Convolve copy function for jnt_comp

Added a copy function (c version and sse2 version) for full-pixel motion
vectors for jnt_comp experiment following existing av1_convolve_2d_copy
function.

Change-Id: I20fd2219799f9c1451f591574fbe97364f40e0f0
diff --git a/av1/common/av1_rtcd_defs.pl b/av1/common/av1_rtcd_defs.pl
index cbf79b8..5ac9460 100755
--- a/av1/common/av1_rtcd_defs.pl
+++ b/av1/common/av1_rtcd_defs.pl
@@ -599,6 +599,11 @@
   if (aom_config("CONFIG_JNT_COMP") eq "yes") {
     add_proto qw/void av1_jnt_convolve_2d/, "const uint8_t *src, int src_stride, CONV_BUF_TYPE *dst, int dst_stride, int w, int h, InterpFilterParams *filter_params_x, InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
     specialize qw/av1_jnt_convolve_2d sse4_1/;
+
+    if (aom_config("CONFIG_COMPOUND_ROUND") ne "yes") {
+      add_proto qw/void av1_jnt_convolve_2d_copy/, "const uint8_t *src, int src_stride, CONV_BUF_TYPE *dst, int dst_stride, int w, int h, InterpFilterParams *filter_params_x, InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
+      specialize qw/av1_jnt_convolve_2d_copy sse4_1/;
+    }
   }
 
   if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
diff --git a/av1/common/convolve.c b/av1/common/convolve.c
index cb11305..13b92fcb 100644
--- a/av1/common/convolve.c
+++ b/av1/common/convolve.c
@@ -756,6 +756,42 @@
     }
   }
 }
+
+void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride,
+                                CONV_BUF_TYPE *dst, int dst_stride, int w,
+                                int h, InterpFilterParams *filter_params_x,
+                                InterpFilterParams *filter_params_y,
+                                const int subpel_x_q4, const int subpel_y_q4,
+                                ConvolveParams *conv_params) {
+  const int bits =
+      FILTER_BITS * 2 - conv_params->round_1 - conv_params->round_0;
+
+  (void)filter_params_x;
+  (void)filter_params_y;
+  (void)subpel_x_q4;
+  (void)subpel_y_q4;
+
+  for (int y = 0; y < h; ++y) {
+    for (int x = 0; x < w; ++x) {
+      CONV_BUF_TYPE res = (1 << bits) * src[y * src_stride + x];
+      if (conv_params->use_jnt_comp_avg) {
+        if (conv_params->do_average) {
+          dst[y * dst_stride + x] += res * conv_params->bck_offset;
+
+          dst[y * dst_stride + x] = ROUND_POWER_OF_TWO(dst[y * dst_stride + x],
+                                                       DIST_PRECISION_BITS - 1);
+        } else {
+          dst[y * dst_stride + x] = res * conv_params->fwd_offset;
+        }
+      } else {
+        if (conv_params->do_average)
+          dst[y * dst_stride + x] += res;
+        else
+          dst[y * dst_stride + x] = res;
+      }
+    }
+  }
+}
 #endif  // CONFIG_JNT_COMP
 
 void av1_convolve_2d_scale_c(const uint8_t *src, int src_stride,
@@ -903,17 +939,44 @@
                     tr_dst_stride, h, w);
   } else {
 #if CONFIG_JNT_COMP
-    if (scaled)
+    if (scaled) {
       av1_convolve_2d_scale(src, src_stride, conv_params->dst,
                             conv_params->dst_stride, w, h, &filter_params_x,
                             &filter_params_y, subpel_x_q4, x_step_q4,
                             subpel_y_q4, y_step_q4, conv_params);
