• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* libs/pixelflinger/fixed.cpp
2 **
3 ** Copyright 2006, The Android Open Source Project
4 **
5 ** Licensed under the Apache License, Version 2.0 (the "License");
6 ** you may not use this file except in compliance with the License.
7 ** You may obtain a copy of the License at
8 **
9 **     http://www.apache.org/licenses/LICENSE-2.0
10 **
11 ** Unless required by applicable law or agreed to in writing, software
12 ** distributed under the License is distributed on an "AS IS" BASIS,
13 ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 ** See the License for the specific language governing permissions and
15 ** limitations under the License.
16 */
17 
18 #include <stdio.h>
19 
20 #include <private/pixelflinger/ggl_context.h>
21 #include <private/pixelflinger/ggl_fixed.h>
22 
23 
24 // ------------------------------------------------------------------------
25 
gglRecipQNormalized(int32_t x,int * exponent)26 int32_t gglRecipQNormalized(int32_t x, int* exponent)
27 {
28     const int32_t s = x>>31;
29     uint32_t a = s ? -x : x;
30 
31     // the result will overflow, so just set it to the biggest/inf value
32     if (ggl_unlikely(a <= 2LU)) {
33         *exponent = 0;
34         return s ? FIXED_MIN : FIXED_MAX;
35     }
36 
37     // Newton-Raphson iteration:
38     // x = r*(2 - a*r)
39 
40     const int32_t lz = gglClz(a);
41     a <<= lz;  // 0.32
42     uint32_t r = a;
43     // note: if a == 0x80000000, this means x was a power-of-2, in this
44     // case we don't need to compute anything. We get the reciprocal for
45     // (almost) free.
46     if (a != 0x80000000) {
47         r = (0x2E800 << (30-16)) - (r>>(2-1)); // 2.30, r = 2.90625 - 2*a
48         // 0.32 + 2.30 = 2.62 -> 2.30
49         // 2.30 + 2.30 = 4.60 -> 2.30
50         r = (((2LU<<30) - uint32_t((uint64_t(a)*r) >> 32)) * uint64_t(r)) >> 30;
51         r = (((2LU<<30) - uint32_t((uint64_t(a)*r) >> 32)) * uint64_t(r)) >> 30;
52     }
53 
54     // shift right 1-bit to make room for the sign bit
55     *exponent = 30-lz-1;
56     r >>= 1;
57     return s ? -r : r;
58 }
59 
gglRecipQ(GGLfixed x,int q)60 int32_t gglRecipQ(GGLfixed x, int q)
61 {
62     int shift;
63     x = gglRecipQNormalized(x, &shift);
64     shift += 16-q;
65     x += 1L << (shift-1);   // rounding
66     x >>= shift;
67     return x;
68 }
69 
70 // ------------------------------------------------------------------------
71 
gglFastDivx(GGLfixed n,GGLfixed d)72 GGLfixed gglFastDivx(GGLfixed n, GGLfixed d)
73 {
74     if ((d>>24) && ((d>>24)+1)) {
75         n >>= 8;
76         d >>= 8;
77     }
78     return gglMulx(n, gglRecip(d));
79 }
80 
81 // ------------------------------------------------------------------------
82 
83 static const GGLfixed ggl_sqrt_reciproc_approx_tab[8] = {
84     // 1/sqrt(x) with x = 1-N/16, N=[8...1]
85     0x16A09, 0x15555, 0x143D1, 0x134BF, 0x1279A, 0x11C01, 0x111AC, 0x10865
86 };
87 
gglSqrtRecipx(GGLfixed x)88 GGLfixed gglSqrtRecipx(GGLfixed x)
89 {
90     if (x == 0)         return FIXED_MAX;
91     if (x == FIXED_ONE) return x;
92     const GGLfixed a = x;
93     const int32_t lz = gglClz(x);
