Merge remote branch 'internal/upstream' into HEAD

build/make/configure.sh

@@ -980,6 +980,9 @@ EOF
         esac
     fi
 
+    # for sysconf(3) and friends.
+    check_header unistd.h
+
     # glibc needs these
     if enabled linux; then
         add_cflags -D_LARGEFILE_SOURCE

configure

@@ -211,6 +211,7 @@ HAVE_LIST="
     alt_tree_layout
     pthread_h
     sys_mman_h
+    unistd_h
 "
 EXPERIMENT_LIST="
     extend_qrange

vp8/common/arm/neon/loopfilter_neon.asm

@@ -308,7 +308,6 @@
 ; q9    q2
 ; q10   q3
 |vp8_loop_filter_neon| PROC
-    ldr         r12, _lf_coeff_
 
     ; vp8_filter_mask
     vabd.u8     q11, q3, q4                 ; abs(p3 - p2)
@@ -339,7 +338,7 @@
     vqadd.u8    q9, q9, q2                  ; a = b + a
     vcge.u8     q9, q0, q9                  ; (a > flimit * 2 + limit) * -1
 
-    vld1.u8     {q0}, [r12]!
+    vmov.u8     q0, #0x80                   ; 0x80
 
     ; vp8_filter() function
     ; convert to signed
@@ -348,7 +347,7 @@
     veor        q5, q5, q0                  ; ps1
     veor        q8, q8, q0                  ; qs1
 
-    vld1.u8     {q10}, [r12]!
+    vmov.u8     q10, #3                     ; #3
 
     vsubl.s8    q2, d14, d12                ; ( qs0 - ps0)
     vsubl.s8    q11, d15, d13
@@ -367,7 +366,7 @@
     vaddw.s8    q2, q2, d2
     vaddw.s8    q11, q11, d3
 
-    vld1.u8     {q9}, [r12]!
+    vmov.u8     q9, #4                      ; #4
 
     ; vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0))
     vqmovn.s16  d2, q2
@@ -399,12 +398,4 @@
 
 ;-----------------
 
-_lf_coeff_
-    DCD     lf_coeff
-lf_coeff
-    DCD     0x80808080, 0x80808080, 0x80808080, 0x80808080
-    DCD     0x03030303, 0x03030303, 0x03030303, 0x03030303
-    DCD     0x04040404, 0x04040404, 0x04040404, 0x04040404
-    DCD     0x01010101, 0x01010101, 0x01010101, 0x01010101
-
     END

vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.asm

@@ -22,20 +22,19 @@
 ; r1    int p, //pitch
 ; r2    const signed char *flimit,
 ; r3    const signed char *limit,
-; stack(r4) const signed char *thresh,
+; stack(r4) const signed char *thresh (unused)
 ; //stack(r5)   int count --unused
 
 |vp8_loop_filter_simple_horizontal_edge_neon| PROC
     sub         r0, r0, r1, lsl #1          ; move src pointer down by 2 lines
 
-    ldr         r12, _lfhy_coeff_
     vld1.u8     {q5}, [r0], r1              ; p1
     vld1.s8     {d2[], d3[]}, [r2]          ; flimit
     vld1.s8     {d26[], d27[]}, [r3]        ; limit -> q13
     vld1.u8     {q6}, [r0], r1              ; p0
-    vld1.u8     {q0}, [r12]!                ; 0x80
+    vmov.u8     q0, #0x80                   ; 0x80
     vld1.u8     {q7}, [r0], r1              ; q0
-    vld1.u8     {q10}, [r12]!               ; 0x03
+    vmov.u8     q10, #0x03                  ; 0x03
     vld1.u8     {q8}, [r0]                  ; q1
 
     ;vp8_filter_mask() function
@@ -66,7 +65,7 @@
     vadd.s16    q11, q2, q2                 ;  3 * ( qs0 - ps0)
     vadd.s16    q12, q3, q3
 
