", " POWF_FRAC_S NL
181 
182 // This simplified powf function is sufficiently accurate.
183 #define POWF(value, exponent) \
184     "tst " exponent ", #0xfc0" NL \
185     "vmovne.f32 " POWF_INT_S ", " value NL \
186     "tst " exponent ", #0x03f" NL \
187     "vmovne.f32 " POWF_FRAC_S ", " value NL \
188     "vmov.f32 " value ", " CONST_ONE_S NL \
189     \
190     POWF_SQR(value, exponent, 0x040, 0xf80) \
191     POWF_SQR(value, exponent, 0x080, 0xf00) \
192     POWF_SQR(value, exponent, 0x100, 0xe00) \
193     POWF_SQR(value, exponent, 0x200, 0xc00) \
194     POWF_SQR(value, exponent, 0x400, 0x800) \
195     "tst " exponent ", #0x800" NL \
196     "vmulne.f32 " value ", " value ", " POWF_INT_S NL \
197     \
198     POWF_SQRT(value, exponent, 0x20, 0x3f) \
199     POWF_SQRT(value, exponent, 0x10, 0x1f) \
200     POWF_SQRT(value, exponent, 0x08, 0x0f) \
201     POWF_SQRT(value, exponent, 0x04, 0x07) \
202     POWF_SQRT(value, exponent, 0x02, 0x03) \
203     POWF_SQRT(value, exponent, 0x01, 0x01)
204 
205 // The following algorithm is an ARM-NEON optimized version of
206 // the main loop found in FELighting.cpp. Since the whole code
207 // is redesigned to be as effective as possible (ARM specific
208 // thinking), it is four times faster than its C++ counterpart.
209 
210 asm ( // NOLINT
211 ".globl " TOSTRING(neonDrawLighting) NL
212 TOSTRING(neonDrawLighting) ":" NL
213     // Because of the clever register allocation, nothing is stored on the stack
214     // except the saved registers.
215     // Stack must be aligned to 8 bytes.
216     "stmdb sp!, {r4-r8, r10, r11, lr}" NL
217     "vstmdb sp!, {d8-d15}" NL
218     "mov " PAINTING_DATA_R ", r0" NL
219 
220     // The following two arguments are loaded to SIMD registers.
221     "ldr r0, [" PAINTING_DATA_R ", #" FLOAT_ARGUMENTS_OFFSET "]" NL
222     "ldr r1, [" PAINTING_DATA_R ", #" DRAWING_CONSTANTS_OFFSET "]" NL
223     "ldr " PIXELS_R ", [" PAINTING_DATA_R ", #" PIXELS_OFFSET "]" NL
224     "ldr " WIDTH_R ", [" PAINTING_DATA_R ", #" WIDTH_OFFSET "]" NL
225     "ldr " HEIGHT_R ", [" PAINTING_DATA_R ", #" HEIGHT_OFFSET "]" NL
226     "ldr " FLAGS_R ", [" PAINTING_DATA_R ", #" FLAGS_OFFSET "]" NL
227     "ldr " SPECULAR_EXPONENT_R ", [" PAINTING_DATA_R ", #" SPECULAR_EXPONENT_OFFSET "]" NL
228     "ldr " CONE_EXPONENT_R ", [" PAINTING_DATA_R ", #" CONE_EXPONENT_OFFSET "]" NL
229 
230     // Load all data to the SIMD registers with the least number of instructions.
231     "vld1.f32 { " READ1_RANGE " }, [r0]!" NL
232     "vld1.f32 { " READ2_RANGE " }, [r0]!" NL
233     "vld1.f32 { " READ3_RANGE " }, [r0]!" NL
234     "vld1.s16 {" READ4_RANGE "}, [r1]!" NL
235     "vld1.s16 {" READ5_RANGE "}, [r1]!" NL
236 
237     // Initializing local variables.
238     "mov " SCANLINE_R ", " WIDTH_R ", lsl #2" NL
239     "add " SCANLINE_R ", " SCANLINE_R ", #8" NL
240     "add " PIXELS_R ", " PIXELS_R ", " SCANLINE_R NL
241     "add " PIXELS_R ", " PIXELS_R ", #3" NL
242     "mov r0, #0" NL
243     "vmov.f32 " CONST_ZERO_S ", r0" NL
244     "vmov.f32 " POSITION_Y_S ", " CONST_ONE_S NL
245     "tst " FLAGS_R ", #" TOSTRING(FLAG_SPOT_LIGHT) NL
246     "vmov.f32 " SPOT_COLOR_Q ", " COLOR_Q NL
247     "mov " RESET_WIDTH_R ", " WIDTH_R NL
248 
249 ".mainloop:" NL
250     "mov r3, #3" NL
251     "vmov.f32 " POSITION_X_S ", " CONST_ONE_S NL
252 
253 ".scanline:" NL
254     // The ROW registers are storing the alpha channel of the last three pixels.
