VC1: merge idct8x8, coeff adjustments and put_pixels.
[ffmpeg.git] / libavcodec / ppc / vc1dsp_altivec.c
index 5a0dddb..b5cc71c 100644 (file)
@@ -130,7 +130,8 @@ do { \
 
 /** Do inverse transform on 8x8 block
 */
-static void vc1_inv_trans_8x8_altivec(DCTELEM block[64])
+static void vc1_inv_trans_8x8_altivec(DCTELEM block[64],
+                                      int sign, int rangered)
 {
     vector signed short src0, src1, src2, src3, src4, src5, src6, src7;
     vector signed int s0, s1, s2, s3, s4, s5, s6, s7;
@@ -144,7 +145,9 @@ static void vc1_inv_trans_8x8_altivec(DCTELEM block[64])
     const vector unsigned int vec_2 = vec_splat_u32(2);
     const vector  signed int vec_1s = vec_splat_s32(1);
     const vector unsigned int vec_1 = vec_splat_u32(1);
-
+    const vector unsigned short rangered_shift = vec_splat_u16(1);
+    const vector   signed short signed_bias = vec_sl(vec_splat_u16(4),
+                                                     vec_splat_u16(4));
 
     src0 = vec_ld(  0, block);
     src1 = vec_ld( 16, block);
@@ -214,6 +217,27 @@ static void vc1_inv_trans_8x8_altivec(DCTELEM block[64])
     src6 = vec_pack(sE, s6);
     src7 = vec_pack(sF, s7);
 
+    if (rangered) {
+        if (!sign) {
+            vec_sub(src0, signed_bias);
+            vec_sub(src1, signed_bias);
+            vec_sub(src2, signed_bias);
+            vec_sub(src3, signed_bias);
+            vec_sub(src4, signed_bias);
+            vec_sub(src5, signed_bias);
+            vec_sub(src6, signed_bias);
+            vec_sub(src7, signed_bias);
+        }
+        vec_sl(src0, rangered_shift);
+        vec_sl(src1, rangered_shift);
+        vec_sl(src2, rangered_shift);
+        vec_sl(src3, rangered_shift);
+        vec_sl(src4, rangered_shift);
+        vec_sl(src5, rangered_shift);
+        vec_sl(src6, rangered_shift);
+        vec_sl(src7, rangered_shift);
+    }
+
     vec_st(src0,  0, block);
     vec_st(src1, 16, block);
     vec_st(src2, 32, block);
@@ -224,6 +248,36 @@ static void vc1_inv_trans_8x8_altivec(DCTELEM block[64])
     vec_st(src7,112, block);
 }
 
+static void vc1_inv_trans_8x8_add_altivec(uint8_t *dest, int stride, DCTELEM *b)
+{
+    vc1_inv_trans_8x8_altivec(b, 0, 0);
+    ff_add_pixels_clamped_c(b, dest, stride);
+}
+
+static void vc1_inv_trans_8x8_put_signed_altivec(uint8_t *dest, int stride, DCTELEM *b)
+{
+    vc1_inv_trans_8x8_altivec(b, 1, 0);
+    ff_put_signed_pixels_clamped_c(b, dest, stride);
+}
+
+static void vc1_inv_trans_8x8_put_signed_rangered_altivec(uint8_t *dest, int stride, DCTELEM *b)
+{
+    vc1_inv_trans_8x8_altivec(b, 1, 1);
+    ff_put_signed_pixels_clamped_c(b, dest, stride);
+}
+
+static void vc1_inv_trans_8x8_put_altivec(uint8_t *dest, int stride, DCTELEM *b)
+{
+    vc1_inv_trans_8x8_altivec(b, 0, 0);
+    ff_put_pixels_clamped_c(b, dest, stride);
+}
+
+static void vc1_inv_trans_8x8_put_rangered_altivec(uint8_t *dest, int stride, DCTELEM *b)
+{
+    vc1_inv_trans_8x8_altivec(b, 0, 1);
+    ff_put_pixels_clamped_c(b, dest, stride);
+}
+
 /** Do inverse transform on 8x4 part of block
 */
 static void vc1_inv_trans_8x4_altivec(uint8_t *dest, int stride, DCTELEM *block)
@@ -342,7 +396,11 @@ void ff_vc1dsp_init_altivec(VC1DSPContext* dsp)
     if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
         return;
 
-    dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_altivec;
+    dsp->vc1_inv_trans_8x8_add = vc1_inv_trans_8x8_add_altivec;
+    dsp->vc1_inv_trans_8x8_put_signed[0] = vc1_inv_trans_8x8_put_signed_altivec;
+    dsp->vc1_inv_trans_8x8_put_signed[1] = vc1_inv_trans_8x8_put_signed_rangered_altivec;
+    dsp->vc1_inv_trans_8x8_put[0] = vc1_inv_trans_8x8_put_altivec;
+    dsp->vc1_inv_trans_8x8_put[1] = vc1_inv_trans_8x8_put_rangered_altivec;
     dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_altivec;
     dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = put_no_rnd_vc1_chroma_mc8_altivec;
     dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = avg_no_rnd_vc1_chroma_mc8_altivec;