evas: add SSE3 op_blend_pixel_color blend functions
authorlucas <lucas@7cbeb6ba-43b4-40fd-8cce-4c39aea84d33>
Fri, 30 Sep 2011 02:36:51 +0000 (02:36 +0000)
committerlucas <lucas@7cbeb6ba-43b4-40fd-8cce-4c39aea84d33>
Fri, 30 Sep 2011 02:36:51 +0000 (02:36 +0000)
_op_blend_p_c_dp_sse3
_op_blend_pan_c_dp_sse3
_op_blend_p_can_dp_sse3
_op_blend_pan_can_dp_sse3
_op_blend_p_caa_dp_sse3
_op_blend_pan_caa_dp_sse3
_op_blend_rel_p_c_dp_sse3

Patch by: Jim Kukunas <james.t.kukunas@linux.intel.com>

git-svn-id: svn+ssh://svn.enlightenment.org/var/svn/e/trunk/evas@63702 7cbeb6ba-43b4-40fd-8cce-4c39aea84d33

src/lib/engines/common/evas_op_blend/op_blend_pixel_color_sse3.c [new file with mode: 0644]

diff --git a/src/lib/engines/common/evas_op_blend/op_blend_pixel_color_sse3.c b/src/lib/engines/common/evas_op_blend/op_blend_pixel_color_sse3.c
new file mode 100644 (file)
index 0000000..4ee31f5
--- /dev/null
@@ -0,0 +1,543 @@
+/* blend pixel x color --> dst */
+
+#ifdef BUILD_SSE3
+
+static void
+_op_blend_p_c_dp_sse3(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c, DATA32 *d, int l) {
+
+   DATA32 alpha;
+
+   const __m128i c_packed = _mm_set_epi32(c, c, c, c);
+
+   LOOP_ALIGNED_U1_A48_SSE3(d, l,
+      { /* UOP */
+
+         DATA32 sc = MUL4_SYM(c, *s);
+         alpha = 256 - (sc >> 24);
+         *d = sc + MUL_256(alpha, *d);
+         d++; s++; l--;
+      },
+      { /* A4OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i sc0 = mul4_sym_sse3(c_packed, s0);
+         __m128i a0  = sub4_alpha_sse3(sc0);
+         __m128i mul0 = mul_256_sse3(a0, d0);
+
+         d0 = _mm_add_epi32(sc0, mul0);
+
+         _mm_store_si128((__m128i *)d, d0);
+
+         d += 4; s += 4; l -= 4;
+      },
+      { /* A8OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i s1 = _mm_lddqu_si128((__m128i *)(s+4));
+         __m128i d1 = _mm_load_si128((__m128i *)(d+4));
+
+         __m128i sc0 = mul4_sym_sse3(c_packed, s0);
+         __m128i sc1 = mul4_sym_sse3(c_packed, s1);
+
+         __m128i a0 = sub4_alpha_sse3(sc0);
+         __m128i a1 = sub4_alpha_sse3(sc1);
+
+         __m128i mul0 = mul_256_sse3(a0, d0);
+         __m128i mul1 = mul_256_sse3(a1, d1);
+
+         d0 = _mm_add_epi32(sc0, mul0);
+         d1 = _mm_add_epi32(sc1, mul1);
+
+         _mm_store_si128((__m128i *)d, d0);
+         _mm_store_si128((__m128i *)(d+4), d1);
+
+         d += 8; s += 8; l -= 8;
+      })
+}
+
+static void
+_op_blend_pan_c_dp_sse3(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c, DATA32 *d, int l) {
+
+   DATA32 c_a = c & 0xFF000000;
+   DATA32 alpha = 256 - (c >> 24);
+
+   const __m128i c_packed = _mm_set_epi32(c, c, c, c);
+   const __m128i c_alpha = _mm_set_epi32(c_a, c_a, c_a, c_a);
+   const __m128i a0 = _mm_set_epi32(alpha, alpha, alpha, alpha);
+
+   LOOP_ALIGNED_U1_A48_SSE3(d, l,
+      { /* UOP */
+
+         *d = ((c & 0xff000000) + MUL3_SYM(c, *s)) + MUL_256(alpha, *d);
+         d++; s++; l--;
+      },
+      { /* A4OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i r0 = _mm_add_epi32(mul3_sym_sse3(c_packed, s0),
+            mul_256_sse3(a0, d0));
+
+         r0 = _mm_add_epi32(r0, c_alpha);
+
+         _mm_store_si128((__m128i *)d, r0);