255     // The alpha channel is stored as signed short (sint16) values. The fourth value
256     // is garbage. The following instructions are shifting out the unnecessary alpha
257     // values and load the next ones.
258     "ldrb r0, [" PIXELS_R ", -" SCANLINE_R "]" NL
259     "ldrb r1, [" PIXELS_R ", +" SCANLINE_R "]" NL
260     "ldrb r2, [" PIXELS_R "], #4" NL
261     "vext.s16 " TOP_ROW_D ", " TOP_ROW_D ", " TOP_ROW_D ", #3" NL
262     "vext.s16 " MIDDLE_ROW_D ", " MIDDLE_ROW_D ", " MIDDLE_ROW_D ", #3" NL
263     "vext.s16 " BOTTOM_ROW_D ", " BOTTOM_ROW_D ", " BOTTOM_ROW_D ", #3" NL
264     "vmov.s16 " TOP_ROW_D "[1], r0" NL
265     "vmov.s16 " MIDDLE_ROW_D "[1], r2" NL
266     "vmov.s16 " BOTTOM_ROW_D "[1], r1" NL
267 
268     // The two border pixels (rightmost and leftmost) are skipped when
269     // the next scanline is reached. It also jumps, when the algorithm
270     // is started, and the first free alpha values are loaded to each row.
271     "subs r3, r3, #1" NL
272     "bne .scanline" NL
273 
274     // The light vector goes to TMP1_Q. It is constant in case of distant light.
275     // The fourth value contains the length of the light vector.
276     "tst " FLAGS_R ", #" TOSTRING(FLAG_POINT_LIGHT | FLAG_SPOT_LIGHT) NL
277     "beq .distantLight" NL
278 
279     "vmov.s16 r3, " MIDDLE_ROW_D "[2]" NL
280     "vmov.f32 " POSITION_Z_S ", r3" NL
281     "vcvt.f32.s32 " POSITION_Z_S ", " POSITION_Z_S NL
282     "vmul.f32 " POSITION_Z_S ", " POSITION_Z_S ", " SCALE_S NL
283 
284     "vsub.f32 " TMP1_Q ", " LIGHT_Q ", " POSITION_Q NL
285     GET_LENGTH(TMP1, TMP2)
286 
287     "tst " FLAGS_R ", #" TOSTRING(FLAG_SPOT_LIGHT) NL
288     "bne .cosineOfAngle" NL
289 ".visiblePixel:" NL
290 
291     //     | -1  0  1 |      | -1 -2 -1 |
292     // X = | -2  0  2 |  Y = |  0  0  0 |
293     //     | -1  0  1 |      |  1  2  1 |
294 
295     // Multiply the alpha values by the X and Y matrices.
296 
297     // Moving the 8 alpha value to TMP3.
298     "vtbl.8 " TMP3_D0 ", " ALL_ROWS_D ", " REMAPX_D NL
299     "vtbl.8 " TMP3_D1 ", " ALL_ROWS_D ", " REMAPY_D NL
300 
301     "vmul.s16 " TMP2_Q ", " TMP3_Q ", " ALPHAX_Q NL
302     "vpadd.s16 " TMP2_D0 ", " TMP2_D0 ", " TMP2_D1 NL
303     "vpadd.s16 " TMP2_D0 ", " TMP2_D0 ", " TMP2_D0 NL
304     "vpadd.s16 " TMP2_D0 ", " TMP2_D0 ", " TMP2_D0 NL
305     "vmov.s16 r0, " TMP2_D0 "[0]" NL
306 
307     "vmul.s16 " TMP2_Q ", " TMP3_Q ", " ALPHAY_Q NL
308     "vpadd.s16 " TMP2_D0 ", " TMP2_D0 ", " TMP2_D1 NL
309     "vpadd.s16 " TMP2_D0 ", " TMP2_D0 ", " TMP2_D0 NL
310     "vpadd.s16 " TMP2_D0 ", " TMP2_D0 ", " TMP2_D0 NL
311     "vmov.s16 r1, " TMP2_D0 "[0]" NL
312 
313     // r0 and r1 contains the X and Y coordinates of the
314     // normal vector, respectively.
