<div dir="ltr">What is the improvement in cycles that we see from testbench from this patch?</div><div class="gmail_extra"><br clear="all"><div><div class="gmail_signature"><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div dir="ltr"><div dir="ltr"><div dir="ltr"><div dir="ltr">Pradeep Ramachandran, PhD<div>Solution Architect at <a href="http://www.multicorewareinc.com/" style="font-size:12.8px" target="_blank">www.multicorewareinc.com/</a></div><div>Visiting Professor at<a href="http://www.cse.iitm.ac.in" target="_blank"> www.cse.iitm.ac.in</a>/</div><div><a href="http://pradeeprama.info/" style="font-size:12.8px" target="_blank">pradeeprama.info/</a><br></div><div><span style="font-size:12.8px">Ph:   +91 99627 82018</span><br></div></div></div></div></div></div></div></div></div></div></div></div>
<br><div class="gmail_quote">On Tue, Mar 22, 2016 at 6:57 PM,  <span dir="ltr"><<a href="mailto:ramya@multicorewareinc.com" target="_blank">ramya@multicorewareinc.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"># HG changeset patch<br>
# User Ramya Sriraman<<a href="mailto:ramya@multicorewareinc.com">ramya@multicorewareinc.com</a>><br>
# Date 1458652316 -19800<br>
#      Tue Mar 22 18:41:56 2016 +0530<br>
# Node ID fd95ed60b242adffbeb0991609271c8a15040ff9<br>
# Parent  a9014e51d47ee5cdfe381d02526b1c94082cd4bf<br>
arm: Implement interp_8tap_vert_ps_NxN NEON<br>
<br>
diff -r a9014e51d47e -r fd95ed60b242 source/common/arm/asm-primitives.cpp<br>
--- a/source/common/arm/asm-primitives.cpp      Tue Mar 22 11:10:43 2016 +0530<br>
+++ b/source/common/arm/asm-primitives.cpp      Tue Mar 22 18:41:56 2016 +0530<br>
@@ -354,6 +354,32 @@<br>
         p.pu[LUMA_24x32].luma_vsp   = PFX(interp_8tap_vert_sp_24x32_neon);<br>
         p.pu[LUMA_48x64].luma_vsp   = PFX(interp_8tap_vert_sp_48x64_neon);<br>
         p.pu[LUMA_12x16].luma_vsp   = PFX(interp_8tap_vert_sp_12x16_neon);<br>
+<br>
+        p.pu[LUMA_4x4].luma_vps     = PFX(interp_8tap_vert_ps_4x4_neon);<br>
+        p.pu[LUMA_4x8].luma_vps     = PFX(interp_8tap_vert_ps_4x8_neon);<br>
+        p.pu[LUMA_4x16].luma_vps    = PFX(interp_8tap_vert_ps_4x16_neon);<br>
+        p.pu[LUMA_8x4].luma_vps     = PFX(interp_8tap_vert_ps_8x4_neon);<br>
+        p.pu[LUMA_8x8].luma_vps     = PFX(interp_8tap_vert_ps_8x8_neon);<br>
+        p.pu[LUMA_8x16].luma_vps    = PFX(interp_8tap_vert_ps_8x16_neon);<br>
+        p.pu[LUMA_8x32].luma_vps    = PFX(interp_8tap_vert_ps_8x32_neon);<br>
+        p.pu[LUMA_16x4].luma_vps    = PFX(interp_8tap_vert_ps_16x4_neon);<br>
+        p.pu[LUMA_16x8].luma_vps    = PFX(interp_8tap_vert_ps_16x8_neon);<br>
+        p.pu[LUMA_16x16].luma_vps   = PFX(interp_8tap_vert_ps_16x16_neon);<br>
+        p.pu[LUMA_16x32].luma_vps   = PFX(interp_8tap_vert_ps_16x32_neon);<br>
+        p.pu[LUMA_16x64].luma_vps   = PFX(interp_8tap_vert_ps_16x64_neon);<br>
+        p.pu[LUMA_16x12].luma_vps   = PFX(interp_8tap_vert_ps_16x12_neon);<br>
+        p.pu[LUMA_32x8].luma_vps    = PFX(interp_8tap_vert_ps_32x8_neon);<br>
+        p.pu[LUMA_32x16].luma_vps   = PFX(interp_8tap_vert_ps_32x16_neon);<br>
+        p.pu[LUMA_32x32].luma_vps   = PFX(interp_8tap_vert_ps_32x32_neon);<br>
+        p.pu[LUMA_32x64].luma_vps   = PFX(interp_8tap_vert_ps_32x64_neon);<br>
+        p.pu[LUMA_32x24].luma_vps   = PFX(interp_8tap_vert_ps_32x24_neon);<br>
+        p.pu[LUMA_64x16].luma_vps   = PFX(interp_8tap_vert_ps_64x16_neon);<br>
+        p.pu[LUMA_64x32].luma_vps   = PFX(interp_8tap_vert_ps_64x32_neon);<br>
+        p.pu[LUMA_64x64].luma_vps   = PFX(interp_8tap_vert_ps_64x64_neon);<br>
+        p.pu[LUMA_64x48].luma_vps   = PFX(interp_8tap_vert_ps_64x48_neon);<br>
+        p.pu[LUMA_24x32].luma_vps   = PFX(interp_8tap_vert_ps_24x32_neon);<br>
+        p.pu[LUMA_48x64].luma_vps   = PFX(interp_8tap_vert_ps_48x64_neon);<br>
+        p.pu[LUMA_12x16].luma_vps   = PFX(interp_8tap_vert_ps_12x16_neon);<br>
     }<br>
     if (cpuMask & X265_CPU_ARMV6)<br>
     {<br>