-    vld1.u8     {q9}, [r12]!                ; 0x04
+    vmov.u8     q9, #0x04                   ; 0x04
 
     vadd.s16    q2, q2, q11
     vadd.s16    q3, q3, q12
@@ -105,11 +104,4 @@
 
 ;-----------------
 
-_lfhy_coeff_
-    DCD     lfhy_coeff
-lfhy_coeff
-    DCD     0x80808080, 0x80808080, 0x80808080, 0x80808080
-    DCD     0x03030303, 0x03030303, 0x03030303, 0x03030303
-    DCD     0x04040404, 0x04040404, 0x04040404, 0x04040404
-
     END

vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.asm

@@ -22,7 +22,7 @@
 ; r1    int p, //pitch
 ; r2    const signed char *flimit,
 ; r3    const signed char *limit,
-; stack(r4) const signed char *thresh,
+; stack(r4) const signed char *thresh (unused)
 ; //stack(r5)   int count --unused
 
 |vp8_loop_filter_simple_vertical_edge_neon| PROC
@@ -32,7 +32,6 @@
     vld1.s8     {d2[], d3[]}, [r2]          ; flimit
     vld1.s8     {d26[], d27[]}, [r3]        ; limit -> q13
     vld4.8      {d6[1], d7[1], d8[1], d9[1]}, [r0], r1
-    ldr         r12, _vlfy_coeff_
     vld4.8      {d6[2], d7[2], d8[2], d9[2]}, [r0], r1
     vld4.8      {d6[3], d7[3], d8[3], d9[3]}, [r0], r1
     vld4.8      {d6[4], d7[4], d8[4], d9[4]}, [r0], r1
@@ -41,11 +40,11 @@
     vld4.8      {d6[7], d7[7], d8[7], d9[7]}, [r0], r1
 
     vld4.8      {d10[0], d11[0], d12[0], d13[0]}, [r0], r1
-    vld1.u8     {q0}, [r12]!                ; 0x80
+    vmov.u8     q0, #0x80                ; 0x80
     vld4.8      {d10[1], d11[1], d12[1], d13[1]}, [r0], r1
-    vld1.u8     {q11}, [r12]!               ; 0x03
+    vmov.u8     q11, #0x03              ; 0x03
     vld4.8      {d10[2], d11[2], d12[2], d13[2]}, [r0], r1
-    vld1.u8     {q12}, [r12]!               ; 0x04
+    vmov.u8     q12, #0x04               ; 0x04
     vld4.8      {d10[3], d11[3], d12[3], d13[3]}, [r0], r1
     vld4.8      {d10[4], d11[4], d12[4], d13[4]}, [r0], r1
     vld4.8      {d10[5], d11[5], d12[5], d13[5]}, [r0], r1
@@ -146,11 +145,4 @@
 
 ;-----------------
 
-_vlfy_coeff_
-    DCD     vlfy_coeff
-vlfy_coeff
-    DCD     0x80808080, 0x80808080, 0x80808080, 0x80808080
-    DCD     0x03030303, 0x03030303, 0x03030303, 0x03030303
-    DCD     0x04040404, 0x04040404, 0x04040404, 0x04040404
-
     END

vp8/common/arm/neon/mbloopfilter_neon.asm

@@ -372,7 +372,6 @@
 ; q10   q3
 
 |vp8_mbloop_filter_neon| PROC
-    ldr         r12, _mblf_coeff_
 
     ; vp8_filter_mask
     vabd.u8     q11, q3, q4                 ; abs(p3 - p2)
@@ -396,7 +395,7 @@
 
     vld1.s8     {d4[], d5[]}, [r2]          ; flimit
 
-    vld1.u8     {q0}, [r12]!
+    vmov.u8     q0, #0x80                   ; 0x80
 
     vadd.u8     q2, q2, q2                  ; flimit * 2
     vadd.u8     q2, q2, q1                  ; flimit * 2 +  limit
@@ -431,12 +430,12 @@
     vadd.s16    q2, q2, q10
     vadd.s16    q13, q13, q11
 
-    vld1.u8     {q12}, [r12]!               ; #3
+    vmov.u8     q12, #3                     ; #3
 
     vaddw.s8    q2, q2, d2                  ; vp8_filter + 3 * ( qs0 - ps0)
     vaddw.s8    q13, q13, d3
 
-    vld1.u8     {q11}, [r12]!               ; #4
+    vmov.u8     q11, #4                     ; #4
 
     ; vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0))
     vqmovn.s16  d2, q2
@@ -444,16 +443,16 @@
 
     vand        q1, q1, q15                 ; vp8_filter &= mask
 
-    vld1.u8     {q15}, [r12]!               ; #63
+    vmov.u16    q15, #63                    ; #63
-    ;
+
     vand        q13, q1, q14                ; Filter2 &= hev
 