+
+         d += 4; s += 4; l -= 4;
+      },
+      { /* A8OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i s1 = _mm_lddqu_si128((__m128i *)(s+4));
+         __m128i d1 = _mm_load_si128((__m128i *)(d+4));
+
+         __m128i r0 = _mm_add_epi32(mul3_sym_sse3(c_packed, s0),
+            mul_256_sse3(a0, d0));
+
+         __m128i r1 = _mm_add_epi32(mul3_sym_sse3(c_packed, s1),
+            mul_256_sse3(a0, d1));
+
+         r0 = _mm_add_epi32(r0, c_alpha);
+         r1 = _mm_add_epi32(r1, c_alpha);
+
+         _mm_store_si128((__m128i *)d, r0);
+         _mm_store_si128((__m128i *)(d+4), r1);
+
+         d += 8; s += 8; l -= 8;
+      })
+}
+
+static void
+_op_blend_p_can_dp_sse3(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c, DATA32 *d, int l) {
+
+   int alpha;
+   const __m128i c_packed = _mm_set_epi32(c, c, c, c);
+
+   LOOP_ALIGNED_U1_A48_SSE3(d, l,
+      { /* UOP */
+
+         alpha = 256 - (*s >> 24);
+         *d = ((*s & 0xff000000) + MUL3_SYM(c, *s)) + MUL_256(alpha, *d);
+         d++; s++; l--;
+      },
+      { /* A4OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i a0 = sub4_alpha_sse3(s0);
+
+         __m128i r0 = _mm_add_epi32(mul3_sym_sse3(c_packed, s0),
+            mul_256_sse3(a0, d0));
+
+         r0 = _mm_add_epi32(r0, _mm_and_si128(s0, A_MASK_SSE3));
+
+         _mm_store_si128((__m128i *)d, r0);
+
+         d += 4; s += 4; l -= 4;
+      },
+      {
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i s1 = _mm_lddqu_si128((__m128i *)(s+4));
+         __m128i d1 = _mm_load_si128((__m128i *)(d+4));
+
+         __m128i a0 = sub4_alpha_sse3(s0);
+         __m128i a1 = sub4_alpha_sse3(s1);
+
+         __m128i r0 = _mm_add_epi32(mul3_sym_sse3(c_packed, s0),
+            mul_256_sse3(a0, d0));
+
+         __m128i r1 = _mm_add_epi32(mul3_sym_sse3(c_packed, s1),
+            mul_256_sse3(a1, d1));
+
+         r0 = _mm_add_epi32(r0, _mm_and_si128(s0, A_MASK_SSE3));
+         r1 = _mm_add_epi32(r1, _mm_and_si128(s1, A_MASK_SSE3));
+
+         _mm_store_si128((__m128i *)d, r0);
+         _mm_store_si128((__m128i *)(d+4), r1);
+
+         d += 8; s += 8; l -= 8;
+      })
+}
+
+static void
+_op_blend_pan_can_dp_sse3(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c, DATA32 *d, int l) {
+
+   const __m128i c_packed = _mm_set_epi32(c, c, c, c);
+
+   LOOP_ALIGNED_U1_A48_SSE3(d, l,
+      { /* UOP */
+
+         *d++ = 0xff000000 + MUL3_SYM(c, *s);
+         s++; l--;
+      },
+      { /* A4OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+
+         __m128i r0 = mul3_sym_sse3(c_packed, s0);
+         r0 = _mm_add_epi32(r0, A_MASK_SSE3);
+
+         _mm_store_si128((__m128i *)d, r0);
+
+         d += 4; s += 4; l -= 4;
+      },
+      { /* A8OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i s1 = _mm_lddqu_si128((__m128i *)(s+4));
+
+         __m128i r0 = mul3_sym_sse3(c_packed, s0);
+         __m128i r1 = mul3_sym_sse3(c_packed, s1);
+
+         r0 = _mm_add_epi32(r0, A_MASK_SSE3);
+         r1 = _mm_add_epi32(r1, A_MASK_SSE3);
+
+         _mm_store_si128((__m128i *)d, r0);
+         _mm_store_si128((__m128i *)(d+4), r1);
+
+         d += 8; s += 8; l -= 8;
+      })
+}
+