315 
316     // Calculating the spot light strength.
317     "tst " FLAGS_R ", #" TOSTRING(FLAG_SPOT_LIGHT) NL
318     "beq .endLight" NL
319 
320     "vneg.f32 " TMP3_S1 ", " COSINE_OF_ANGLE NL
321     "tst " FLAGS_R ", #" TOSTRING(FLAG_CONE_EXPONENT_IS_1) NL
322     "beq .coneExpPowf" NL
323 ".coneExpPowfFinished:" NL
324 
325     // Smoothing the cone edge if necessary.
326     "vcmp.f32 " COSINE_OF_ANGLE ", " CONE_FULL_LIGHT_S NL
327     "fmstat" NL
328     "bhi .cutOff" NL
329 ".cutOffFinished:" NL
330 
331     "vmin.f32 " TMP3_D0 ", " TMP3_D0 ", " CONST_ONE_HI_D NL
332     "vmul.f32 " COLOR_Q ", " SPOT_COLOR_Q ", " TMP3_D0 "[1]" NL
333 
334 ".endLight:" NL
335     // Summarize:
336     // r0 and r1 contains the normalVector.
337     // TMP1_Q contains the light vector and its length.
338     // COLOR_Q contains the color of the light vector.
339 
340     // Test whether both r0 and r1 are zero (Normal vector is (0, 0, 1)).
341     "orrs r2, r0, r1" NL
342     "bne .normalVectorIsNonZero" NL
343 
344     "tst " FLAGS_R ", #" TOSTRING(FLAG_SPECULAR_LIGHT) NL
345     "bne .specularLight1" NL
346 
347     // Calculate diffuse light strength.
348     MULTIPLY_BY_DIFFUSE_CONST(TMP1_S2, TMP1_S3)
349     "b .lightStrengthCalculated" NL
350 
351 ".specularLight1:" NL
352     // Calculating specular light strength.
353     "vadd.f32 " TMP1_S2 ", " TMP1_S2 ", " TMP1_S3 NL
354     GET_LENGTH(TMP1, TMP2)
355 
356     // When the exponent is 1, we don't need to call an expensive powf function.
357     "tst " FLAGS_R ", #" TOSTRING(FLAG_SPECULAR_EXPONENT_IS_1) NL
358     "vdiveq.f32 " TMP2_S1 ", " TMP1_S2 ", " TMP1_S3 NL
359     "beq .specularExpPowf" NL
360 
361     MULTIPLY_BY_DIFFUSE_CONST(TMP1_S2, TMP1_S3)
362     "b .lightStrengthCalculated" NL
363 
364 ".normalVectorIsNonZero:" NL
365     // Normal vector goes to TMP2, and its length is calculated as well.
366     "vmov.s32 " TMP2_S0 ", r0" NL
367     "vcvt.f32.s32 " TMP2_S0 ", " TMP2_S0 NL
368     "vmul.f32 " TMP2_S0 ", " TMP2_S0 ", " SCALE_DIV4_S NL
369     "vmov.s32 " TMP2_S1 ", r1" NL
370     "vcvt.f32.s32 " TMP2_S1 ", " TMP2_S1 NL
371     "vmul.f32 " TMP2_S1 ", " TMP2_S1 ", " SCALE_DIV4_S NL
372     "vmov.f32 " TMP2_S2 ", " CONST_ONE_S NL
373     GET_LENGTH(TMP2, TMP3)
374 
375     "tst " FLAGS_R ", #" TOSTRING(FLAG_SPECULAR_LIGHT) NL
376     "bne .specularLight2" NL
377 
378     // Calculating diffuse light strength.
379     DOT_PRODUCT(TMP3, TMP2, TMP1)
380     MULTIPLY_BY_DIFFUSE_CONST(TMP3_S0, TMP3_S3)
381     "b .lightStrengthCalculated" NL
382 
383 ".specularLight2:" NL
384     // Calculating specular light strength.
385     "vadd.f32 " TMP1_S2 ", " TMP1_S2 ", " TMP1_S3 NL
386     GET_LENGTH(TMP1, TMP3)
387     DOT_PRODUCT(TMP3, TMP2, TMP1)
388 
389     // When the exponent is 1, we don't need to call an expensive powf function.
