aboutsummaryrefslogtreecommitdiffhomepage
path: root/src/core/SkUtilsArm.cpp
diff options
context:
space:
mode:
authorGravatar mtklein <mtklein@google.com>2016-04-15 08:40:22 -0700
committerGravatar Commit bot <commit-bot@chromium.org>2016-04-15 08:40:23 -0700
commit86498fbfcb93a9048bbe1c28cc0df40d8d0c96e9 (patch)
treea579fdee9a3cf330050dbe7fd0a3a25d49783693 /src/core/SkUtilsArm.cpp
parent2c7f24093a394ccbe54a7db60ba79af14682e7fa (diff)
Revert of Move CPU feature detection to its own file. (patchset #7 id:120001 of https://codereview.chromium.org/1890483002/ )
Reason for revert: many unexpected GM diffs across GPU+CPU configs on Windows (hopefully just text masks on GPU?). seems like we pick a different srcover variant in some places. Original issue's description: > Move CPU feature detection to its own file. > > - Moves CPU feature detection to its own file. > - Cleans up some redundant feature detection scattered around core/ and opts/. > - Can now detect a few new CPU features: > * F16C -> Intel f16<->f32 instructions, added between AVX and AVX2 > * FMA -> Intel FMA instructions, added at the same time as AVX2 > * VFP_FP16 -> ARM f16<->f32 instructions, quite common > * NEON_FMA -> ARM FMA instructions, also quite common > * SSE and SSE3... why not? > > This new internal API makes it very cheap to do fine-grained runtime CPU > feature detection. Redundant calls to SkCpu::Supports() should be eliminated > and it's hoistable out of loops. It compiles away entirely when we have the > appropriate instructions available at compile time. > > This means we can call it to guard even a little snippet of 1 or 2 instructions > right where needed and let inlining hoist the check (if any at all) up to > somewhere that doesn't hurt performance. I've explained how I made this work > in the private section of the new header. > > Once this lands and bakes a bit, I'll start following up with CLs to use it more > and to add a bunch of those little 1-2 instruction snippets we've been wanting, > e.g. cvtps2ph, cvtph2ps, ptest, pmulld, pmovzxbd, blendvps, pshufb, roundps > (for floor) on x86, and vcvt.f32.f16, vcvt.f16.f32 on ARM. > > BUG=skia: > GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1890483002 > CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot > > Committed: https://skia.googlesource.com/skia/+/872ea29357439f05b1f6995dd300fc054733e607 TBR=fmalita@chromium.org,herb@google.com,reed@google.com,mtklein@chromium.org # Skipping CQ checks because original CL landed less than 1 days ago. NOPRESUBMIT=true NOTREECHECKS=true NOTRY=true BUG=skia: Review URL: https://codereview.chromium.org/1892643003
Diffstat (limited to 'src/core/SkUtilsArm.cpp')
-rw-r--r--src/core/SkUtilsArm.cpp139
1 files changed, 138 insertions, 1 deletions
diff --git a/src/core/SkUtilsArm.cpp b/src/core/SkUtilsArm.cpp
index c29938fdfc..bf98fed476 100644
--- a/src/core/SkUtilsArm.cpp
+++ b/src/core/SkUtilsArm.cpp
@@ -5,4 +5,141 @@
* found in the LICENSE file.
*/
-// This file no longer needs to exist, but it's still referenced by Chrome's GYP / GN builds.
+#include "SkUtilsArm.h"
+
+#if SK_ARM_NEON_IS_DYNAMIC
+
+#include <unistd.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <string.h>
+#include <pthread.h>
+
+#if defined(SK_BUILD_FOR_ANDROID)
+# include <cpu-features.h>
+#endif
+
+// A function used to determine at runtime if the target CPU supports
+// the ARM NEON instruction set. This implementation is Linux-specific.
+static bool sk_cpu_arm_check_neon(void) {
+ // If we fail any of the following, assume we don't have NEON instructions
+ // This allows us to return immediately in case of error.
+ bool result = false;
+
+// Use the Android NDK's cpu-features helper library to detect NEON at runtime.
+// See http://crbug.com/164154 to see why this is needed in Chromium for Android.
