evas: add SSE3 op_blend_mask_color blend functions
authorlucas <lucas@7cbeb6ba-43b4-40fd-8cce-4c39aea84d33>
Fri, 30 Sep 2011 02:36:15 +0000 (02:36 +0000)
committerlucas <lucas@7cbeb6ba-43b4-40fd-8cce-4c39aea84d33>
Fri, 30 Sep 2011 02:36:15 +0000 (02:36 +0000)
_op_blend_mas_c_dp_sse3
_op_blend_mas_can_dp_sse3
_op_blend_rel_mas_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@63700 7cbeb6ba-43b4-40fd-8cce-4c39aea84d33

src/lib/engines/common/evas_op_blend/op_blend_mask_color_sse3.c [new file with mode: 0644]

diff --git a/src/lib/engines/common/evas_op_blend/op_blend_mask_color_sse3.c b/src/lib/engines/common/evas_op_blend/op_blend_mask_color_sse3.c
new file mode 100644 (file)
index 0000000..13d71f8
--- /dev/null
@@ -0,0 +1,319 @@
+/* blend mask x color -> dst */
+
+#ifdef BUILD_SSE3
+
+static void
+_op_blend_mas_c_dp_sse3(DATA32 *s __UNUSED__, DATA8 *m, 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 */
+
+         DATA32 a = *m;
+         DATA32 mc = MUL_SYM(a, c);
+         a = 256 - (mc >> 24);
+         *d = mc + MUL_256(a, *d);
+         m++;  d++; l--;
+      },
+      { /* A4OP */
+
+         if ((m[3] | m[2] | m[1] | m[0]) == 0) {
+            m += 4; d += 4; l -= 4;
+            continue;
+         }
+
+         __m128i m0 = _mm_set_epi32(m[3], m[2], m[1], m[0]);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i mc0 = mul_sym_sse3(m0, c_packed);
+         __m128i  a0 = sub4_alpha_sse3(mc0);
+         __m128i mul0 = mul_256_sse3(a0, d0);
+
+         mul0 = _mm_add_epi32(mul0, mc0);
+
+         _mm_store_si128((__m128i *)d, mul0);
+
+         m += 4; d += 4; l -= 4;
+      },
+      { /* A8OP */
+
+         if((m[7] | m[6] | m[5] | m[4] | m[3] | m[2] | m[1] | m[0]) == 0) {
+            m += 8; d += 8; l -= 8;
+            continue;
+         }
+
+         __m128i m0 = _mm_set_epi32(m[3], m[2], m[1], m[0]);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i m1 = _mm_set_epi32(m[7], m[6], m[5], m[4]);
+         __m128i d1 = _mm_load_si128((__m128i *)(d+4));
+
+         __m128i mc0 = mul_sym_sse3(m0, c_packed);
+         __m128i a0  = sub4_alpha_sse3(mc0);
+         __m128i mul0 = mul_256_sse3(a0, d0);
+
+         mul0 = _mm_add_epi32(mc0, mul0);
+
+         __m128i mc1 = mul_sym_sse3(m1, c_packed);
+         __m128i a1  = sub4_alpha_sse3(mc1);
+         __m128i mul1 = mul_256_sse3(a1, d1);
+
+         mul1 = _mm_add_epi32(mc1, mul1);
+
+         _mm_store_si128((__m128i *)d, mul0);
+         _mm_store_si128((__m128i *)(d+4), mul1);
+
+         m += 8; d += 8; l -= 8;
+      })
+}
+
+static void
+_op_blend_mas_can_dp_sse3(DATA32 *s __UNUSED__, DATA8 *m, DATA32 c, DATA32 *d, int l) {
+
+   DATA32 alpha;
+
+   const __m128i one = _mm_set_epi32(1, 1, 1, 1);
+   const __m128i c_packed = _mm_set_epi32(c, c, c, c);
+
+   LOOP_ALIGNED_U1_A48_SSE3(d, l,
+      { /* UOP */
+
+         alpha = *m;
+         switch(alpha)
+           {
+           case 0:
+              break;