+static void
+_op_blend_p_caa_dp_sse3(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c, DATA32 *d, int l) {
+
+   int alpha;
+   c = 1 + (c & 0xff);
+   const __m128i c_packed = _mm_set_epi32(c, c, c, c);
+
+   LOOP_ALIGNED_U1_A48_SSE3(d, l,
+      { /* UOP */
+
+         DATA32 sc = MUL_256(c, *s);
+         alpha = 256 - (sc >> 24);
+         *d = sc + MUL_256(alpha, *d);
+         d++;
+         s++;
+         l--;
+      },
+      { /* A4OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128 ((__m128i *)d);
+
+         __m128i sc0 = mul_256_sse3(c_packed, s0);
+         __m128i a0 = sub4_alpha_sse3(sc0);
+
+         __m128i r0 = _mm_add_epi32(mul_256_sse3(a0, d0), sc0);
+
+         _mm_store_si128((__m128i *)d, r0);
+
+         d += 4; s += 4; l -= 4;
+      },
+      {
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i s1 = _mm_lddqu_si128((__m128i *)(s+4));
+         __m128i d1 = _mm_load_si128((__m128i *)(d+4));
+
+         __m128i sc0 = mul_256_sse3(c_packed, s0);
+         __m128i sc1 = mul_256_sse3(c_packed, s1);
+
+         __m128i a0 = sub4_alpha_sse3(sc0);
+         __m128i a1 = sub4_alpha_sse3(sc1);
+
+         __m128i r0 = _mm_add_epi32(mul_256_sse3(a0, d0), sc0);
+         __m128i r1 = _mm_add_epi32(mul_256_sse3(a1, d1), sc1);
+
+         _mm_store_si128((__m128i *)d, r0);
+         _mm_store_si128((__m128i *)(d+4), r1);
+
+         d += 8; s += 8; l -= 8;
+      })
+}
+
+static void
+_op_blend_pan_caa_dp_sse3(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c, DATA32 *d, int l) {
+
+   c = 1 + (c & 0xff);
+   const __m128i c_packed = _mm_set_epi32(c, c, c,c);
+
+   LOOP_ALIGNED_U1_A48_SSE3(d, l,
+      { /* UOP */
+
+         *d = INTERP_256(c, *s, *d);
+         d++; s++; l--;
+      },
+      { /* A4OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i r0 = interp4_256_sse3(c_packed, s0, d0);
+
+         _mm_store_si128((__m128i *)d, r0);
+
+         d += 4; s += 4; l -= 4;
+      },
+      {
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i s1 = _mm_lddqu_si128((__m128i *)(s+4));
+         __m128i d1 = _mm_load_si128((__m128i *)(d+4));
+
+         __m128i r0 = interp4_256_sse3(c_packed, s0, d0);
+         __m128i r1 = interp4_256_sse3(c_packed, s1, d1);
+
+         _mm_store_si128((__m128i *)d, r0);
+         _mm_store_si128((__m128i *)(d+4), r1);
+
+         d += 8; s += 8; l -= 8;
+      })
+}
+
+#define _op_blend_pas_c_dp_sse3 _op_blend_p_c_dp_sse3
+#define _op_blend_pas_can_dp_sse3 _op_blend_p_can_dp_sse3
+#define _op_blend_pas_caa_dp_sse3 _op_blend_p_caa_dp_sse3
+
+#define _op_blend_p_c_dpan_sse3 _op_blend_p_c_dp_sse3
+#define _op_blend_pas_c_dpan_sse3 _op_blend_pas_c_dp_sse3
+#define _op_blend_pan_c_dpan_sse3 _op_blend_pan_c_dp_sse3
+#define _op_blend_p_can_dpan_sse3 _op_blend_p_can_dp_sse3
+#define _op_blend_pas_can_dpan_sse3 _op_blend_pas_can_dp_sse3
+#define _op_blend_pan_can_dpan_sse3 _op_blend_pan_can_dp_sse3
+#define _op_blend_p_caa_dpan_sse3 _op_blend_p_caa_dp_sse3
+#define _op_blend_pas_caa_dpan_sse3 _op_blend_pas_caa_dp_sse3
+#define _op_blend_pan_caa_dpan_sse3 _op_blend_pan_caa_dp_sse3
+
+static void
+init_blend_pixel_color_span_funcs_sse3(void)
+{
