xref: /openbmc/qemu/util/bufferiszero.c (revision 750541c4)
1 /*
2  * Simple C functions to supplement the C library
3  *
4  * Copyright (c) 2006 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "qemu/osdep.h"
25 #include "qemu-common.h"
26 #include "qemu/cutils.h"
27 #include "qemu/bswap.h"
28 
29 static bool
30 buffer_zero_int(const void *buf, size_t len)
31 {
32     if (unlikely(len < 8)) {
33         /* For a very small buffer, simply accumulate all the bytes.  */
34         const unsigned char *p = buf;
35         const unsigned char *e = buf + len;
36         unsigned char t = 0;
37 
38         do {
39             t |= *p++;
40         } while (p < e);
41 
42         return t == 0;
43     } else {
44         /* Otherwise, use the unaligned memory access functions to
45            handle the beginning and end of the buffer, with a couple
46            of loops handling the middle aligned section.  */
47         uint64_t t = ldq_he_p(buf);
48         const uint64_t *p = (uint64_t *)(((uintptr_t)buf + 8) & -8);
49         const uint64_t *e = (uint64_t *)(((uintptr_t)buf + len) & -8);
50 
51         for (; p + 8 <= e; p += 8) {
52             __builtin_prefetch(p + 8);
53             if (t) {
54                 return false;
55             }
56             t = p[0] | p[1] | p[2] | p[3] | p[4] | p[5] | p[6] | p[7];
57         }
58         while (p < e) {
59             t |= *p++;
60         }
61         t |= ldq_he_p(buf + len - 8);
62 
63         return t == 0;
64     }
65 }
66 
67 #if defined(CONFIG_AVX2_OPT) || defined(__SSE2__)
68 /* Do not use push_options pragmas unnecessarily, because clang
69  * does not support them.
70  */
71 #ifdef CONFIG_AVX2_OPT
72 #pragma GCC push_options
73 #pragma GCC target("sse2")
74 #endif
75 #include <emmintrin.h>
76 
77 /* Note that each of these vectorized functions require len >= 64.  */
78 
79 static bool
80 buffer_zero_sse2(const void *buf, size_t len)
81 {
82     __m128i t = _mm_loadu_si128(buf);
83     __m128i *p = (__m128i *)(((uintptr_t)buf + 5 * 16) & -16);
84     __m128i *e = (__m128i *)(((uintptr_t)buf + len) & -16);
85     __m128i zero = _mm_setzero_si128();
86 
87     /* Loop over 16-byte aligned blocks of 64.  */
88     while (likely(p <= e)) {
89         __builtin_prefetch(p);
90         t = _mm_cmpeq_epi8(t, zero);
91         if (unlikely(_mm_movemask_epi8(t) != 0xFFFF)) {
92             return false;
93         }
94         t = p[-4] | p[-3] | p[-2] | p[-1];
95         p += 4;
96     }
97 
98     /* Finish the aligned tail.  */
99     t |= e[-3];
100     t |= e[-2];
101     t |= e[-1];
102 
103     /* Finish the unaligned tail.  */
104     t |= _mm_loadu_si128(buf + len - 16);
105 
106     return _mm_movemask_epi8(_mm_cmpeq_epi8(t, zero)) == 0xFFFF;
107 }
108 #ifdef CONFIG_AVX2_OPT
109 #pragma GCC pop_options
110 #endif
111 
112 #ifdef CONFIG_AVX2_OPT
113 /* Note that due to restrictions/bugs wrt __builtin functions in gcc <= 4.8,
114  * the includes have to be within the corresponding push_options region, and
115  * therefore the regions themselves have to be ordered with increasing ISA.
116  */
117 #pragma GCC push_options
118 #pragma GCC target("sse4")
119 #include <smmintrin.h>
120 
121 static bool
122 buffer_zero_sse4(const void *buf, size_t len)
123 {
124     __m128i t = _mm_loadu_si128(buf);
125     __m128i *p = (__m128i *)(((uintptr_t)buf + 5 * 16) & -16);
126     __m128i *e = (__m128i *)(((uintptr_t)buf + len) & -16);
127 
128     /* Loop over 16-byte aligned blocks of 64.  */
129     while (likely(p <= e)) {
130         __builtin_prefetch(p);
131         if (unlikely(!_mm_testz_si128(t, t))) {
132             return false;
133         }
134         t = p[-4] | p[-3] | p[-2] | p[-1];
135         p += 4;
136     }
137 
138     /* Finish the aligned tail.  */
139     t |= e[-3];
140     t |= e[-2];
141     t |= e[-1];
142 
143     /* Finish the unaligned tail.  */
144     t |= _mm_loadu_si128(buf + len - 16);
145 
146     return _mm_testz_si128(t, t);
147 }
148 
149 #pragma GCC pop_options
150 #pragma GCC push_options
151 #pragma GCC target("avx2")
152 #include <immintrin.h>
153 
154 static bool
155 buffer_zero_avx2(const void *buf, size_t len)
156 {
157     /* Begin with an unaligned head of 32 bytes.  */
158     __m256i t = _mm256_loadu_si256(buf);
159     __m256i *p = (__m256i *)(((uintptr_t)buf + 5 * 32) & -32);
160     __m256i *e = (__m256i *)(((uintptr_t)buf + len) & -32);
161 
162     if (likely(p <= e)) {
163         /* Loop over 32-byte aligned blocks of 128.  */
164         do {
165             __builtin_prefetch(p);
166             if (unlikely(!_mm256_testz_si256(t, t))) {
167                 return false;
168             }
169             t = p[-4] | p[-3] | p[-2] | p[-1];
170             p += 4;
171         } while (p <= e);
172     } else {
173         t |= _mm256_loadu_si256(buf + 32);
174         if (len <= 128) {
175             goto last2;
176         }
177     }
178 
179     /* Finish the last block of 128 unaligned.  */
180     t |= _mm256_loadu_si256(buf + len - 4 * 32);
181     t |= _mm256_loadu_si256(buf + len - 3 * 32);
182  last2:
183     t |= _mm256_loadu_si256(buf + len - 2 * 32);
184     t |= _mm256_loadu_si256(buf + len - 1 * 32);
185 
186     return _mm256_testz_si256(t, t);
187 }
188 #pragma GCC pop_options
189 #endif /* CONFIG_AVX2_OPT */
190 
191 /* Note that for test_buffer_is_zero_next_accel, the most preferred
192  * ISA must have the least significant bit.
