Skip to content

Commit a09fc18

Browse files
thiagomacieirathiagomacieira
committed
Add a configure-time check for QT_COMPILER_SUPPORTS_SIMD_ALWAYS
This has two main benefits: 1) introduces a qmake CONFIG we can use in .pro/.pri/.prf files 2) removes the need to keep an up-to-date list of which compilers support the feature The test is implemented as trying to compile every single SIMD test we currently have, but without passing the -mXXX option. The reason for trying all of them is that some people may have modified their mkspecs to add -mXXX options or -march=XXX, which could enable the particular feature we tried, resulting in a false positive outcome. Change-Id: I938b024e38bf4aac9154fffd14f7784dc8d1f020 Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
1 parent ff1ffa7 commit a09fc18

File tree

5 files changed

+291
-6
lines changed

5 files changed

+291
-6
lines changed

‎config.tests/x86_simd/main.cpp

+253
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,253 @@
1+
/****************************************************************************
2+
**
3+
** Copyright (C) 2017 Intel Corporation.
4+
** Contact: https://www.qt.io/licensing/
5+
**
6+
** This file is part of the QtCore module of the Qt Toolkit.
7+
**
8+
** $QT_BEGIN_LICENSE:LGPL$
9+
** Commercial License Usage
10+
** Licensees holding valid commercial Qt licenses may use this file in
11+
** accordance with the commercial license agreement provided with the
12+
** Software or, alternatively, in accordance with the terms contained in
13+
** a written agreement between you and The Qt Company. For licensing terms
14+
** and conditions see https://www.qt.io/terms-conditions. For further
15+
** information use the contact form at https://www.qt.io/contact-us.
16+
**
17+
** GNU Lesser General Public License Usage
18+
** Alternatively, this file may be used under the terms of the GNU Lesser
19+
** General Public License version 3 as published by the Free Software
20+
** Foundation and appearing in the file LICENSE.LGPL3 included in the
21+
** packaging of this file. Please review the following information to
22+
** ensure the GNU Lesser General Public License version 3 requirements
23+
** will be met: https://www.gnu.org/licenses/lgpl-3.0.html.
24+
**
25+
** GNU General Public License Usage
26+
** Alternatively, this file may be used under the terms of the GNU
27+
** General Public License version 2.0 or (at your option) the GNU General
28+
** Public license version 3 or any later version approved by the KDE Free
29+
** Qt Foundation. The licenses are as published by the Free Software
30+
** Foundation and appearing in the file LICENSE.GPL2 and LICENSE.GPL3
31+
** included in the packaging of this file. Please review the following
32+
** information to ensure the GNU General Public License requirements will
33+
** be met: https://www.gnu.org/licenses/gpl-2.0.html and
34+
** https://www.gnu.org/licenses/gpl-3.0.html.
35+
**
36+
** $QT_END_LICENSE$
37+
**
38+
****************************************************************************/
39+
40+
// All of our supported compilers support <immintrin.h>
41+
#include <immintrin.h>
42+
#define T(x) (QT_COMPILER_SUPPORTS_ ## x)
43+
44+
#if !defined(__INTEL_COMPILER) && !defined(_MSC_VER)
45+
/* GCC requires attributes for a function */
46+
# define attribute_target(x) __attribute__((__target__(x)))
47+
#else
48+
# define attribute_target(x)
49+
#endif
50+
51+
#if T(SSE2)
52+
attribute_target("sse2") void test_sse2()
53+
{
54+
__m128i a = _mm_setzero_si128();
55+
_mm_maskmoveu_si128(a, _mm_setzero_si128(), 0);
56+
}
57+
#endif
58+
59+
#if T(SSE3)
60+
attribute_target("sse3") void test_sse3()
61+
{
62+
__m128d a = _mm_set1_pd(6.28);
63+
__m128d b = _mm_set1_pd(3.14);
64+
__m128d result = _mm_addsub_pd(a, b);
65+
(void) _mm_movedup_pd(result);
66+
}
67+
#endif
68+
69+
#if T(SSSE3)
70+
attribute_target("ssse3") void test_ssse3()
71+
{
72+
__m128i a = _mm_set1_epi32(42);
73+
_mm_abs_epi8(a);
74+
(void) _mm_sign_epi16(a, _mm_set1_epi32(64));
75+
}
76+
#endif
77+
78+
#if T(SSE4_1)
79+
attribute_target("sse4.1") void test_sse4_1()
80+
{
81+
__m128 a = _mm_setzero_ps();
82+