diff -r a9014e51d47e -r fd95ed60b242 source/common/arm/ipfilter8.S<br>
--- a/source/common/arm/ipfilter8.S     Tue Mar 22 11:10:43 2016 +0530<br>
+++ b/source/common/arm/ipfilter8.S     Tue Mar 22 18:41:56 2016 +0530<br>
@@ -698,7 +698,7 @@<br>
     bgt         .loop_filterP2S_48x64<br>
     bx          lr<br>
 endfunc<br>
-<br>
+//**************luma_vpp************<br>
 .macro LUMA_VPP_4xN h<br>
 function x265_interp_8tap_vert_pp_4x\h\()_neon<br>
     push           {r4, r5, r6}<br>
@@ -1606,4 +1606,333 @@<br>
     pop             {r4, r5, r6, r7}<br>
     bx              lr<br>
 endfunc<br>
+//**************luma_vps*****************<br>
+.macro LUMA_VPS_4xN h<br>
+function x265_interp_8tap_vert_ps_4x\h\()_neon<br>
+    push           {r4, r5, r6}<br>
+    ldr             r4, [sp, #4 * 3]<br>
+    lsl             r3, #1<br>
+    mov             r5, r4, lsl #6<br>
+    mov             r4, r1, lsl #2<br>
+    sub             r4, r1<br>
+    sub             r0, r4<br>
<br>
+    mov             r4, #8192<br>
+    vdup.32         q8, r4<br>
+    mov             r4, #\h<br>
+<br>
+.loop_vps_4x\h:<br>
+    movrel          r12, g_lumaFilter<br>
+    add             r12, r5<br>
+    mov             r6, r0<br>
+<br>
+    pld [r6]<br>
+    vld1.u32        d0[0], [r6], r1<br>
+    pld [r6]<br>
+    vld1.u32        d0[1], [r6], r1<br>
+    pld [r6]<br>
+    vld1.u32        d1[0], [r6], r1<br>
+    pld [r6]<br>
+    vld1.u32        d1[1], [r6], r1<br>
+    pld [r6]<br>
+    vld1.u32        d2[0], [r6], r1<br>
+    pld [r6]<br>
+    vld1.u32        d2[1], [r6], r1<br>
+    pld [r6]<br>
+    vld1.u32        d3[0], [r6], r1<br>
+    pld [r6]<br>
+    vld1.u32        d3[1], [r6], r1<br>
+<br>
+    veor.u8         q9, q9<br>
+<br>
+    vmovl.u8        q11, d0<br>
+    vmovl.u16       q12, d22<br>
+    vmovl.u16       q13, d23<br>
+    vld1.s32        d20, [r12]!<br>
+    vmov.s32        d21, d20<br>
+    vmla.s32        q9, q12, q10<br>
+    vld1.s32        d20, [r12]!<br>
+    vmov.s32        d21, d20<br>
+    vmla.s32        q9, q13, q10<br>
+<br>
+    vmovl.u8        q11, d1<br>
+    vmovl.u16       q12, d22<br>
+    vmovl.u16       q13, d23<br>
+    vld1.s32        d20, [r12]!<br>
+    vmov.s32        d21, d20<br>
+    vmla.s32        q9, q12, q10<br>
+    vld1.s32        d20, [r12]!<br>
+    vmov.s32        d21, d20<br>
+    vmla.s32        q9, q13, q10<br>
+<br>
+    vmovl.u8        q11, d2<br>
+    vmovl.u16       q12, d22<br>
+    vmovl.u16       q13, d23<br>
+    vld1.s32        d20, [r12]!<br>
+    vmov.s32        d21, d20<br>
+    vmla.s32        q9, q12, q10<br>
+    vld1.s32        d20, [r12]!<br>
+    vmov.s32        d21, d20<br>
+    vmla.s32        q9, q13, q10<br>
+<br>
+    vmovl.u8        q11, d3<br>
+    vmovl.u16       q12, d22<br>
+    vmovl.u16       q13, d23<br>
+    vld1.s32        d20, [r12]!<br>
+    vmov.s32        d21, d20<br>
+    vmla.s32        q9, q12, q10<br>
+    vld1.s32        d20, [r12]!<br>
+    vmov.s32        d21, d20<br>
+    vmla.s32        q9, q13, q10<br>
+<br>
+    vsub.s32        q9, q8<br>
+    vqmovn.s32      d0, q9<br>
+    vst1.u16        d0, [r2], r3<br>
+<br>
+    add             r0, r1<br>
+    subs            r4, #1<br>
+    bne             .loop_vps_4x\h<br>
+<br>
+    pop             {r4, r5, r6}<br>
+    bx              lr<br>
+    .ltorg<br>
+endfunc<br>
+.endm<br>
+<br>
+LUMA_VPS_4xN 4<br>
+LUMA_VPS_4xN 8<br>
+LUMA_VPS_4xN 16<br>
+<br>
+<br>
+.macro FILTER_VPS a b filterv<br>
+<br>
+.loop_ps_\filterv\()_\a\()x\b:<br>
+<br>
+    mov             r7, r2<br>
+    mov             r6, r0<br>
+    eor             r8, r8<br>
+<br>
+.loop_ps_w8_\filterv\()_\a\()x\b:<br>
+<br>
+    add             r6, r0, r8<br>
+<br>
+    pld [r6]<br>
+    vld1.u8         d0, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d1, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d2, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d3, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d4, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d5, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d6, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d7, [r6], r1<br>