94     x = ggl_sqrt_reciproc_approx_tab[(a>>(28-lz))&0x7];
95     const int32_t exp = lz - 16;
96     if (exp <= 0)   x >>= -exp>>1;
97     else            x <<= (exp>>1) + (exp & 1);
98     if (exp & 1) {
99         x = gglMulx(x, ggl_sqrt_reciproc_approx_tab[0])>>1;
100     }
101     // 2 Newton-Raphson iterations: x = x/2*(3-(a*x)*x)
102     x = gglMulx((x>>1),(0x30000 - gglMulx(gglMulx(a,x),x)));
103     x = gglMulx((x>>1),(0x30000 - gglMulx(gglMulx(a,x),x)));
104     return x;
105 }
106 
gglSqrtx(GGLfixed a)107 GGLfixed gglSqrtx(GGLfixed a)
108 {
109     // Compute a full precision square-root (24 bits accuracy)
110     GGLfixed r = 0;
111     GGLfixed bit = 0x800000;
112     int32_t bshift = 15;
113     do {
114         GGLfixed temp = bit + (r<<1);
115         if (bshift >= 8)    temp <<= (bshift-8);
116         else                temp >>= (8-bshift);
117         if (a >= temp) {
118             r += bit;
119             a -= temp;
120         }
121         bshift--;
122     } while (bit>>=1);
123     return r;
124 }
125 
126 // ------------------------------------------------------------------------
127 
128 static const GGLfixed ggl_log_approx_tab[] = {
129     // -ln(x)/ln(2) with x = N/16, N=[8...16]
130     0xFFFF, 0xd47f, 0xad96, 0x8a62, 0x6a3f, 0x4caf, 0x3151, 0x17d6, 0x0000
131 };
132 
133 static const GGLfixed ggl_alog_approx_tab[] = { // domain [0 - 1.0]
134 	0xffff, 0xeac0, 0xd744, 0xc567, 0xb504, 0xa5fe, 0x9837, 0x8b95, 0x8000
135 };
136 
gglPowx(GGLfixed x,GGLfixed y)137 GGLfixed gglPowx(GGLfixed x, GGLfixed y)
138 {
139     // prerequisite: 0 <= x <= 1, and y >=0
140 
141     // pow(x,y) = 2^(y*log2(x))
142     // =  2^(y*log2(x*(2^exp)*(2^-exp))))
143     // =  2^(y*(log2(X)-exp))
144     // =  2^(log2(X)*y - y*exp)
145     // =  2^( - (-log2(X)*y + y*exp) )
146 
147     int32_t exp = gglClz(x) - 16;
148     GGLfixed f = x << exp;
149     x = (f & 0x0FFF)<<4;
150     f = (f >> 12) & 0x7;
151     GGLfixed p = gglMulAddx(
152             ggl_log_approx_tab[f+1] - ggl_log_approx_tab[f], x,
153             ggl_log_approx_tab[f]);
154     p = gglMulAddx(p, y, y*exp);
155     exp = gglFixedToIntFloor(p);
156     if (exp < 31) {
157         p = gglFracx(p);
158         x = (p & 0x1FFF)<<3;
159         p >>= 13;
160         p = gglMulAddx(
161                 ggl_alog_approx_tab[p+1] - ggl_alog_approx_tab[p], x,
162                 ggl_alog_approx_tab[p]);
163         p >>= exp;
164     } else {
165         p = 0;
166     }
167     return p;
168         // ( powf((a*65536.0f), (b*65536.0f)) ) * 65536.0f;
169 }
170 
171 // ------------------------------------------------------------------------
172 
gglDivQ(GGLfixed n,GGLfixed d,int32_t i)173 int32_t gglDivQ(GGLfixed n, GGLfixed d, int32_t i)
174 {
175     //int32_t r =int32_t((int64_t(n)<<i)/d);
176     const int32_t ds = n^d;
177     if (n<0) n = -n;
178     if (d<0) d = -d;
179     int nd = gglClz(d) - gglClz(n);
180     i += nd + 1;
181     if (nd > 0) d <<= nd;
182     else        n <<= -nd;
183     uint32_t q = 0;
184 
185     int j = i & 7;
186     i >>= 3;
187 
188     // gcc deals with the code below pretty well.