+           case 255:
+              *d = c;
+              break;
+           default:
+              alpha++;
+              *d = INTERP_256(alpha, c, *d);
+              break;
+           }
+         m++;  d++; l--;
+      },
+      { /* A4OP */
+
+         if ((m[3] | m[2] | m[1] | m[0]) == 0) {
+            m += 4; d += 4; l -= 4;
+            continue;
+         }
+
+         __m128i m0 = _mm_set_epi32(m[3], m[2], m[1], m[0]);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i zm0 = _mm_cmpeq_epi32(m0, _mm_setzero_si128());
+
+         m0 = _mm_add_epi32(one, m0);
+
+         __m128i r0 = interp4_256_sse3(m0, c_packed, d0);
+
+         r0 = _mm_and_si128(~zm0, r0);
+         d0 = _mm_and_si128(zm0, d0);
+
+         d0 = _mm_add_epi32(r0, d0);
+
+         _mm_store_si128((__m128i *)d, d0);
+
+         m += 4; d += 4; l -= 4;
+      },
+      { /* A8OP */
+
+         if ((m[7] | m[6] | m[5] | m[4] | m[3] | m[2] | m[1] | m[0]) == 0) {
+            m += 8; d += 8; l -= 8;
+            continue;
+         }
+
+         __m128i m0 = _mm_set_epi32(m[3], m[2], m[1], m[0]);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i m1 = _mm_set_epi32(m[7], m[6], m[5], m[4]);
+         __m128i d1 = _mm_load_si128((__m128i *)(d+4));
+
+         __m128i zm0 = _mm_cmpeq_epi32(m0, _mm_setzero_si128());
+         __m128i zm1 = _mm_cmpeq_epi32(m1, _mm_setzero_si128());
+
+         m0 = _mm_add_epi32(one, m0);
+         m1 = _mm_add_epi32(one, m1);
+
+         __m128i r0 = interp4_256_sse3(m0, c_packed, d0);
+         __m128i r1 = interp4_256_sse3(m1, c_packed, d1);
+
+         r0 = _mm_and_si128(~zm0, r0);
+         d0 = _mm_and_si128(zm0, d0);
+
+         r1 = _mm_and_si128(~zm1, r1);
+         d1 = _mm_and_si128(zm1, d1);
+
+         d0 = _mm_add_epi32(d0, r0);
+         d1 = _mm_add_epi32(d1, r1);
+
+         _mm_store_si128((__m128i *)d, d0);
+         _mm_store_si128((__m128i *)(d+4), d1);
+
+         m += 8; d += 8; l -= 8;
+      })
+}
+
+#define _op_blend_mas_cn_dp_sse3 _op_blend_mas_can_dp_sse3
+#define _op_blend_mas_caa_dp_sse3 _op_blend_mas_c_dp_sse3
+
+#define _op_blend_mas_c_dpan_sse3 _op_blend_mas_c_dp_sse3
+#define _op_blend_mas_cn_dpan_sse3 _op_blend_mas_cn_dp_sse3
+#define _op_blend_mas_can_dpan_sse3 _op_blend_mas_can_dp_sse3
+#define _op_blend_mas_caa_dpan_sse3 _op_blend_mas_caa_dp_sse3
+
+static void
+init_blend_mask_color_span_funcs_sse3(void)
+{
+   op_blend_span_funcs[SP_N][SM_AS][SC][DP][CPU_SSE3] = _op_blend_mas_c_dp_sse3;
+   op_blend_span_funcs[SP_N][SM_AS][SC_N][DP][CPU_SSE3] = _op_blend_mas_cn_dp_sse3;
+   op_blend_span_funcs[SP_N][SM_AS][SC_AN][DP][CPU_SSE3] = _op_blend_mas_can_dp_sse3;
+   op_blend_span_funcs[SP_N][SM_AS][SC_AA][DP][CPU_SSE3] = _op_blend_mas_caa_dp_sse3;
+
+   op_blend_span_funcs[SP_N][SM_AS][SC][DP_AN][CPU_SSE3] = _op_blend_mas_c_dpan_sse3;
+   op_blend_span_funcs[SP_N][SM_AS][SC_N][DP_AN][CPU_SSE3] = _op_blend_mas_cn_dpan_sse3;
