Merge commit '792a5a5dd1946a327f22b2daba05c6645dfa8037'

src/Makefile

@@ -9,15 +9,20 @@ DFLAGS = -O2 -g -Werror -march=native
 
 # ARCH related flags
 ifeq ($(ARCH), x86_64)
-    ASMFLAGS += -DARCH_X86_64=1
+    TARGET_CPU_BITS := 64
-    DFLAGS += -m64
+    TARGET_CPU_ARCH := x86
-    LDFLAGS += -m64
+else ifeq ($(ARCH), ppc64)
+    TARGET_CPU_BITS := 64
+    TARGET_CPU_ARCH := ppc
 else
-    ASMFLAGS += -DARCH_X86_64=0
+    #safe (?) defaults
-    DFLAGS += -m32
+    TARGET_CPU_BITS := 32
-    LDFLAGS += -m32
+    TARGET_CPU_ARCH := x86
 endif
 
+DFLAGS += -m$(TARGET_CPU_BITS)
+LDFLAGS += -m$(TARGET_CPU_BITS)
+
 # Windows (cygwin/mingw) specific flags
 ifneq ( ,$(findstring Windows, $(OS)))
   ifeq ($(ARCH), x86_64)
@@ -48,8 +53,6 @@ CC = gcc
 CCFLAGS = $(DFLAGS) -I. -Wall -Wtype-limits
 LDFLAGS += -lm
 LD = gcc -pthread -lrt
-YASM = yasm
-ASMOBJS = cpu.o
 OBJS = interface_main.o encmain.o bitstream.o cabac.o config.o context.o encoder.o filter.o inter.o intra.o nal.o picture.o rdo.o sao.o scalinglist.o search.o strategyselector.o tables.o threadqueue.o transform.o
 PROG  = ./kvazaar
 PROGS = $(PROG)
@@ -62,9 +65,6 @@ all: $(PROGS)
 $(PROG): $(OBJS) $(ASMOBJS)
 	$(LD) $^ $(LDFLAGS) -o $@
 
-cpu.o: x86/cpu.asm
-	$(YASM) $(ASMFLAGS) x86/cpu.asm -o cpu.o
-
 %.o: %.c Makefile
 	$(CC) $(CCFLAGS) -c $< -o $@
 
@@ -73,6 +73,7 @@ cpu.o: x86/cpu.asm
 
 
 clean:
+	echo $(ARCH)
 	rm -f $(OBJS) $(PROGS) $(ASMOBJS) $(DEPS)
 
 -include $(DEPS)

src/global.h

@@ -135,7 +135,7 @@ typedef int16_t coefficient;
 #define SIZE_NxN   3
 #define SIZE_NONE  15
 
-#define MAX_TILES_PER_DIM 16
+#define MAX_TILES_PER_DIM 48
 #define MAX_SLICES 16
 
 /* Inlining functions */

src/intra.c

@@ -718,7 +718,6 @@ void intra_recon_lcu_chroma(encoder_state * const encoder_state, int x, int y, i
   cu_info *cur_cu = &lcu->cu[LCU_CU_OFFSET + (lcu_px.x>>3) + (lcu_px.y>>3)*LCU_T_CU_WIDTH];
   const int8_t width = LCU_WIDTH >> depth;
   const int8_t width_c = (depth == MAX_PU_DEPTH ? width : width / 2);
-  const int pu_index = PU_INDEX(x >> 2, y >> 2);
 
   if (depth == 0 || cur_cu->tr_depth > depth) {
     int offset = width / 2;

src/strategyselector.c

@@ -155,7 +155,28 @@ static void* strategyselector_choose_for(const strategy_list * const strategies,
 }
 
 #if COMPILE_INTEL
-#include "x86/cpu.h"
+
+#if defined(__GNUC__)
+#include <cpuid.h>
+#else
+#include <intrin.h>
+//Adapter from __cpuid (VS) to __get_cpuid (GNU C).
+__inline int __get_cpuid(unsigned int __level, unsigned int *__eax, unsigned int *__ebx, unsigned int *__ecx, unsigned int *__edx) {
+  int CPUInfo[4] = {*__eax, *__ebx, *__ecx, *__edx};
+  __cpuid(CPUInfo, 0);
+  // check if the CPU supports the cpuid instruction.
+  if (CPUInfo[0] != 0) {
+    __cpuid(CPUInfo, __level);
+    *__eax = CPUInfo[0];
+    *__ebx = CPUInfo[1];
+    *__ecx = CPUInfo[2];
+    *__edx = CPUInfo[3];
+    return 1;
+  }
+  return 0;
+}
+#endif //defined(__GNUC__)
+
 #endif
 
 static void set_hardware_flags() {
@@ -167,12 +188,12 @@ static void set_hardware_flags() {
   
 #if COMPILE_INTEL
   {
-    int ecx = 0,edx =0;
+    unsigned int eax = 0, ebx = 0, ecx = 0, edx =0;
     /* CPU feature bits */
     enum { BIT_SSE3 = 0,BIT_SSSE3 = 9, BIT_SSE41 = 19, BIT_SSE42 = 20, BIT_MMX = 24, BIT_SSE = 25, BIT_SSE2 = 26, BIT_AVX = 28};
 
     // Dig CPU features with cpuid
-    kvz_cpu_cpuid(&ecx,&edx);
+    __get_cpuid(1, &eax, &ebx, &ecx, &edx);
     
     // EDX
     if (edx & (1<<BIT_MMX))   g_hardware_flags.intel_flags.mmx = 1;

tools/plot-threadqueue-log.py

@@ -106,7 +106,7 @@ class LogParser:
     else:
       return float(value)
 
-  def __init__(self):
+  def __init__(self, filename):
     re_thread = re.compile(r'^\t([0-9]+)\t-\t([0-9\.]+)\t(\+?)([0-9\.]+)\t-\tthread$')
     re_job = re.compile(r'^([^\t]+)\t([0-9]+)\t([0-9\.]+)\t(\+?)([0-9\.]+)\t(\+?)([0-9\.]+)\t(\+?)([0-9\.]+)\t(.*)$')
     re_dep = re.compile(r'^(.+)->(.+)$')
@@ -118,7 +118,7 @@ class LogParser:
     objects = []
     threads = {}
 
-    for line in open('threadqueue.log','r').readlines():
+    for line in open(filename,'r').readlines():
       m = re_thread.match(line)
       if m:
         g = m.groups()
@@ -281,6 +281,10 @@ class LogParser:
 
 
 if __name__ == '__main__':
-  l = LogParser()
+  import sys
+  if len(sys.argv) > 1:
+    l = LogParser(sys.argv[1])
+  else:
+    l = LogParser('threadqueue.log')
   l.plot_picture_wise_wpp()
   #l.plot_threads()