390     "tst " FLAGS_R ", #" TOSTRING(FLAG_SPECULAR_EXPONENT_IS_1) NL
391     "vdiveq.f32 " TMP2_S1 ", " TMP3_S0 ", " TMP3_S3 NL
392     "beq .specularExpPowf" NL
393     MULTIPLY_BY_DIFFUSE_CONST(TMP3_S0, TMP3_S3)
394 
395 ".lightStrengthCalculated:" NL
396     // TMP2_S1 contains the light strength. Clamp it to [0, 1]
397     "vmax.f32 " TMP2_D0 ", " TMP2_D0 ", " CONST_ZERO_HI_D NL
398     "vmin.f32 " TMP2_D0 ", " TMP2_D0 ", " CONST_ONE_HI_D NL
399     "vmul.f32 " TMP3_Q ", " COLOR_Q ", " TMP2_D0 "[1]" NL
400     "vcvt.u32.f32 " TMP3_Q ", " TMP3_Q NL
401     "vmov.u32 r2, r3, " TMP3_S0 ", " TMP3_S1 NL
402     // The color values are stored in-place.
403     "strb r2, [" PIXELS_R ", #-11]" NL
404     "strb r3, [" PIXELS_R ", #-10]" NL
405     "vmov.u32 r2, " TMP3_S2 NL
406     "strb r2, [" PIXELS_R ", #-9]" NL
407 
408     // Continue to the next pixel.
409 ".blackPixel:" NL
410     "vadd.f32 " POSITION_X_S ", " CONST_ONE_S NL
411     "mov r3, #1" NL
412     "subs " WIDTH_R ", " WIDTH_R ", #1" NL
413     "bne .scanline" NL
414 
415     // If the end of the scanline is reached, we continue
416     // to the next scanline.
417     "vadd.f32 " POSITION_Y_S ", " CONST_ONE_S NL
418     "mov " WIDTH_R ", " RESET_WIDTH_R NL
419     "subs " HEIGHT_R ", " HEIGHT_R ", #1" NL
420     "bne .mainloop" NL
421 
422     // Return.
423     "vldmia sp!, {d8-d15}" NL
424     "ldmia sp!, {r4-r8, r10, r11, pc}" NL
425 
426 ".distantLight:" NL
427     // In case of distant light, the light vector is constant,
428     // we simply copy it.
429     "vmov.f32 " TMP1_Q ", " LIGHT_Q NL
430     "b .visiblePixel" NL
431 
432 ".cosineOfAngle:" NL
433     // If the pixel is outside of the cone angle, it is simply a black pixel.
434     DOT_PRODUCT(TMP3, TMP1, DIRECTION)
435     "vdiv.f32 " COSINE_OF_ANGLE ", " TMP3_S0 ", " TMP1_S3 NL
436     "vcmp.f32 " COSINE_OF_ANGLE ", " CONE_CUT_OFF_S NL
437     "fmstat" NL
438     "bls .visiblePixel" NL
439     "mov r0, #0" NL
440     "strh r0, [" PIXELS_R ", #-11]" NL
441     "strb r0, [" PIXELS_R ", #-9]" NL
442     "b .blackPixel" NL
443 
444 ".cutOff:" NL
445     // Smoothing the light strength on the cone edge.
446     "vsub.f32 " TMP3_S0 ", " CONE_CUT_OFF_S ", " COSINE_OF_ANGLE NL
447     "vdiv.f32 " TMP3_S0 ", " TMP3_S0 ", " CONE_CUT_OFF_RANGE_S NL
448     "vmul.f32 " TMP3_S1 ", " TMP3_S1 ", " TMP3_S0 NL
449     "b .cutOffFinished" NL
450 
451 ".coneExpPowf:" NL
452     POWF(TMP3_S1, CONE_EXPONENT_R)
453     "b .coneExpPowfFinished" NL
454 
455 ".specularExpPowf:" NL
456     POWF(TMP2_S1, SPECULAR_EXPONENT_R)
457     "tst " FLAGS_R ", #" TOSTRING(FLAG_DIFFUSE_CONST_IS_1) NL
458     "vmuleq.f32 " TMP2_S1 ", " TMP2_S1 ", " DIFFUSE_CONST_S NL
459     "b .lightStrengthCalculated" NL
460 ); // NOLINT
461 
462 } // namespace WebCore
463 
464 #endif // CPU(ARM_NEON) && COMPILER(GCC)
465