+   op_blend_span_funcs[SP][SM_N][SC][DP][CPU_SSE3] = _op_blend_p_c_dp_sse3;
+   op_blend_span_funcs[SP_AS][SM_N][SC][DP][CPU_SSE3] = _op_blend_pas_c_dp_sse3;
+   op_blend_span_funcs[SP_AN][SM_N][SC][DP][CPU_SSE3] = _op_blend_pan_c_dp_sse3;
+   op_blend_span_funcs[SP][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_p_can_dp_sse3;
+   op_blend_span_funcs[SP_AS][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_pas_can_dp_sse3;
+   op_blend_span_funcs[SP_AN][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_pan_can_dp_sse3;
+   op_blend_span_funcs[SP][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_p_caa_dp_sse3;
+   op_blend_span_funcs[SP_AS][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_pas_caa_dp_sse3;
+   op_blend_span_funcs[SP_AN][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_pan_caa_dp_sse3;
+
+   op_blend_span_funcs[SP][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_p_c_dpan_sse3;
+   op_blend_span_funcs[SP_AS][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_pas_c_dpan_sse3;
+   op_blend_span_funcs[SP_AN][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_pan_c_dpan_sse3;
+   op_blend_span_funcs[SP][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_p_can_dpan_sse3;
+   op_blend_span_funcs[SP_AS][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_pas_can_dpan_sse3;
+   op_blend_span_funcs[SP_AN][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_pan_can_dpan_sse3;
+   op_blend_span_funcs[SP][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_p_caa_dpan_sse3;
+   op_blend_span_funcs[SP_AS][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_pas_caa_dpan_sse3;
+   op_blend_span_funcs[SP_AN][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_pan_caa_dpan_sse3;
+}
+
+#define _op_blend_pt_p_c_dp_sse3 NULL
+
+#define _op_blend_pt_pas_c_dp_sse3 _op_blend_pt_p_c_dp_sse3
+#define _op_blend_pt_pan_c_dp_sse3 _op_blend_pt_p_c_dp_sse3
+#define _op_blend_pt_p_can_dp_sse3 _op_blend_pt_p_c_dp_sse3
+#define _op_blend_pt_pas_can_dp_sse3 _op_blend_pt_p_c_dp_sse3
+#define _op_blend_pt_pan_can_dp_sse3 _op_blend_pt_p_c_dp_sse3
+#define _op_blend_pt_p_caa_dp_sse3 _op_blend_pt_p_c_dp_sse3
+#define _op_blend_pt_pas_caa_dp_sse3 _op_blend_pt_p_c_dp_sse3
+#define _op_blend_pt_pan_caa_dp_sse3 _op_blend_pt_p_c_dp_sse3
+
+#define _op_blend_pt_p_c_dpan_sse3 _op_blend_pt_p_c_dp_sse3
+#define _op_blend_pt_pas_c_dpan_sse3 _op_blend_pt_pas_c_dp_sse3
+#define _op_blend_pt_pan_c_dpan_sse3 _op_blend_pt_pan_c_dp_sse3
+#define _op_blend_pt_p_can_dpan_sse3 _op_blend_pt_p_can_dp_sse3
+#define _op_blend_pt_pas_can_dpan_sse3 _op_blend_pt_pas_can_dp_sse3
+#define _op_blend_pt_pan_can_dpan_sse3 _op_blend_pt_pan_can_dp_sse3
+#define _op_blend_pt_p_caa_dpan_sse3 _op_blend_pt_p_caa_dp_sse3
+#define _op_blend_pt_pas_caa_dpan_sse3 _op_blend_pt_pas_caa_dp_sse3
+#define _op_blend_pt_pan_caa_dpan_sse3 _op_blend_pt_pan_caa_dp_sse3
+
+static void
+init_blend_pixel_color_pt_funcs_sse3(void)
+{
+   op_blend_pt_funcs[SP][SM_N][SC][DP][CPU_SSE3] = _op_blend_pt_p_c_dp_sse3;
+   op_blend_pt_funcs[SP_AS][SM_N][SC][DP][CPU_SSE3] = _op_blend_pt_pas_c_dp_sse3;
+   op_blend_pt_funcs[SP_AN][SM_N][SC][DP][CPU_SSE3] = _op_blend_pt_pan_c_dp_sse3;