-    vld1.u8     {d7}, [r12]!                ; #9
+    vmov.u8     d7, #9                      ; #9
 
     vqadd.s8    q2, q13, q11                ; Filter1 = clamp(Filter2+4)
     vqadd.s8    q13, q13, q12               ; Filter2 = clamp(Filter2+3)
 
-    vld1.u8     {d6}, [r12]!                ; #18
+    vmov.u8     d6, #18                     ; #18
 
     vshr.s8     q2, q2, #3                  ; Filter1 >>= 3
     vshr.s8     q13, q13, #3                ; Filter2 >>= 3
@@ -463,7 +462,7 @@
 
     vqsub.s8    q7, q7, q2                  ; qs0 = clamp(qs0 - Filter1)
 
-    vld1.u8     {d5}, [r12]!                ; #27
+    vmov.u8     d5, #27                     ; #27
 
     vqadd.s8    q6, q6, q13                 ; ps0 = clamp(ps0 + Filter2)
 
@@ -507,14 +506,4 @@
 
 ;-----------------
 
-_mblf_coeff_
-    DCD     mblf_coeff
-mblf_coeff
-    DCD     0x80808080, 0x80808080, 0x80808080, 0x80808080
-    DCD     0x03030303, 0x03030303, 0x03030303, 0x03030303
-    DCD     0x04040404, 0x04040404, 0x04040404, 0x04040404
-    DCD     0x003f003f, 0x003f003f, 0x003f003f, 0x003f003f
-    DCD     0x09090909, 0x09090909, 0x12121212, 0x12121212
-    DCD     0x1b1b1b1b, 0x1b1b1b1b
-
     END

vp8/common/generic/systemdependent.c

@@ -17,9 +17,54 @@
 #include "vp8/common/idct.h"
 #include "vp8/common/onyxc_int.h"
 
+#if CONFIG_MULTITHREAD
+#if HAVE_UNISTD_H
+#include <unistd.h>
+#elif defined(_WIN32)
+#include <windows.h>
+typedef void (WINAPI *PGNSI)(LPSYSTEM_INFO);
+#endif
+#endif
+
 extern void vp8_arch_x86_common_init(VP8_COMMON *ctx);
 extern void vp8_arch_arm_common_init(VP8_COMMON *ctx);
 
+#if CONFIG_MULTITHREAD
+static int get_cpu_count()
+{
+    int core_count = 16;
+
+#if HAVE_UNISTD_H
+#if defined(_SC_NPROCESSORS_ONLN)
+    core_count = sysconf(_SC_NPROCESSORS_ONLN);
+#elif defined(_SC_NPROC_ONLN)
+    core_count = sysconf(_SC_NPROC_ONLN);
+#endif
+#elif defined(_WIN32)
+    {
+        PGNSI pGNSI;
+        SYSTEM_INFO sysinfo;
+
+        /* Call GetNativeSystemInfo if supported or
+         * GetSystemInfo otherwise. */
+
+        pGNSI = (PGNSI) GetProcAddress(
+                GetModuleHandle(TEXT("kernel32.dll")), "GetNativeSystemInfo");
+        if (pGNSI != NULL)
+            pGNSI(&sysinfo);
+        else
+            GetSystemInfo(&sysinfo);
+
+        core_count = sysinfo.dwNumberOfProcessors;
+    }
+#else
+    /* other platforms */
+#endif
+
+    return core_count > 0 ? core_count : 1;
+}
+#endif
+
 void vp8_machine_specific_config(VP8_COMMON *ctx)
 {
 #if CONFIG_RUNTIME_CPU_DETECT
@@ -98,4 +143,7 @@ void vp8_machine_specific_config(VP8_COMMON *ctx)
 #endif
 