-    else
+    } else {
+#if CONFIG_COMPOUND_ROUND
       av1_jnt_convolve_2d(src, src_stride, conv_params->dst,
                           conv_params->dst_stride, w, h, &filter_params_x,
                           &filter_params_y, subpel_x_q4, subpel_y_q4,
                           conv_params);
 #else
+      if (subpel_x_q4 == 0 && subpel_y_q4 == 0) {
+        av1_jnt_convolve_2d_copy(src, src_stride, conv_params->dst,
+                                 conv_params->dst_stride, w, h,
+                                 &filter_params_x, &filter_params_y,
+                                 subpel_x_q4, subpel_y_q4, conv_params);
+      } else if (subpel_x_q4 == 0) {
+        // place holder
+        av1_jnt_convolve_2d(src, src_stride, conv_params->dst,
+                            conv_params->dst_stride, w, h, &filter_params_x,
+                            &filter_params_y, subpel_x_q4, subpel_y_q4,
+                            conv_params);
+      } else if (subpel_y_q4 == 0) {
+        // place holder
+        av1_jnt_convolve_2d(src, src_stride, conv_params->dst,
+                            conv_params->dst_stride, w, h, &filter_params_x,
+                            &filter_params_y, subpel_x_q4, subpel_y_q4,
+                            conv_params);
+      } else {
+        av1_jnt_convolve_2d(src, src_stride, conv_params->dst,
+                            conv_params->dst_stride, w, h, &filter_params_x,
+                            &filter_params_y, subpel_x_q4, subpel_y_q4,
+                            conv_params);
+      }
+#endif  // CONFIG_COMPOUND_ROUND
+    }
+#else
     if (scaled) {
       av1_convolve_2d_scale(src, src_stride, conv_params->dst,
                             conv_params->dst_stride, w, h, &filter_params_x,
diff --git a/av1/common/x86/convolve_2d_sse4.c b/av1/common/x86/convolve_2d_sse4.c
index 71c32e7..893036b 100644
--- a/av1/common/x86/convolve_2d_sse4.c
+++ b/av1/common/x86/convolve_2d_sse4.c
@@ -450,5 +450,194 @@
     }
   }
 }
+
+void av1_jnt_convolve_2d_copy_sse4_1(const uint8_t *src, int src_stride,
+                                     CONV_BUF_TYPE *dst, int dst_stride, int w,
+                                     int h, InterpFilterParams *filter_params_x,
+                                     InterpFilterParams *filter_params_y,
+                                     const int subpel_x_q4,
+                                     const int subpel_y_q4,
+                                     ConvolveParams *conv_params) {
+  const int bits =
+      FILTER_BITS * 2 - conv_params->round_1 - conv_params->round_0;
+  const int do_average = conv_params->do_average;
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i left_shift = _mm_cvtsi32_si128(bits);
+  int i, j;
+  (void)filter_params_x;
+  (void)filter_params_y;
+  (void)subpel_x_q4;
+  (void)subpel_y_q4;
+
+  const int w0 = conv_params->fwd_offset;
+  const int w1 = conv_params->bck_offset;
+  const __m128i wt0 = _mm_set1_epi32(w0);
+  const __m128i wt1 = _mm_set1_epi32(w1);
+  const int jnt_round_const = 1 << (DIST_PRECISION_BITS - 2);
+  const __m128i jnt_r = _mm_set1_epi32(jnt_round_const);
+
+  if (!(w % 16)) {