+<br>
+    veor.u8         q9, q9<br>
+    veor.u8         q10, q10<br>
+<br>
+   \filterv<br>
+<br>
+    mov             r12,#8192<br>
+    vdup.32         q8, r12<br>
+    vsub.s32        q9, q8<br>
+    vqmovn.s32      d0, q9<br>
+    vsub.s32        q10, q8<br>
+    vqmovn.s32      d1, q10<br>
+    vst1.u16         {q0}, [r7]!<br>
+<br>
+    add             r8, #8<br>
+    cmp             r8, #\a<br>
+    blt             .loop_ps_w8_\filterv\()_\a\()x\b<br>
+<br>
+    add             r0, r1<br>
+    add             r2, r3<br>
+    subs            r4, #1<br>
+    bne             .loop_ps_\filterv\()_\a\()x\b<br>
+<br>
+.endm<br>
+<br>
+.macro LUMA_VPS  w h<br>
+function x265_interp_8tap_vert_ps_\w\()x\h\()_neon<br>
+<br>
+    push            {r4, r5, r6, r7, r8}<br>
+    ldr             r5, [sp, #4 * 5]<br>
+    lsl             r3, #1<br>
+    mov             r4, r1, lsl #2<br>
+    sub             r4, r1<br>
+    sub             r0, r4<br>
+    mov             r4, #\h<br>
+<br>
+    cmp             r5, #0<br>
+    beq              0f<br>
+    cmp             r5, #1<br>
+    beq              1f<br>
+    cmp             r5, #2<br>
+    beq              2f<br>
+    cmp             r5, #3<br>
+    beq              3f<br>
+0:<br>
+    FILTER_VPS  \w \h qpel_filter_0_32b<br>
+    b            5f<br>
+1:<br>
+    FILTER_VPS  \w \h qpel_filter_1_32b<br>
+    b            5f<br>
+2:<br>
+    FILTER_VPS  \w \h qpel_filter_2_32b<br>
+    b            5f<br>
+3:<br>
+    FILTER_VPS  \w \h qpel_filter_3_32b<br>
+    b            5f<br>
+5:<br>
+    pop             {r4, r5, r6, r7, r8}<br>
+    bx              lr<br>
+endfunc<br>
+.endm<br>
+<br>
+LUMA_VPS 8 4<br>
+LUMA_VPS 8 8<br>
+LUMA_VPS 8 16<br>
+LUMA_VPS 8 32<br>
+LUMA_VPS 16 4<br>
+LUMA_VPS 16 8<br>
+LUMA_VPS 16 16<br>
+LUMA_VPS 16 32<br>
+LUMA_VPS 16 64<br>
+LUMA_VPS 16 12<br>
+LUMA_VPS 32 8<br>
+LUMA_VPS 32 16<br>
+LUMA_VPS 32 32<br>
+LUMA_VPS 32 64<br>
+LUMA_VPS 32 24<br>
+LUMA_VPS 64 16<br>
+LUMA_VPS 64 32<br>
+LUMA_VPS 64 64<br>
+LUMA_VPS 64 48<br>
+LUMA_VPS 24 32<br>
+LUMA_VPS 48 64<br>
+<br>
+function x265_interp_8tap_vert_ps_12x16_neon<br>
+    push            {r4, r5, r6, r7}<br>
+    lsl             r3, #1<br>
+    ldr             r5, [sp, #4 * 4]<br>
+    mov             r4, r1, lsl #2<br>
+    sub             r4, r1<br>
+    sub             r0, r4<br>
+<br>
+    mov             r4, #16<br>
+.loop_vps_12x16:<br>
+<br>
+    mov             r6, r0<br>
+    mov             r7, r2<br>
+<br>
+    pld [r6]<br>
+    vld1.u8         d0, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d1, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d2, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d3, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d4, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d5, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d6, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d7, [r6], r1<br>
+<br>
+    veor.u8         q9, q9<br>
+    veor.u8         q10, q10<br>
+<br>
+    cmp             r5,#0<br>
+    beq              0f<br>
+    cmp             r5,#1<br>
+    beq              1f<br>
+    cmp             r5,#2<br>
+    beq              2f<br>
+    cmp             r5,#3<br>
+    beq              3f<br>
+0:<br>
+    qpel_filter_0_32b<br>
+    b            5f<br>
+1:<br>
+    qpel_filter_1_32b<br>
+    b            5f<br>
+2:<br>
+    qpel_filter_2_32b<br>
+    b            5f<br>
+3:<br>
+    qpel_filter_3_32b<br>
+    b            5f<br>
+5:<br>
+    mov             r12,#8192<br>
+    vdup.32         q8, r12<br>
+    vsub.s32        q9, q8<br>
+    vqmovn.s32      d0, q9<br>
+    vsub.s32        q10, q8<br>
+    vqmovn.s32      d1, q10<br>
+    vst1.u8         {q0}, [r7]!<br>
+<br>