189     // we get 3.75 cycles per bit in the main loop
190     // and 8 cycles per bit in the termination loop
191     if (ggl_likely(i)) {
192         n -= d;
193         do {
194             q <<= 8;
195             if (n>=0)   q |= 128;
196             else        n += d;
197             n = n*2 - d;
198             if (n>=0)   q |= 64;
199             else        n += d;
200             n = n*2 - d;
201             if (n>=0)   q |= 32;
202             else        n += d;
203             n = n*2 - d;
204             if (n>=0)   q |= 16;
205             else        n += d;
206             n = n*2 - d;
207             if (n>=0)   q |= 8;
208             else        n += d;
209             n = n*2 - d;
210             if (n>=0)   q |= 4;
211             else        n += d;
212             n = n*2 - d;
213             if (n>=0)   q |= 2;
214             else        n += d;
215             n = n*2 - d;
216             if (n>=0)   q |= 1;
217             else        n += d;
218 
219             if (--i == 0)
220                 goto finish;
221 
222             n = n*2 - d;
223         } while(true);
224         do {
225             q <<= 1;
226             n = n*2 - d;
227             if (n>=0)   q |= 1;
228             else        n += d;
229         finish: ;
230         } while (j--);
231         return (ds<0) ? -q : q;
232     }
233 
234     n -= d;
235     if (n>=0)   q |= 1;
236     else        n += d;
237     j--;
238     goto finish;
239 }
240 
241 // ------------------------------------------------------------------------
242 
243 // assumes that the int32_t values of a, b, and c are all positive
244 // use when both a and b are larger than c
245 
246 template <typename T>
swap(T & a,T & b)247 static inline void swap(T& a, T& b) {
248     T t(a);
249     a = b;
250     b = t;
251 }
252 
253 static __attribute__((noinline))
slow_muldiv(uint32_t a,uint32_t b,uint32_t c)254 int32_t slow_muldiv(uint32_t a, uint32_t b, uint32_t c)
255 {
256 	// first we compute a*b as a 64-bit integer
257     // (GCC generates umull with the code below)
258     uint64_t ab = uint64_t(a)*b;
259     uint32_t hi = ab>>32;
260     uint32_t lo = ab;
261     uint32_t result;
262 
263 	// now perform the division
264 	if (hi >= c) {
265 	overflow:
266 		result = 0x7fffffff;  // basic overflow
267 	} else if (hi == 0) {
268 		result = lo/c;  // note: c can't be 0
269 		if ((result >> 31) != 0)  // result must fit in 31 bits
270 			goto overflow;
271 	} else {
272 		uint32_t r = hi;
273 		int bits = 31;
274 	    result = 0;
275 		do {
276 			r = (r << 1) | (lo >> 31);
277 			lo <<= 1;
278 			result <<= 1;
279 			if (r >= c) {
280 				r -= c;
281 				result |= 1;
282 			}
283 		} while (bits--);
284 	}
285 	return int32_t(result);
286 }
287 
288 // assumes a >= 0 and c >= b >= 0
289 static inline
quick_muldiv(int32_t a,int32_t b,int32_t c)290 int32_t quick_muldiv(int32_t a, int32_t b, int32_t c)
291 {
292     int32_t r = 0, q = 0, i;
293     int leading = gglClz(a);
294     i = 32 - leading;
295     a <<= leading;
296     do {
297         r <<= 1;
298         if (a < 0)
299             r += b;
300         a <<= 1;
301         q <<= 1;
302         if (r >= c) {
303             r -= c;
304             q++;
305         }
306         asm(""::); // gcc generates better code this way
307         if (r >= c) {
308             r -= c;
309             q++;
310         }
311     }
312     while (--i);
313     return q;
314 }
315 
316 // this function computes a*b/c with 64-bit intermediate accuracy
317 // overflows (e.g. division by 0) are handled and return INT_MAX
318 
gglMulDivi(int32_t a,int32_t b,int32_t c)319 int32_t gglMulDivi(int32_t a, int32_t b, int32_t c)
320 {
321 	int32_t result;
322 	int32_t sign = a^b^c;
323 
324 	if (a < 0) a = -a;
325 	if (b < 0) b = -b;
326 	if (c < 0) c = -c;
327 
328     if (a < b) {
329         swap(a, b);
330     }
331 
332 	if (b <= c) result = quick_muldiv(a, b, c);
333 	else        result = slow_muldiv((uint32_t)a, (uint32_t)b, (uint32_t)c);
334 
335 	if (sign < 0)
336 		result = -result;
337 
338     return result;
339 }
340