+   op_blend_pt_funcs[SP][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_pt_p_can_dp_sse3;
+   op_blend_pt_funcs[SP_AS][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_pt_pas_can_dp_sse3;
+   op_blend_pt_funcs[SP_AN][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_pt_pan_can_dp_sse3;
+   op_blend_pt_funcs[SP][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_pt_p_caa_dp_sse3;
+   op_blend_pt_funcs[SP_AS][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_pt_pas_caa_dp_sse3;
+   op_blend_pt_funcs[SP_AN][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_pt_pan_caa_dp_sse3;
+
+   op_blend_pt_funcs[SP][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_pt_p_c_dpan_sse3;
+   op_blend_pt_funcs[SP_AS][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_pt_pas_c_dpan_sse3;
+   op_blend_pt_funcs[SP_AN][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_pt_pan_c_dpan_sse3;
+   op_blend_pt_funcs[SP][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_pt_p_can_dpan_sse3;
+   op_blend_pt_funcs[SP_AS][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_pt_pas_can_dpan_sse3;
+   op_blend_pt_funcs[SP_AN][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_pt_pan_can_dpan_sse3;
+   op_blend_pt_funcs[SP][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_pt_p_caa_dpan_sse3;
+   op_blend_pt_funcs[SP_AS][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_pt_pas_caa_dpan_sse3;
+   op_blend_pt_funcs[SP_AN][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_pt_pan_caa_dpan_sse3;
+}
+
+/*-----*/
+
+/* blend_rel pixel x color -> dst */
+
+static void
+_op_blend_rel_p_c_dp_sse3(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c, DATA32 *d, int l) {
+
+   int alpha;
+
+   const __m128i c_packed = _mm_set_epi32(c, c, c, c);
+
+   LOOP_ALIGNED_U1_A48_SSE3(d, l,
+      { /* UOP */
+
+         DATA32 sc = MUL4_SYM(c, *s);
+         alpha = 256 - (sc >> 24);
+         *d = MUL_SYM(*d >> 24, sc) + MUL_256(alpha, *d);
+         d++; s++; l--;
+      },
+      { /* A4OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i sc0 = mul4_sym_sse3(c_packed, s0);
+         __m128i a0 = sub4_alpha_sse3(sc0);
+
+         __m128i l0 = mul_sym_sse3(_mm_srli_epi32(d0, 24), sc0);
+         __m128i r0 = mul_256_sse3(a0, d0);
+
+         r0 = _mm_add_epi32(l0, r0);
+
+         _mm_store_si128((__m128i *)d, r0);
+
+         d += 4; s += 4; l -= 4;
+      },
+      {  /* A8OP */
+
+         __m128i s0 = _mm_lddqu_si128((__m128i *)s);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i s1 = _mm_lddqu_si128((__m128i *)(s+4));
+         __m128i d1 = _mm_load_si128((__m128i *)(d+4));
+
+         __m128i sc0 = mul4_sym_sse3(c_packed, s0);
+         __m128i sc1 = mul4_sym_sse3(c_packed, s1);
+
+         __m128i a0 = sub4_alpha_sse3(sc0);
+         __m128i a1 = sub4_alpha_sse3(sc1);
+
+         __m128i l0 = mul_sym_sse3(_mm_srli_epi32(d0, 24), sc0);
+         __m128i r0 = mul_256_sse3(a0, d0);
+
+         __m128i l1 = mul_sym_sse3(_mm_srli_epi32(d1, 24), sc1);
+         __m128i r1 = mul_256_sse3(a1, d1);
+
+         r0 = _mm_add_epi32(l0, r0);
+         r1 = _mm_add_epi32(l1, r1);
+
+         _mm_store_si128((__m128i *)d, r0);