 
+#if CONFIG_MULTITHREAD
+    ctx->processor_core_count = get_cpu_count();
+#endif /* CONFIG_MULTITHREAD */
 }

vp8/common/onyxc_int.h

@@ -195,6 +195,9 @@ typedef struct VP8Common
 
 #if CONFIG_RUNTIME_CPU_DETECT
     VP8_COMMON_RTCD rtcd;
+#endif
+#if CONFIG_MULTITHREAD
+    int processor_core_count;
 #endif
     struct postproc_state  postproc_state;
 } VP8_COMMON;

vp8/decoder/threading.c

@@ -439,12 +439,18 @@ void vp8_decoder_create_threads(VP8D_COMP *pbi)
 
     pbi->b_multithreaded_rd = 0;
     pbi->allocated_decoding_thread_count = 0;
-    core_count = (pbi->max_threads > 16) ? 16 : pbi->max_threads;
+
+    /* limit decoding threads to the max number of token partitions */
+    core_count = (pbi->max_threads > 8) ? 8 : pbi->max_threads;
+
+    /* limit decoding threads to the available cores */
+    if (core_count > pbi->common.processor_core_count)
+        core_count = pbi->common.processor_core_count;
 
     if (core_count > 1)
     {
         pbi->b_multithreaded_rd = 1;
-        pbi->decoding_thread_count = core_count -1;
+        pbi->decoding_thread_count = core_count - 1;
 
         CHECK_MEM_ERROR(pbi->h_decoding_thread, vpx_malloc(sizeof(pthread_t) * pbi->decoding_thread_count));
         CHECK_MEM_ERROR(pbi->h_event_start_decoding, vpx_malloc(sizeof(sem_t) * pbi->decoding_thread_count));

vp8/encoder/ethreading.c

@@ -459,15 +459,15 @@ void vp8cx_create_encoder_threads(VP8_COMP *cpi)
 
     cpi->b_multi_threaded = 0;
     cpi->encoding_thread_count = 0;
-    cpi->processor_core_count = 32; //vp8_get_proc_core_count();
 
-    if (cpi->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
+    if (cm->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
     {
         int ithread;
         int th_count = cpi->oxcf.multi_threaded - 1;
 
-        if (cpi->oxcf.multi_threaded > cpi->processor_core_count)
+        /* don't allocate more threads than cores available */
-            th_count = cpi->processor_core_count - 1;
+        if (cpi->oxcf.multi_threaded > cm->processor_core_count)
+            th_count = cm->processor_core_count - 1;
 
         /* we have th_count + 1 (main) threads processing one row each */
         /* no point to have more threads than the sync range allows */
@@ -514,6 +514,7 @@ void vp8cx_create_encoder_threads(VP8_COMP *cpi)
             LPFTHREAD_DATA * lpfthd = &cpi->lpf_thread_data;
 
             sem_init(&cpi->h_event_start_lpf, 0, 0);
+            sem_init(&cpi->h_event_end_picklpf, 0, 0);
             sem_init(&cpi->h_event_end_lpf, 0, 0);
 
             lpfthd->ptr1 = (void *)cpi;
@@ -547,6 +548,7 @@ void vp8cx_remove_encoder_threads(VP8_COMP *cpi)
 
         sem_destroy(&cpi->h_event_end_encoding);
         sem_destroy(&cpi->h_event_end_lpf);
+        sem_destroy(&cpi->h_event_end_picklpf);
         sem_destroy(&cpi->h_event_start_lpf);
 
         //free thread related resources

vp8/encoder/onyx_if.c

@@ -3211,7 +3211,7 @@ void loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm)
 
 #if CONFIG_MULTITHREAD
     if (cpi->b_multi_threaded)
-        sem_post(&cpi->h_event_end_lpf); /* signal that we have set filter_level */
+        sem_post(&cpi->h_event_end_picklpf); /* signal that we have set filter_level */
 #endif
 
     if (cm->filter_level > 0)
@@ -4221,7 +4221,7 @@ static void encode_frame_to_data_rate
 #if CONFIG_MULTITHREAD
     /* wait that filter_level is picked so that we can continue with stream packing */
     if (cpi->b_multi_threaded)
-        sem_wait(&cpi->h_event_end_lpf);
+        sem_wait(&cpi->h_event_end_picklpf);
 #endif
 
     // build the bitstream

vp8/encoder/onyx_int.h

@@ -580,7 +580,6 @@ typedef struct
     // multithread data
     int * mt_current_mb_col;
     int mt_sync_range;
-    int processor_core_count;
     int b_multi_threaded;
     int encoding_thread_count;
 
@@ -595,6 +594,7 @@ typedef struct
     sem_t *h_event_start_encoding;
     sem_t h_event_end_encoding;
     sem_t h_event_start_lpf;
+    sem_t h_event_end_picklpf;
     sem_t h_event_end_lpf;
 #endif