193  */
194 #define CACHE_AVX2    1
195 #define CACHE_SSE4    2
196 #define CACHE_SSE2    4
197 
198 /* Make sure that these variables are appropriately initialized when
199  * SSE2 is enabled on the compiler command-line, but the compiler is
200  * too old to support CONFIG_AVX2_OPT.
201  */
202 #ifdef CONFIG_AVX2_OPT
203 # define INIT_CACHE 0
204 # define INIT_ACCEL buffer_zero_int
205 #else
206 # ifndef __SSE2__
207 #  error "ISA selection confusion"
208 # endif
209 # define INIT_CACHE CACHE_SSE2
210 # define INIT_ACCEL buffer_zero_sse2
211 #endif
212 
213 static unsigned cpuid_cache = INIT_CACHE;
214 static bool (*buffer_accel)(const void *, size_t) = INIT_ACCEL;
215 
216 static void init_accel(unsigned cache)
217 {
218     bool (*fn)(const void *, size_t) = buffer_zero_int;
219     if (cache & CACHE_SSE2) {
220         fn = buffer_zero_sse2;
221     }
222 #ifdef CONFIG_AVX2_OPT
223     if (cache & CACHE_SSE4) {
224         fn = buffer_zero_sse4;
225     }
226     if (cache & CACHE_AVX2) {
227         fn = buffer_zero_avx2;
228     }
229 #endif
230     buffer_accel = fn;
231 }
232 
233 #ifdef CONFIG_AVX2_OPT
234 #include "qemu/cpuid.h"
235 
236 static void __attribute__((constructor)) init_cpuid_cache(void)
237 {
238     int max = __get_cpuid_max(0, NULL);
239     int a, b, c, d;
240     unsigned cache = 0;
241 
242     if (max >= 1) {
243         __cpuid(1, a, b, c, d);
244         if (d & bit_SSE2) {
245             cache |= CACHE_SSE2;
246         }
247         if (c & bit_SSE4_1) {
248             cache |= CACHE_SSE4;
249         }
250 
251         /* We must check that AVX is not just available, but usable.  */
252         if ((c & bit_OSXSAVE) && (c & bit_AVX) && max >= 7) {
253             int bv;
254             __asm("xgetbv" : "=a"(bv), "=d"(d) : "c"(0));
255             __cpuid_count(7, 0, a, b, c, d);
256             if ((bv & 6) == 6 && (b & bit_AVX2)) {
257                 cache |= CACHE_AVX2;
258             }
259         }
260     }
261     cpuid_cache = cache;
262     init_accel(cache);
263 }
264 #endif /* CONFIG_AVX2_OPT */
265 
266 bool test_buffer_is_zero_next_accel(void)
267 {
268     /* If no bits set, we just tested buffer_zero_int, and there
269        are no more acceleration options to test.  */
270     if (cpuid_cache == 0) {
271         return false;
272     }
273     /* Disable the accelerator we used before and select a new one.  */
274     cpuid_cache &= cpuid_cache - 1;
275     init_accel(cpuid_cache);
276     return true;
277 }
278 
279 static bool select_accel_fn(const void *buf, size_t len)
280 {
281     if (likely(len >= 64)) {
282         return buffer_accel(buf, len);
283     }
284     return buffer_zero_int(buf, len);
285 }
286 
287 #else
288 #define select_accel_fn  buffer_zero_int
289 bool test_buffer_is_zero_next_accel(void)
290 {
291     return false;
292 }
293 #endif
294 
295 /*
296  * Checks if a buffer is all zeroes
297  */
298 bool buffer_is_zero(const void *buf, size_t len)
299 {
300     if (unlikely(len == 0)) {
301         return true;
302     }
303 
304     /* Fetch the beginning of the buffer while we select the accelerator.  */
305     __builtin_prefetch(buf);
306 
307     /* Use an optimized zero check if possible.  Note that this also
308        includes a check for an unrolled loop over 64-bit integers.  */
309     return select_accel_fn(buf, len);
310 }
311