libvpx: Pull from upstream

source/libvpx/vp9/encoder/x86/vp9_quantize_ssse3_x86_64.asm

 ;
 ;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 ;
 ;  Use of this source code is governed by a BSD-style license
 ;  that can be found in the LICENSE file in the root of the source
 ;  tree. An additional intellectual property rights grant can be found
 ;  in the file PATENTS.  All contributing project authors may
 ;  be found in the AUTHORS file in the root of the source tree.
 ;
 
@@ skipping 216 matching lines @@
 
   ; actual quantize loop - setup pointers, rounders, etc.
   movifnidn                   coeffq, coeffmp
   movifnidn                  ncoeffq, ncoeffmp
   mov                             r2, dequantmp
   movifnidn                    zbinq, zbinmp
   movifnidn                   roundq, roundmp
   movifnidn                   quantq, quantmp
   mova                            m1, [roundq]             ; m1 = round
   mova                            m2, [quantq]             ; m2 = quant
-%ifidn %1, b_32x32
-; TODO(jingning) to be continued with 32x32 quantization process
+%ifidn %1, fp_32x32
   pcmpeqw                         m5, m5
   psrlw                           m5, 15
-  paddw                           m0, m5
   paddw                           m1, m5
-  psrlw                           m0, 1                    ; m0 = (m0 + 1) / 2
   psrlw                           m1, 1                    ; m1 = (m1 + 1) / 2
 %endif
   mova                            m3, [r2q]                ; m3 = dequant
   mov                             r3, qcoeffmp
   mov                             r4, dqcoeffmp
   mov                             r5, iscanmp
-%ifidn %1, b_32x32
-  psllw                           m4, 1
+%ifidn %1, fp_32x32
+  psllw                           m2, 1
 %endif
   pxor                            m5, m5                   ; m5 = dedicated zero
   DEFINE_ARGS coeff, ncoeff, d1, qcoeff, dqcoeff, iscan, d2, d3, d4, d5, d6, eob
   lea                         coeffq, [  coeffq+ncoeffq*2]
   lea                         iscanq, [  iscanq+ncoeffq*2]
   lea                        qcoeffq, [ qcoeffq+ncoeffq*2]
   lea                       dqcoeffq, [dqcoeffq+ncoeffq*2]
   neg                        ncoeffq
 
   ; get DC and first 15 AC coeffs
   mova                            m9, [  coeffq+ncoeffq*2+ 0] ; m9 = c[i]
   mova                           m10, [  coeffq+ncoeffq*2+16] ; m10 = c[i]
   pabsw                           m6, m9                   ; m6 = abs(m9)
   pabsw                          m11, m10                  ; m11 = abs(m10)
   pcmpeqw                         m7, m7
 
   paddsw                          m6, m1                   ; m6 += round
   punpckhqdq                      m1, m1
   paddsw                         m11, m1                   ; m11 += round
   pmulhw                          m8, m6, m2               ; m8 = m6*q>>16
   punpckhqdq                      m2, m2
   pmulhw                         m13, m11, m2              ; m13 = m11*q>>16
   psignw                          m8, m9                   ; m8 = reinsert sign
   psignw                         m13, m10                  ; m13 = reinsert sign
   mova        [qcoeffq+ncoeffq*2+ 0], m8
   mova        [qcoeffq+ncoeffq*2+16], m13
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
   pabsw                           m8, m8
   pabsw                          m13, m13
 %endif
   pmullw                          m8, m3                   ; dqc[i] = qc[i] * q
   punpckhqdq                      m3, m3
   pmullw                         m13, m3                   ; dqc[i] = qc[i] * q
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
   psrlw                           m8, 1
   psrlw                          m13, 1
   psignw                          m8, m9
   psignw                         m13, m10
+  psrlw                           m0, m3, 2
 %endif
   mova       [dqcoeffq+ncoeffq*2+ 0], m8
   mova       [dqcoeffq+ncoeffq*2+16], m13
   pcmpeqw                         m8, m5                   ; m8 = c[i] == 0
   pcmpeqw                        m13, m5                   ; m13 = c[i] == 0
   mova                            m6, [  iscanq+ncoeffq*2+ 0] ; m6 = scan[i]
   mova                           m11, [  iscanq+ncoeffq*2+16] ; m11 = scan[i]
   psubw                           m6, m7                   ; m6 = scan[i] + 1
   psubw                          m11, m7                   ; m11 = scan[i] + 1
   pandn                           m8, m6                   ; m8 = max(eob)
   pandn                          m13, m11                  ; m13 = max(eob)
   pmaxsw                          m8, m13
   add                        ncoeffq, mmsize
   jz .accumulate_eob
 