_mm_ceil_ps(a);
83+
__m128i result = _mm_mullo_epi32(_mm_set1_epi32(42), _mm_set1_epi32(64));
84+
(void)result;
85+
}
86+
#endif
87+
88+
#if T(SSE4_2)
89+
attribute_target("sse4.2") void test_sse4_2()
90+
{
91+
__m128i a = _mm_setzero_si128();
92+
__m128i b = _mm_set1_epi32(42);
93+
(void) _mm_cmpestrm(a, 16, b, 16, 0);
94+
}
95+
#endif
96+
97+
#if T(AESNI)
98+
attribute_target("aes,sse4.2") void test_aesni()
99+
{
100+
__m128i a = _mm_setzero_si128();
101+
__m128i b = _mm_aesenc_si128(a, a);
102+
__m128i c = _mm_aesdec_si128(a, b);
103+
(void)c;
104+
}
105+
#endif
106+
107+
#if T(F16C)
108+
attribute_target("f16c") void test_f16c()
109+
{
110+
__m128i a = _mm_setzero_si128();
111+
__m128 b = _mm_cvtph_ps(a);
112+
__m256 b256 = _mm256_cvtph_ps(a);
113+
(void) _mm_cvtps_ph(b, 0);
114+
(void) _mm256_cvtps_ph(b256, 0);
115+
}
116+
#endif
117+
118+
#if T(RDRND)
119+
attribute_target("rdrnd") int test_rdrnd()
120+
{
121+
unsigned short us;
122+
unsigned int ui;
123+
if (_rdrand16_step(&us))
124+
return 1;
125+
if (_rdrand32_step(&ui))
126+
return 1;
127+
# if defined(__x86_64) || defined(__x86_64__) || defined(__amd64) || defined(_M_X64)
128+
unsigned long long ull;
129+
if (_rdrand64_step(&ull))
130+
return 1;
131+
# endif
132+
}
133+
#endif
134+
135+
#if T(SHANI)
136+
attribute_target("sha") void test_shani()
137+
{
138+
__m128i a = _mm_setzero_si128();
139+
__m128i b = _mm_sha1rnds4_epu32(a, a, 0);
140+
__m128i c = _mm_sha1msg1_epu32(a, b);
141+
__m128i d = _mm_sha256msg2_epu32(b, c);
142+
(void)d;
143+
}
144+
#endif
145+
146+
#if T(AVX)
147+
attribute_target("avx") void test_avx()
148+
{
149+
__m256d a = _mm256_setzero_pd();
150+
__m256d b = _mm256_set1_pd(42.42);
151+
(void) _mm256_add_pd(a, b);
152+
}
153+
#endif
154+
155+
#if T(AVX2)
156+
attribute_target("avx2") void test_avx2()
157+
{
158+
_mm256_zeroall();
159+
__m256i a = _mm256_setzero_si256();
160+
__m256i b = _mm256_and_si256(a, a);
161+
(void) _mm256_add_epi8(a, b);
162+
}
163+
#endif
164+
165+
#if T(AVX512F)
166+
attribute_target("avx512f") void test_avx512f(char *ptr)
167+
{
168+
/* AVX512 Foundation */
169+
__mmask16 m = ~1;
170+
__m512i i;
171+
__m512d d;
172+
__m512 f;
173+
i = _mm512_maskz_loadu_epi32(0, ptr);
174+
d = _mm512_loadu_pd((double *)ptr + 64);
175+
f = _mm512_loadu_ps((float *)ptr + 128);
176+
_mm512_mask_storeu_epi64(ptr, m, i);
177+
_mm512_mask_storeu_ps(ptr + 64, m, f);
178+
_mm512_mask_storeu_pd(ptr + 128, m, d);
179+
}
180+
#endif
181+
182+
#if T(AVX512ER)
183+
attribute_target("avx512er") void test_avx512er()
184+
{
185+
/* AVX512 Exponential and Reciprocal */
186+
__m512 f;
187+
f = _mm512_exp2a23_round_ps(f, 8);
188+
}
189+
#endif
190+
191+
#if T(AVX512CD)
192+
attribute_target("avx512cd") void test_avx512cd()
193+
{
194+
/* AVX512 Conflict Detection */
195+
__mmask16 m = ~1;
196+
__m512i i;
197+
i = _mm512_maskz_conflict_epi32(m, i);
198+
}
199+
#endif
200+
201+
#if T(AVX512PF)
202+
attribute_target("avx512pf") void test_avx512pf(void *ptr)
203+
{
204+
/* AVX512 Prefetch */
205+
__m512i i;
206+
__mmask16 m = 0xf;
207+
_mm512_mask_prefetch_i64scatter_pd(ptr, m, i, 2, 2);
208+
}
209+
#endif
210+
211+
#if T(AVX512DQ)
212+
attribute_target("avx512dq") void test_avx512dq()
213+
{
214+
/* AVX512 Doubleword and Quadword support */
215+
__m512i i;
216+
__mmask16 m = ~1;
217+
m = _mm512_movepi32_mask(i);
218+
}
219+
#endif
220+
221+
#if T(AVX512BW)
222+
attribute_target("avx512bw") void test_avx512bw(char *ptr)
223+
{
224+
/* AVX512 Byte and Word support */
225+
__m512i i;
226+
__mmask16 m = ~1;
227+
i = _mm512_mask_loadu_epi8(i, m, ptr - 8);
228+
}
229+
#endif
230+
231+
#if T(AVX512VL)
232+
attribute_target("avx512vl") void test_avx512vl(char *ptr)
233+
{
234+
/* AVX512 Vector Length */
235+
__mmask16 m = ~1;
236+
__m256i i2 = _mm256_maskz_loadu_epi32(0, ptr);
237+
_mm256_mask_storeu_epi32(ptr + 1, m, i2);
238+
}
239+
#endif
240+
241+
#if T(AVX512IFMA)
242+
attribute_target("avx512ifma") void test_avx512ifma()
243+
{
244+
/* AVX512 Integer Fused Multiply-Add */
245+
__m512i i;
246+
i = _mm512_madd52lo_epu64(i, i, i);
247+
}
248+
#endif
249+
250+
int main()
251+
{
252+
return 0;
253+
}