+    for (i = 0; i < h; ++i) {
+      for (j = 0; j < w; j += 16) {
+        const __m128i d8 = _mm_loadu_si128((__m128i *)&src[j]);
+        const __m128i d16_0 = _mm_unpacklo_epi8(d8, zero);
+        const __m128i d16_1 = _mm_unpackhi_epi8(d8, zero);
+        __m128i d32_0 = _mm_unpacklo_epi16(d16_0, zero);
+        __m128i d32_1 = _mm_unpackhi_epi16(d16_0, zero);
+        __m128i d32_2 = _mm_unpacklo_epi16(d16_1, zero);
+        __m128i d32_3 = _mm_unpackhi_epi16(d16_1, zero);
+
+        d32_0 = _mm_sll_epi32(d32_0, left_shift);
+        d32_1 = _mm_sll_epi32(d32_1, left_shift);
+        d32_2 = _mm_sll_epi32(d32_2, left_shift);
+        d32_3 = _mm_sll_epi32(d32_3, left_shift);
+
+        __m128i *const p = (__m128i *)&dst[j];
+
+        if (conv_params->use_jnt_comp_avg) {
+          if (do_average) {
+            __m128i weighted_res = _mm_mullo_epi32(d32_0, wt1);
+            __m128i sum = _mm_add_epi32(_mm_loadu_si128(p + 0), weighted_res);
+            d32_0 = _mm_srai_epi32(_mm_add_epi32(sum, jnt_r),
+                                   DIST_PRECISION_BITS - 1);
+
+            weighted_res = _mm_mullo_epi32(d32_1, wt1);
+            sum = _mm_add_epi32(_mm_loadu_si128(p + 1), weighted_res);
+            d32_1 = _mm_srai_epi32(_mm_add_epi32(sum, jnt_r),
+                                   DIST_PRECISION_BITS - 1);
+
+            weighted_res = _mm_mullo_epi32(d32_2, wt1);
+            sum = _mm_add_epi32(_mm_loadu_si128(p + 2), weighted_res);
+            d32_2 = _mm_srai_epi32(_mm_add_epi32(sum, jnt_r),
+                                   DIST_PRECISION_BITS - 1);
+
+            weighted_res = _mm_mullo_epi32(d32_3, wt1);
+            sum = _mm_add_epi32(_mm_loadu_si128(p + 3), weighted_res);
+            d32_3 = _mm_srai_epi32(_mm_add_epi32(sum, jnt_r),
+                                   DIST_PRECISION_BITS - 1);
+          } else {
+            d32_0 = _mm_mullo_epi32(d32_0, wt0);
+            d32_1 = _mm_mullo_epi32(d32_1, wt0);
+            d32_2 = _mm_mullo_epi32(d32_2, wt0);
+            d32_3 = _mm_mullo_epi32(d32_3, wt0);
+          }
+        } else {
+          if (do_average) {
+            d32_0 = _mm_add_epi32(_mm_loadu_si128(p + 0), d32_0);
+            d32_1 = _mm_add_epi32(_mm_loadu_si128(p + 1), d32_1);
+            d32_2 = _mm_add_epi32(_mm_loadu_si128(p + 2), d32_2);
+            d32_3 = _mm_add_epi32(_mm_loadu_si128(p + 3), d32_3);
+          }
+        }
+
+        _mm_storeu_si128(p + 0, d32_0);
+        _mm_storeu_si128(p + 1, d32_1);
+        _mm_storeu_si128(p + 2, d32_2);
+        _mm_storeu_si128(p + 3, d32_3);
+      }
+      src += src_stride;
+      dst += dst_stride;
+    }
+  } else if (!(w % 8)) {
+    for (i = 0; i < h; ++i) {
+      for (j = 0; j < w; j += 8) {
+        const __m128i d8 = _mm_loadl_epi64((__m128i *)&src[j]);
+        const __m128i d16_0 = _mm_unpacklo_epi8(d8, zero);
+        __m128i d32_0 = _mm_unpacklo_epi16(d16_0, zero);
+        __m128i d32_1 = _mm_unpackhi_epi16(d16_0, zero);
+
+        d32_0 = _mm_sll_epi32(d32_0, left_shift);
+        d32_1 = _mm_sll_epi32(d32_1, left_shift);
+
+        __m128i *const p = (__m128i *)&dst[j];
+        if (conv_params->use_jnt_comp_avg) {