+         _mm_store_si128((__m128i *)(d+4), r1);
+
+         d += 8; s += 8; l -= 8;
+      })
+}
+
+#define _op_blend_rel_pas_c_dp_sse3 _op_blend_rel_p_c_dp_sse3
+#define _op_blend_rel_pan_c_dp_sse3 _op_blend_rel_p_c_dp_sse3
+#define _op_blend_rel_p_can_dp_sse3 _op_blend_rel_p_c_dp_sse3
+#define _op_blend_rel_pas_can_dp_sse3 _op_blend_rel_p_c_dp_sse3
+#define _op_blend_rel_pan_can_dp_sse3 _op_blend_rel_p_c_dp_sse3
+#define _op_blend_rel_p_caa_dp_sse3 _op_blend_rel_p_c_dp_sse3
+#define _op_blend_rel_pas_caa_dp_sse3 _op_blend_rel_p_c_dp_sse3
+#define _op_blend_rel_pan_caa_dp_sse3 _op_blend_rel_p_c_dp_sse3
+
+#define _op_blend_rel_p_c_dpan_sse3 _op_blend_p_c_dpan_sse3
+#define _op_blend_rel_pas_c_dpan_sse3 _op_blend_pas_c_dpan_sse3
+#define _op_blend_rel_pan_c_dpan_sse3 _op_blend_pan_c_dpan_sse3
+#define _op_blend_rel_p_can_dpan_sse3 _op_blend_p_can_dpan_sse3
+#define _op_blend_rel_pas_can_dpan_sse3 _op_blend_pas_can_dpan_sse3
+#define _op_blend_rel_pan_can_dpan_sse3 _op_blend_pan_can_dpan_sse3
+#define _op_blend_rel_p_caa_dpan_sse3 _op_blend_p_caa_dpan_sse3
+#define _op_blend_rel_pas_caa_dpan_sse3 _op_blend_pas_caa_dpan_sse3
+#define _op_blend_rel_pan_caa_dpan_sse3 _op_blend_pan_caa_dpan_sse3
+
+static void
+init_blend_rel_pixel_color_span_funcs_sse3(void)
+{
+   op_blend_rel_span_funcs[SP][SM_N][SC][DP][CPU_SSE3] = _op_blend_rel_p_c_dp_sse3;
+   op_blend_rel_span_funcs[SP_AS][SM_N][SC][DP][CPU_SSE3] = _op_blend_rel_pas_c_dp_sse3;
+   op_blend_rel_span_funcs[SP_AN][SM_N][SC][DP][CPU_SSE3] = _op_blend_rel_pan_c_dp_sse3;
+   op_blend_rel_span_funcs[SP][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_rel_p_can_dp_sse3;
+   op_blend_rel_span_funcs[SP_AS][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_rel_pas_can_dp_sse3;
+   op_blend_rel_span_funcs[SP_AN][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_rel_pan_can_dp_sse3;
+   op_blend_rel_span_funcs[SP][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_rel_p_caa_dp_sse3;
+   op_blend_rel_span_funcs[SP_AS][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_rel_pas_caa_dp_sse3;
+   op_blend_rel_span_funcs[SP_AN][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_rel_pan_caa_dp_sse3;
+
+   op_blend_rel_span_funcs[SP][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_rel_p_c_dpan_sse3;
+   op_blend_rel_span_funcs[SP_AS][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_rel_pas_c_dpan_sse3;
+   op_blend_rel_span_funcs[SP_AN][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_rel_pan_c_dpan_sse3;
+   op_blend_rel_span_funcs[SP][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_rel_p_can_dpan_sse3;
+   op_blend_rel_span_funcs[SP_AS][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_rel_pas_can_dpan_sse3;
+   op_blend_rel_span_funcs[SP_AN][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_rel_pan_can_dpan_sse3;
+   op_blend_rel_span_funcs[SP][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_rel_p_caa_dpan_sse3;
+   op_blend_rel_span_funcs[SP_AS][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_rel_pas_caa_dpan_sse3;
+   op_blend_rel_span_funcs[SP_AN][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_rel_pan_caa_dpan_sse3;
+}
+
