[x265] [PATCH] Fix testbench crash with dhdr10 enabled

[bhavna at multicorewareinc.com]

# HG changeset patch
# User Bhavna Hariharan <bhavna at multicorewareinc.com>
# Date 1501680985 -19800
#      Wed Aug 02 19:06:25 2017 +0530
# Node ID 041fcaf820659d504d887cfaa239a02988812c14
# Parent  d11482e5fedbcdaf62ee3c6872f43827d99ad181
Fix testbench crash with dhdr10 enabled

__rdtsc() was being defined multiple times causing failure in testbench.
This patch renames the function and duplicates the REPORT_SPEEDUP macro that
supports GNU compiler.

Display warning when dhdr10 is enabled with MSVC version that does not
support c++11.

diff -r d11482e5fedb -r 041fcaf82065 source/CMakeLists.txt
--- a/source/CMakeLists.txt	Mon Jul 24 11:15:38 2017 +0530
+++ b/source/CMakeLists.txt	Wed Aug 02 19:06:25 2017 +0530
@@ -184,6 +184,9 @@
 endif()
 # this option is to enable the inclusion of dynamic HDR10 library to the libx265 compilation
 option(ENABLE_HDR10_PLUS "Enable dynamic HDR10 compilation" OFF)
+if(MSVC AND (MSVC_VERSION LESS 1800) AND ENABLE_HDR10_PLUS)
+    message(FATAL_ERROR "MSVC version 12.0 or above required to support hdr10plus")
+endif()
 if(GCC)
     add_definitions(-Wall -Wextra -Wshadow)
     add_definitions(-D__STDC_LIMIT_MACROS=1)
diff -r d11482e5fedb -r 041fcaf82065 source/dynamicHDR10/metadataFromJson.cpp
--- a/source/dynamicHDR10/metadataFromJson.cpp	Mon Jul 24 11:15:38 2017 +0530
+++ b/source/dynamicHDR10/metadataFromJson.cpp	Wed Aug 02 19:06:25 2017 +0530
@@ -34,8 +34,6 @@
 #include "BasicStructures.h"
 #include "SeiMetadataDictionary.h"
 
-#define M_PI 3.14159265358979323846
-
 using namespace SeiMetadataDictionary;
 
 class metadataFromJson::DynamicMetaIO
diff -r d11482e5fedb -r 041fcaf82065 source/test/testharness.h
--- a/source/test/testharness.h	Mon Jul 24 11:15:38 2017 +0530
+++ b/source/test/testharness.h	Wed Aug 02 19:06:25 2017 +0530
@@ -70,7 +70,7 @@
 #include <intrin.h>
 #elif defined(__GNUC__)
 /* fallback for older GCC/MinGW */
-static inline uint32_t __rdtsc(void)
+static inline uint32_t rdtsc(void)
 {
     uint32_t a = 0;
 
@@ -92,6 +92,41 @@
 // Adapted from checkasm.c, runs each optimized primitive four times, measures rdtsc
 // and discards invalid times.  Repeats 1000 times to get a good average.  Then measures
 // the C reference with fewer runs and reports X factor and average cycles.
+
+// Duplicate macro to avoid conflict with in-built function __rdtsc()
+#if defined(__GNUC__)
+#define REPORT_SPEEDUP(RUNOPT, RUNREF, ...) \
+    { \
+        uint32_t cycles = 0; int runs = 0; \
+        RUNOPT(__VA_ARGS__); \
+        for (int ti = 0; ti < BENCH_RUNS; ti++) { \
+            uint32_t t0 = (uint32_t)rdtsc(); \
+            RUNOPT(__VA_ARGS__); \
+            RUNOPT(__VA_ARGS__); \
+            RUNOPT(__VA_ARGS__); \
+            RUNOPT(__VA_ARGS__); \
+            uint32_t t1 = (uint32_t)rdtsc() - t0; \
+            if (t1 * runs <= cycles * 4 && ti > 0) { cycles += t1; runs++; } \
+                } \
+        uint32_t refcycles = 0; int refruns = 0; \
+        RUNREF(__VA_ARGS__); \
+        for (int ti = 0; ti < BENCH_RUNS / 4; ti++) { \
+            uint32_t t0 = (uint32_t)rdtsc(); \
+            RUNREF(__VA_ARGS__); \
+            RUNREF(__VA_ARGS__); \
+            RUNREF(__VA_ARGS__); \
+            RUNREF(__VA_ARGS__); \
+            uint32_t t1 = (uint32_t)rdtsc() - t0; \
+            if (t1 * refruns <= refcycles * 4 && ti > 0) { refcycles += t1; refruns++; } \
+                } \
+        x265_emms(); \
+        float optperf = (10.0f * cycles / runs) / 4; \
+        float refperf = (10.0f * refcycles / refruns) / 4; \
+        printf("\t%3.2fx ", refperf / optperf); \
+        printf("\t %-8.2lf \t %-8.2lf\n", optperf, refperf); \
+    }
+
+#else
 #define REPORT_SPEEDUP(RUNOPT, RUNREF, ...) \
     { \
         uint32_t cycles = 0; int runs = 0; \
@@ -123,6 +158,7 @@
         printf("\t %-8.2lf \t %-8.2lf\n", optperf, refperf); \
     }
 
+#endif
 extern "C" {
 #if X265_ARCH_X86
 int PFX(stack_pagealign)(int (*func)(), int align);
-------------- next part --------------
A non-text attachment was scrubbed...
Name: x265-clone.patch
Type: text/x-patch
Size: 3954 bytes
Desc: not available
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20170802/e6e7031d/attachment.bin>