+          if (do_average) {
+            __m128i weighted_res = _mm_mullo_epi32(d32_0, wt1);
+            __m128i sum = _mm_add_epi32(_mm_loadu_si128(p + 0), weighted_res);
+            d32_0 = _mm_srai_epi32(_mm_add_epi32(sum, jnt_r),
+                                   DIST_PRECISION_BITS - 1);
+
+            weighted_res = _mm_mullo_epi32(d32_1, wt1);
+            sum = _mm_add_epi32(_mm_loadu_si128(p + 1), weighted_res);
+            d32_1 = _mm_srai_epi32(_mm_add_epi32(sum, jnt_r),
+                                   DIST_PRECISION_BITS - 1);
+          } else {
+            d32_0 = _mm_mullo_epi32(d32_0, wt0);
+            d32_1 = _mm_mullo_epi32(d32_1, wt0);
+          }
+        } else {
+          if (do_average) {
+            d32_0 = _mm_add_epi32(_mm_loadu_si128(p + 0), d32_0);
+            d32_1 = _mm_add_epi32(_mm_loadu_si128(p + 1), d32_1);
+          }
+        }
+
+        _mm_storeu_si128(p + 0, d32_0);
+        _mm_storeu_si128(p + 1, d32_1);
+      }
+      src += src_stride;
+      dst += dst_stride;
+    }
+  } else if (!(w % 4)) {
+    for (i = 0; i < h; ++i) {
+      for (j = 0; j < w; j += 4) {
+        const __m128i d8 = _mm_loadl_epi64((__m128i *)&src[j]);
+        const __m128i d16_0 = _mm_unpacklo_epi8(d8, zero);
+        __m128i d32_0 = _mm_unpacklo_epi16(d16_0, zero);
+
+        d32_0 = _mm_sll_epi32(d32_0, left_shift);
+
+        __m128i *const p = (__m128i *)&dst[j];
+        if (conv_params->use_jnt_comp_avg) {
+          if (do_average) {
+            __m128i weighted_res = _mm_mullo_epi32(d32_0, wt1);
+            __m128i sum = _mm_add_epi32(_mm_loadu_si128(p + 0), weighted_res);
+            d32_0 = _mm_srai_epi32(_mm_add_epi32(sum, jnt_r),
+                                   DIST_PRECISION_BITS - 1);
+          } else {
+            d32_0 = _mm_mullo_epi32(d32_0, wt0);
+          }
+        } else {
+          if (do_average) {
+            d32_0 = _mm_add_epi32(_mm_loadu_si128(p + 0), d32_0);
+          }
+        }
+
+        _mm_storeu_si128(p, d32_0);
+      }
+      src += src_stride;
+      dst += dst_stride;
+    }
+  } else {
+    for (i = 0; i < h; ++i) {
+      for (j = 0; j < w; j += 2) {
+        const __m128i d8 = _mm_cvtsi32_si128(*(const int *)&src[j]);
+        const __m128i d16_0 = _mm_unpacklo_epi8(d8, zero);
+        __m128i d32_0 = _mm_unpacklo_epi16(d16_0, zero);
+
+        d32_0 = _mm_sll_epi32(d32_0, left_shift);
+        __m128i *const p = (__m128i *)&dst[j];
+        if (conv_params->use_jnt_comp_avg) {
+          if (do_average) {
+            __m128i weighted_res = _mm_mullo_epi32(d32_0, wt1);
+            __m128i sum = _mm_add_epi32(_mm_loadl_epi64(p), weighted_res);
+            d32_0 = _mm_srai_epi32(_mm_add_epi32(sum, jnt_r),
+                                   DIST_PRECISION_BITS - 1);
+          } else {
+            d32_0 = _mm_mullo_epi32(d32_0, wt0);
+          }
+        } else {
+          if (do_average) {
+            d32_0 = _mm_add_epi32(_mm_loadl_epi64(p), d32_0);
+          }
+        }
+
+        _mm_storel_epi64(p, d32_0);
+      }
+      src += src_stride;
+      dst += dst_stride;
+    }
+  }
+}
 #endif  // CONFIG_COMPOUND_ROUND
 #endif  // CONFIG_JNT_COMP