 .ac_only_loop:
   mova                            m9, [  coeffq+ncoeffq*2+ 0] ; m9 = c[i]
   mova                           m10, [  coeffq+ncoeffq*2+16] ; m10 = c[i]
   pabsw                           m6, m9                   ; m6 = abs(m9)
   pabsw                          m11, m10                  ; m11 = abs(m10)
-  pcmpeqw                         m7, m7
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
+  pcmpgtw                         m7, m6,  m0
+  pcmpgtw                        m12, m11, m0
   pmovmskb                        r6, m7
-  pmovmskb                        r2, m7
+  pmovmskb                        r2, m12
+
   or                              r6, r2
   jz .skip_iter
 %endif
+  pcmpeqw                         m7, m7
+
   paddsw                          m6, m1                   ; m6 += round
   paddsw                         m11, m1                   ; m11 += round
   pmulhw                         m14, m6, m2               ; m14 = m6*q>>16
   pmulhw                         m13, m11, m2              ; m13 = m11*q>>16
   psignw                         m14, m9                   ; m14 = reinsert sign
   psignw                         m13, m10                  ; m13 = reinsert sign
   mova        [qcoeffq+ncoeffq*2+ 0], m14
   mova        [qcoeffq+ncoeffq*2+16], m13
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
   pabsw                          m14, m14
   pabsw                          m13, m13
 %endif
   pmullw                         m14, m3                   ; dqc[i] = qc[i] * q
   pmullw                         m13, m3                   ; dqc[i] = qc[i] * q
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
   psrlw                          m14, 1
   psrlw                          m13, 1
   psignw                         m14, m9
   psignw                         m13, m10
 %endif
   mova       [dqcoeffq+ncoeffq*2+ 0], m14
   mova       [dqcoeffq+ncoeffq*2+16], m13
   pcmpeqw                        m14, m5                   ; m14 = c[i] == 0
   pcmpeqw                        m13, m5                   ; m13 = c[i] == 0
   mova                            m6, [  iscanq+ncoeffq*2+ 0] ; m6 = scan[i]
   mova                           m11, [  iscanq+ncoeffq*2+16] ; m11 = scan[i]
   psubw                           m6, m7                   ; m6 = scan[i] + 1
   psubw                          m11, m7                   ; m11 = scan[i] + 1
   pandn                          m14, m6                   ; m14 = max(eob)
   pandn                          m13, m11                  ; m13 = max(eob)
   pmaxsw                          m8, m14
   pmaxsw                          m8, m13
   add                        ncoeffq, mmsize
   jl .ac_only_loop
 
-%ifidn %1, b_32x32
+%ifidn %1, fp_32x32
   jmp .accumulate_eob
 .skip_iter:
   mova        [qcoeffq+ncoeffq*2+ 0], m5
   mova        [qcoeffq+ncoeffq*2+16], m5
   mova       [dqcoeffq+ncoeffq*2+ 0], m5
   mova       [dqcoeffq+ncoeffq*2+16], m5
   add                        ncoeffq, mmsize
   jl .ac_only_loop
 %endif
 
@@ skipping 27 matching lines @@
   mova        [qcoeffq+ncoeffq*2+ 0], m7
   mova        [qcoeffq+ncoeffq*2+16], m7
   add                        ncoeffq, mmsize
   jl .blank_loop
   mov                    word [eobq], 0
   RET
 %endmacro
 
 INIT_XMM ssse3
 QUANTIZE_FP fp, 7
+QUANTIZE_FP fp_32x32, 7