+   op_blend_span_funcs[SP_N][SM_AS][SC_AN][DP_AN][CPU_SSE3] = _op_blend_mas_can_dpan_sse3;
+   op_blend_span_funcs[SP_N][SM_AS][SC_AA][DP_AN][CPU_SSE3] = _op_blend_mas_caa_dpan_sse3;
+}
+
+#define _op_blend_pt_mas_c_dp_sse3 NULL
+#define _op_blend_pt_mas_can_dp_sse3 NULL
+
+#define _op_blend_pt_mas_cn_dp_sse3 _op_blend_pt_mas_can_dp_sse3
+#define _op_blend_pt_mas_caa_dp_sse3 _op_blend_pt_mas_c_dp_sse3
+
+#define _op_blend_pt_mas_c_dpan_sse3 _op_blend_pt_mas_c_dp_sse3
+#define _op_blend_pt_mas_cn_dpan_sse3 _op_blend_pt_mas_cn_dp_sse3
+#define _op_blend_pt_mas_can_dpan_sse3 _op_blend_pt_mas_can_dp_sse3
+#define _op_blend_pt_mas_caa_dpan_sse3 _op_blend_pt_mas_caa_dp_sse3
+
+static void
+init_blend_mask_color_pt_funcs_sse3(void)
+{
+   op_blend_pt_funcs[SP_N][SM_AS][SC][DP][CPU_SSE3] = _op_blend_pt_mas_c_dp_sse3;
+   op_blend_pt_funcs[SP_N][SM_AS][SC_N][DP][CPU_SSE3] = _op_blend_pt_mas_cn_dp_sse3;
+   op_blend_pt_funcs[SP_N][SM_AS][SC_AN][DP][CPU_SSE3] = _op_blend_pt_mas_can_dp_sse3;
+   op_blend_pt_funcs[SP_N][SM_AS][SC_AA][DP][CPU_SSE3] = _op_blend_pt_mas_caa_dp_sse3;
+
+   op_blend_pt_funcs[SP_N][SM_AS][SC][DP_AN][CPU_SSE3] = _op_blend_pt_mas_c_dpan_sse3;
+   op_blend_pt_funcs[SP_N][SM_AS][SC_N][DP_AN][CPU_SSE3] = _op_blend_pt_mas_cn_dpan_sse3;
+   op_blend_pt_funcs[SP_N][SM_AS][SC_AN][DP_AN][CPU_SSE3] = _op_blend_pt_mas_can_dpan_sse3;
+   op_blend_pt_funcs[SP_N][SM_AS][SC_AA][DP_AN][CPU_SSE3] = _op_blend_pt_mas_caa_dpan_sse3;
+}
+
+/*-----*/
+
+/* blend_rel mask x color --> dst */
+
+static void
+_op_blend_rel_mas_c_dp_sse3(DATA32 *s __UNUSED__, DATA8 *m, 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 */
+
+         DATA32 mc = MUL_SYM(*m, c);
+         int alpha = 256 - (mc >> 24);
+         *d = MUL_SYM(*d >> 24, mc) + MUL_256(alpha, *d);
+         d++; m++; l--;
+      },
+      { /* A4OP */
+
+         __m128i m0 = _mm_set_epi32(m[3], m[2], m[1], m[0]);
+         __m128i d0 = _mm_load_si128((__m128i *) d);
+
+         __m128i mc0 = mul_sym_sse3(m0, c_packed);
+         __m128i a0 = sub4_alpha_sse3(mc0);
+
+         __m128i d0_sym = mul_sym_sse3(_mm_srli_epi32(d0, 24), mc0);
+         d0 = mul_256_sse3(a0, d0);
+
+         d0 = _mm_add_epi32(d0, d0_sym);
+
+         _mm_store_si128((__m128i *)d, d0);
+
+         d += 4; m += 4; l -= 4;
+      },
+      { /* A8OP */
+
+         __m128i m0 = _mm_set_epi32(m[3], m[2], m[1], m[0]);
+         __m128i d0 = _mm_load_si128((__m128i *)d);
+
+         __m128i m1 = _mm_set_epi32(m[7], m[6], m[5], m[4]);
+         __m128i d1 = _mm_load_si128((__m128i *)(d+4));
+
+         __m128i mc0 = mul_sym_sse3(m0, c_packed);
+         __m128i mc1 = mul_sym_sse3(m1, c_packed);
+
+         __m128i a0 = sub4_alpha_sse3(mc0);
+         __m128i a1 = sub4_alpha_sse3(mc1);
+
+         __m128i d0_sym = mul_sym_sse3(_mm_srli_epi32(d0, 24), mc0);
+         __m128i d1_sym = mul_sym_sse3(_mm_srli_epi32(d1, 24), mc1);
+
+         d0 = mul_256_sse3(a0, d0);
+         d1 = mul_256_sse3(a1, d1);
+