+#ifdef SK_BUILD_FOR_ANDROID
+
+ result = (android_getCpuFeatures() & ANDROID_CPU_ARM_FEATURE_NEON) != 0;
+
+#else // SK_BUILD_FOR_ANDROID
+
+ // There is no user-accessible CPUID instruction on ARM that we can use.
+ // Instead, we must parse /proc/cpuinfo and look for the 'neon' feature.
+ // For example, here's a typical output (Nexus S running ICS 4.0.3):
+ /*
+ Processor : ARMv7 Processor rev 2 (v7l)
+ BogoMIPS : 994.65
+ Features : swp half thumb fastmult vfp edsp thumbee neon vfpv3
+ CPU implementer : 0x41
+ CPU architecture: 7
+ CPU variant : 0x2
+ CPU part : 0xc08
+ CPU revision : 2
+
+ Hardware : herring
+ Revision : 000b
+ Serial : 3833c77d6dc000ec
+ */
+ char buffer[4096];
+
+ do {
+ // open /proc/cpuinfo
+ int fd = TEMP_FAILURE_RETRY(open("/proc/cpuinfo", O_RDONLY));
+ if (fd < 0) {
+ SkDebugf("Could not open /proc/cpuinfo: %s\n", strerror(errno));
+ break;
+ }
+
+ // Read the file. To simplify our search, we're going to place two
+ // sentinel '\n' characters: one at the start of the buffer, and one at
+ // the end. This means we reserve the first and last buffer bytes.
+ buffer[0] = '\n';
+ int size = TEMP_FAILURE_RETRY(read(fd, buffer+1, sizeof(buffer)-2));
+ close(fd);
+
+ if (size < 0) { // should not happen
+ SkDebugf("Could not read /proc/cpuinfo: %s\n", strerror(errno));
+ break;
+ }
+
+ SkDebugf("START /proc/cpuinfo:\n%.*s\nEND /proc/cpuinfo\n",
+ size, buffer+1);
+
+ // Compute buffer limit, and place final sentinel
+ char* buffer_end = buffer + 1 + size;
+ buffer_end[0] = '\n';
+
+ // Now, find a line that starts with "Features", i.e. look for
+ // '\nFeatures ' in our buffer.
+ const char features[] = "\nFeatures\t";
+ const size_t features_len = sizeof(features)-1;
+
+ char* line = (char*) memmem(buffer, buffer_end - buffer,
+ features, features_len);
+ if (line == nullptr) { // Weird, no Features line, bad kernel?
+ SkDebugf("Could not find a line starting with 'Features'"
+ "in /proc/cpuinfo ?\n");
+ break;
+ }
+
+ line += features_len; // Skip the "\nFeatures\t" prefix
+
+ // Find the end of the current line
+ char* line_end = (char*) memchr(line, '\n', buffer_end - line);
+ if (line_end == nullptr)
+ line_end = buffer_end;
+
+ // Now find an instance of 'neon' in the flags list. We want to
+ // ensure it's only 'neon' and not something fancy like 'noneon'
+ // so check that it follows a space.
+ const char neon[] = " neon";
+ const size_t neon_len = sizeof(neon)-1;
+ const char* flag = (const char*) memmem(line, line_end - line,
+ neon, neon_len);
+ if (flag == nullptr)
+ break;
+
+ // Ensure it is followed by a space or a newline.
+ if (flag[neon_len] != ' ' && flag[neon_len] != '\n')
+ break;
+
+ // Fine, we support Arm NEON !
+ result = true;
+
+ } while (0);
+
+#endif // SK_BUILD_FOR_ANDROID
+
+ if (result) {
+ SkDEBUGF(("Device supports ARM NEON instructions!\n"));
+ } else {
+ SkDEBUGF(("Device does NOT support ARM NEON instructions!\n"));
+ }
+ return result;
+}
+
+static pthread_once_t sOnce;
+static bool sHasArmNeon;
+
+// called through pthread_once()
+void sk_cpu_arm_probe_features(void) {
+ sHasArmNeon = sk_cpu_arm_check_neon();
+}
+
+bool sk_cpu_arm_has_neon(void) {
+ pthread_once(&sOnce, sk_cpu_arm_probe_features);
+ return sHasArmNeon;
+}
+
+#endif // SK_ARM_NEON_IS_DYNAMIC