+    add             r6, r0, #8<br>
+<br>
+    pld [r6]<br>
+    vld1.u8         d0, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d1, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d2, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d3, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d4, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d5, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d6, [r6], r1<br>
+    pld [r6]<br>
+    vld1.u8         d7, [r6], r1<br>
+<br>
+    veor.u8         q9, q9<br>
+    veor.u8         q10, q10<br>
+<br>
+    cmp             r5,#0<br>
+    beq              0f<br>
+    cmp             r5,#1<br>
+    beq              1f<br>
+    cmp             r5,#2<br>
+    beq              2f<br>
+    cmp             r5,#3<br>
+    beq              3f<br>
+0:<br>
+    qpel_filter_0_32b<br>
+    b            5f<br>
+1:<br>
+    qpel_filter_1_32b<br>
+    b            5f<br>
+2:<br>
+    qpel_filter_2_32b<br>
+    b            5f<br>
+3:<br>
+    qpel_filter_3_32b<br>
+    b            5f<br>
+5:<br>
+    mov             r12,#8192<br>
+    vdup.32         q8, r12<br>
+    vsub.s32        q9, q8<br>
+    vqmovn.s32      d0, q9<br>
+    vst1.u8         d0, [r7]!<br>
+<br>
+    add             r0, r1<br>
+    add             r2, r3<br>
+    subs            r4, #1<br>
+    bne             .loop_vps_12x16<br>
+<br>
+    pop             {r4, r5, r6, r7}<br>
+    bx              lr<br>
+endfunc<br>
diff -r a9014e51d47e -r fd95ed60b242 source/common/arm/ipfilter8.h<br>
--- a/source/common/arm/ipfilter8.h     Tue Mar 22 11:10:43 2016 +0530<br>
+++ b/source/common/arm/ipfilter8.h     Tue Mar 22 18:41:56 2016 +0530<br>
@@ -102,4 +102,30 @@<br>
 void x265_interp_8tap_vert_sp_24x32_neon(const int16_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx);<br>
 void x265_interp_8tap_vert_sp_48x64_neon(const int16_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx);<br>
 void x265_interp_8tap_vert_sp_12x16_neon(const int16_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx);<br>
+<br>
+void x265_interp_8tap_vert_ps_4x4_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_4x8_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_4x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_8x4_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_8x8_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_8x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_8x32_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_16x4_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_16x8_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_16x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_16x32_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_16x64_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_16x12_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_32x8_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_32x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_32x32_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_32x64_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_32x24_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_64x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_64x32_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_64x64_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_64x48_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_24x32_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_48x64_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
+void x265_interp_8tap_vert_ps_12x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx);<br>
 #endif // ifndef X265_IPFILTER8_ARM_H<br>
_______________________________________________<br>
x265-devel mailing list<br>
<a href="mailto:x265-devel@videolan.org">x265-devel@videolan.org</a><br>
<a href="https://mailman.videolan.org/listinfo/x265-devel" rel="noreferrer" target="_blank">https://mailman.videolan.org/listinfo/x265-devel</a><br>
</blockquote></div><br></div>