+#define _op_blend_rel_pt_p_c_dp_sse3 NULL
+
+#define _op_blend_rel_pt_pas_c_dp_sse3 _op_blend_rel_pt_p_c_dp_sse3
+#define _op_blend_rel_pt_pan_c_dp_sse3 _op_blend_rel_pt_p_c_dp_sse3
+#define _op_blend_rel_pt_p_can_dp_sse3 _op_blend_rel_pt_p_c_dp_sse3
+#define _op_blend_rel_pt_pas_can_dp_sse3 _op_blend_rel_pt_p_c_dp_sse3
+#define _op_blend_rel_pt_pan_can_dp_sse3 _op_blend_rel_pt_p_c_dp_sse3
+#define _op_blend_rel_pt_p_caa_dp_sse3 _op_blend_rel_pt_p_c_dp_sse3
+#define _op_blend_rel_pt_pas_caa_dp_sse3 _op_blend_rel_pt_p_c_dp_sse3
+#define _op_blend_rel_pt_pan_caa_dp_sse3 _op_blend_rel_pt_p_c_dp_sse3
+
+#define _op_blend_rel_pt_p_c_dpan_sse3 _op_blend_pt_p_c_dpan_sse3
+#define _op_blend_rel_pt_pas_c_dpan_sse3 _op_blend_pt_pas_c_dpan_sse3
+#define _op_blend_rel_pt_pan_c_dpan_sse3 _op_blend_pt_pan_c_dpan_sse3
+#define _op_blend_rel_pt_p_can_dpan_sse3 _op_blend_pt_p_can_dpan_sse3
+#define _op_blend_rel_pt_pas_can_dpan_sse3 _op_blend_pt_pas_can_dpan_sse3
+#define _op_blend_rel_pt_pan_can_dpan_sse3 _op_blend_pt_pan_can_dpan_sse3
+#define _op_blend_rel_pt_p_caa_dpan_sse3 _op_blend_pt_p_caa_dpan_sse3
+#define _op_blend_rel_pt_pas_caa_dpan_sse3 _op_blend_pt_pas_caa_dpan_sse3
+#define _op_blend_rel_pt_pan_caa_dpan_sse3 _op_blend_pt_pan_caa_dpan_sse3
+
+static void
+init_blend_rel_pixel_color_pt_funcs_sse3(void)
+{
+   op_blend_rel_pt_funcs[SP][SM_N][SC][DP][CPU_SSE3] = _op_blend_rel_pt_p_c_dp_sse3;
+   op_blend_rel_pt_funcs[SP_AS][SM_N][SC][DP][CPU_SSE3] = _op_blend_rel_pt_pas_c_dp_sse3;
+   op_blend_rel_pt_funcs[SP_AN][SM_N][SC][DP][CPU_SSE3] = _op_blend_rel_pt_pan_c_dp_sse3;
+   op_blend_rel_pt_funcs[SP][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_rel_pt_p_can_dp_sse3;
+   op_blend_rel_pt_funcs[SP_AS][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_rel_pt_pas_can_dp_sse3;
+   op_blend_rel_pt_funcs[SP_AN][SM_N][SC_AN][DP][CPU_SSE3] = _op_blend_rel_pt_pan_can_dp_sse3;
+   op_blend_rel_pt_funcs[SP][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_rel_pt_p_caa_dp_sse3;
+   op_blend_rel_pt_funcs[SP_AS][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_rel_pt_pas_caa_dp_sse3;
+   op_blend_rel_pt_funcs[SP_AN][SM_N][SC_AA][DP][CPU_SSE3] = _op_blend_rel_pt_pan_caa_dp_sse3;
+
+   op_blend_rel_pt_funcs[SP][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_rel_pt_p_c_dpan_sse3;
+   op_blend_rel_pt_funcs[SP_AS][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_rel_pt_pas_c_dpan_sse3;
+   op_blend_rel_pt_funcs[SP_AN][SM_N][SC][DP_AN][CPU_SSE3] = _op_blend_rel_pt_pan_c_dpan_sse3;
+   op_blend_rel_pt_funcs[SP][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_rel_pt_p_can_dpan_sse3;
+   op_blend_rel_pt_funcs[SP_AS][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_rel_pt_pas_can_dpan_sse3;
+   op_blend_rel_pt_funcs[SP_AN][SM_N][SC_AN][DP_AN][CPU_SSE3] = _op_blend_rel_pt_pan_can_dpan_sse3;
+   op_blend_rel_pt_funcs[SP][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_rel_pt_p_caa_dpan_sse3;
+   op_blend_rel_pt_funcs[SP_AS][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_rel_pt_pas_caa_dpan_sse3;
+   op_blend_rel_pt_funcs[SP_AN][SM_N][SC_AA][DP_AN][CPU_SSE3] = _op_blend_rel_pt_pan_caa_dpan_sse3;
+}
+
+#endif