+         d0 = _mm_add_epi32(d0, d0_sym);
+         d1 = _mm_add_epi32(d1, d1_sym);
+
+         _mm_store_si128((__m128i *)d, d0);
+         _mm_store_si128((__m128i *)(d+4), d1);
+
+         d += 8; m += 8; l -= 8;
+      })
+}
+
+#define _op_blend_rel_mas_cn_dp_sse3 _op_blend_rel_mas_c_dp_sse3
+#define _op_blend_rel_mas_can_dp_sse3 _op_blend_rel_mas_c_dp_sse3
+#define _op_blend_rel_mas_caa_dp_sse3 _op_blend_rel_mas_c_dp_sse3
+
+#define _op_blend_rel_mas_c_dpan_sse3 _op_blend_mas_c_dpan_sse3
+#define _op_blend_rel_mas_cn_dpan_sse3 _op_blend_mas_cn_dpan_sse3
+#define _op_blend_rel_mas_can_dpan_sse3 _op_blend_mas_can_dpan_sse3
+#define _op_blend_rel_mas_caa_dpan_sse3 _op_blend_mas_caa_dpan_sse3
+
+static void
+init_blend_rel_mask_color_span_funcs_sse3(void)
+{
+   op_blend_rel_span_funcs[SP_N][SM_AS][SC][DP][CPU_SSE3] = _op_blend_rel_mas_c_dp_sse3;
+   op_blend_rel_span_funcs[SP_N][SM_AS][SC_N][DP][CPU_SSE3] = _op_blend_rel_mas_can_dp_sse3;
+   op_blend_rel_span_funcs[SP_N][SM_AS][SC_AN][DP][CPU_SSE3] = _op_blend_rel_mas_can_dp_sse3;
+   op_blend_rel_span_funcs[SP_N][SM_AS][SC_AA][DP][CPU_SSE3] = _op_blend_rel_mas_caa_dp_sse3;
+
+   op_blend_rel_span_funcs[SP_N][SM_AS][SC][DP_AN][CPU_SSE3] = _op_blend_rel_mas_c_dpan_sse3;
+   op_blend_rel_span_funcs[SP_N][SM_AS][SC_N][DP_AN][CPU_SSE3] = _op_blend_rel_mas_cn_dpan_sse3;
+   op_blend_rel_span_funcs[SP_N][SM_AS][SC_AN][DP_AN][CPU_SSE3] = _op_blend_rel_mas_can_dpan_sse3;
+   op_blend_rel_span_funcs[SP_N][SM_AS][SC_AA][DP_AN][CPU_SSE3] = _op_blend_rel_mas_caa_dpan_sse3;
+}
+
+#define _op_blend_rel_pt_mas_c_dp_sse3 NULL
+
+#define _op_blend_rel_pt_mas_cn_dp_sse3 _op_blend_rel_pt_mas_c_dp_sse3
+#define _op_blend_rel_pt_mas_can_dp_sse3 _op_blend_rel_pt_mas_c_dp_sse3
+#define _op_blend_rel_pt_mas_caa_dp_sse3 _op_blend_rel_pt_mas_c_dp_sse3
+
+#define _op_blend_rel_pt_mas_c_dpan_sse3 _op_blend_pt_mas_c_dpan_sse3
+#define _op_blend_rel_pt_mas_cn_dpan_sse3 _op_blend_pt_mas_cn_dpan_sse3
+#define _op_blend_rel_pt_mas_can_dpan_sse3 _op_blend_pt_mas_can_dpan_sse3
+#define _op_blend_rel_pt_mas_caa_dpan_sse3 _op_blend_pt_mas_caa_dpan_sse3
+
+static void
+init_blend_rel_mask_color_pt_funcs_sse3(void)
+{
+   op_blend_rel_pt_funcs[SP_N][SM_AS][SC][DP][CPU_SSE3] = _op_blend_rel_pt_mas_c_dp_sse3;
+   op_blend_rel_pt_funcs[SP_N][SM_AS][SC_N][DP][CPU_SSE3] = _op_blend_rel_pt_mas_cn_dp_sse3;
+   op_blend_rel_pt_funcs[SP_N][SM_AS][SC_AN][DP][CPU_SSE3] = _op_blend_rel_pt_mas_can_dp_sse3;
+   op_blend_rel_pt_funcs[SP_N][SM_AS][SC_AA][DP][CPU_SSE3] = _op_blend_rel_pt_mas_caa_dp_sse3;
+
+   op_blend_rel_pt_funcs[SP_N][SM_AS][SC][DP_AN][CPU_SSE3] = _op_blend_rel_pt_mas_c_dpan_sse3;
+   op_blend_rel_pt_funcs[SP_N][SM_AS][SC_N][DP_AN][CPU_SSE3] = _op_blend_rel_pt_mas_cn_dpan_sse3;
+   op_blend_rel_pt_funcs[SP_N][SM_AS][SC_AN][DP_AN][CPU_SSE3] = _op_blend_rel_pt_mas_can_dpan_sse3;
+   op_blend_rel_pt_funcs[SP_N][SM_AS][SC_AA][DP_AN][CPU_SSE3] = _op_blend_rel_pt_mas_caa_dpan_sse3;
+}
+
+#endif