‎config.tests/x86_simd/x86_simd.pro

+5
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
SOURCES = main.cpp
2+
for (config, SIMD) {
3+
uc = $$upper($$config)
4+
DEFINES += QT_COMPILER_SUPPORTS_$${uc}
5+
}

‎configure.json

+19-1
Original file line numberDiff line numberDiff line change
@@ -218,7 +218,7 @@
218218
},
219219

220220
"testTypeAliases": {
221-
"compile": [ "library", "architecture" ]
221+
"compile": [ "library", "architecture", "x86SimdAlways" ]
222222
},
223223

224224
"tests": {
@@ -563,6 +563,11 @@
563563
"test": "avx512",
564564
"args": "AVX512=VBMI"
565565
},
566+
"x86SimdAlways": {
567+
"label": "Intrinsics without -mXXX argument",
568+
"type": "x86SimdAlways",
569+
"test": "x86_simd"
570+
},
566571
"mips_dsp": {
567572
"label": "MIPS DSP instructions",
568573
"type": "subarch",
@@ -1111,6 +1116,14 @@
11111116
{ "type": "define", "name": "QT_COMPILER_SUPPORTS_SHA", "value": 1 }
11121117
]
11131118
},
1119+
"x86SimdAlways": {
1120+
"label": "Intrinsics without -mXXX option",
1121+
"condition": "(arch.i386 || arch.x86_64) && tests.x86SimdAlways",
1122+
"output": [
1123+
"privateConfig",
1124+
{ "type": "define", "name": "QT_COMPILER_SUPPORTS_SIMD_ALWAYS", "value": 1 }
1125+
]
1126+
},
11141127
"mips_dsp": {
11151128
"label": "DSP",
11161129
"condition": "arch.mips && tests.mips_dsp",
@@ -1397,6 +1410,11 @@ Configure with '-qreal float' to create a build that is binary-compatible with 5
13971410
"args": "aesni f16c rdrnd shani",
13981411
"condition": "(arch.i386 || arch.x86_64)"
13991412
},
1413+
{
1414+
"type": "feature",
1415+
"args": "x86SimdAlways",
1416+
"condition": "(arch.i386 || arch.x86_64) && !config.msvc"
1417+
},
14001418
{
14011419
"type": "feature",
14021420
"args": "neon",

‎configure.pri

+13
Original file line numberDiff line numberDiff line change
@@ -395,6 +395,19 @@ defineTest(qtConfTest_buildParts) {
395395
return(true)
396396
}
397397

398+
defineTest(qtConfTest_x86SimdAlways) {
399+
configs =
400+
fpfx = $${currentConfig}.features
401+
simd = sse2 sse3 ssse3 sse4_1 sse4_2 rdrnd aesni shani avx avx2 avx512f \
402+
avx512er avx512cd avx512pf avx512dq avx512bw avx512vl avx512ifma avx512vbmi
403+
for (f, simd) {
404+
qtConfCheckFeature($$f)
405+
equals($${fpfx}.$${f}.available, true): configs += $$f
406+
}
407+
$${1}.literal_args = $$system_quote(SIMD=$$join(configs, " "))
408+
qtConfTest_compile($${1})
409+
}
410+
398411
# custom outputs
399412

400413
# this reloads the qmakespec as completely as reasonably possible.

‎src/corelib/tools/qsimd_p.h

+1-5
Original file line numberDiff line numberDiff line change
@@ -164,11 +164,7 @@
164164
# if !defined(__MIPS_DSPR2__) && defined(__mips_dspr2) && defined(Q_PROCESSOR_MIPS_32)
165165
# define __MIPS_DSPR2__
166166
# endif
167-
#elif (defined(Q_CC_INTEL) || defined(Q_CC_MSVC) \
168-
|| (defined(Q_CC_GNU) && !defined(Q_CC_CLANG) && Q_CC_GNU >= 409) \
169-
|| (defined(Q_CC_CLANG) && Q_CC_CLANG >= 308)) \
170-
&& !defined(QT_BOOTSTRAPPED)
171-
# define QT_COMPILER_SUPPORTS_SIMD_ALWAYS
167+
#elif defined(Q_PROCESSOR_X86) && defined(QT_COMPILER_SUPPORTS_SIMD_ALWAYS)
172168
# define QT_COMPILER_SUPPORTS_HERE(x) ((__ ## x ## __) || QT_COMPILER_SUPPORTS(x))
173169
# if defined(Q_CC_GNU) && !defined(Q_CC_INTEL)
174170
/* GCC requires attributes for a function */

0 commit comments

Comments
 (0)