more texturing optimizations

git-svn-id: svn://svn.icculus.org/twilight/trunk/darkplaces@10768 d7cf8633-e32d-0410-b094-e92efae38249

diff --git a/dpsoftrast.c b/dpsoftrast.c
index e2a8b0a5feac2c627a0f144ce3d2269b62808d45..83f1b94d639fd9671f806cb1354ba845783cc08b 100644 (file)
--- a/dpsoftrast.c
+++ b/dpsoftrast.c
@@ -1733,7 +1733,7 @@ void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(const DPSOFTRAST_State_Draw_Span
        tcscale = _mm_cvtepi32_ps(tcsize);
        data = _mm_mul_ps(_mm_shuffle_ps(data, data, _MM_SHUFFLE(1, 0, 1, 0)), tcscale);
        slope = _mm_mul_ps(_mm_shuffle_ps(slope, slope, _MM_SHUFFLE(1, 0, 1, 0)), tcscale);
-       endtc = _mm_sub_ps(_mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(startx))), _mm_set1_ps(zf[startx])), _mm_set1_ps(0.5f));
+       endtc = _mm_sub_ps(_mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(startx))), _mm_load1_ps(&zf[startx])), _mm_set1_ps(0.5f));
        endsubtc = _mm_cvtps_epi32(_mm_mul_ps(endtc, _mm_set1_ps(65536.0f)));
        tcoffset = _mm_add_epi32(_mm_slli_epi32(_mm_shuffle_epi32(tcsize, _MM_SHUFFLE(0, 0, 0, 0)), 18), _mm_set1_epi32(4));
        tcmax = filter ? _mm_packs_epi32(tcmask, tcmask) : _mm_slli_epi32(tcmask, 16);  
@@ -1748,16 +1748,16 @@ void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(const DPSOFTRAST_State_Draw_Span
                }       
                tc = endtc;
                subtc = endsubtc;
-               endtc = _mm_sub_ps(_mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(nextsub))), _mm_set1_ps(zf[nextsub])), _mm_set1_ps(0.5f));
+               endtc = _mm_sub_ps(_mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(nextsub))), _mm_load1_ps(&zf[nextsub])), _mm_set1_ps(0.5f));
                substep = _mm_cvtps_epi32(_mm_mul_ps(_mm_sub_ps(endtc, tc), subscale));
                endsubtc = _mm_cvtps_epi32(_mm_mul_ps(endtc, _mm_set1_ps(65536.0f)));
+               subtc = _mm_unpacklo_epi64(subtc, _mm_add_epi32(subtc, substep));
+               substep = _mm_slli_epi32(substep, 1);
                if (filter)
                {
                        __m128i tcrange = _mm_srai_epi32(_mm_unpacklo_epi64(subtc, _mm_add_epi32(endsubtc, substep)), 16);
                        if (_mm_movemask_epi8(_mm_andnot_si128(_mm_cmplt_epi32(tcrange, _mm_setzero_si128()), _mm_cmplt_epi32(tcrange, tcmask))) == 0xFFFF)
                        {
-                               subtc = _mm_unpacklo_epi64(subtc, _mm_add_epi32(subtc, substep));
-                               substep = _mm_slli_epi32(substep, 1);
                                for (; x + 1 <= endsub; x += 2, subtc = _mm_add_epi32(subtc, substep))
                                {
                                        __m128i tci = _mm_shufflehi_epi16(_mm_shufflelo_epi16(subtc, _MM_SHUFFLE(3, 1, 3, 1)), _MM_SHUFFLE(3, 1, 3, 1)), pix1, pix2, pix3, pix4, fracm;
@@ -1800,9 +1800,43 @@ void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(const DPSOFTRAST_State_Draw_Span
                        }
                        else if (flags & DPSOFTRAST_TEXTURE_FLAG_CLAMPTOEDGE)
                        {
-                               for (; x <= endsub; x++, subtc = _mm_add_epi32(subtc, substep))
+                               for (; x + 1 <= endsub; x += 2, subtc = _mm_add_epi32(subtc, substep))
                                {
-                                       __m128i tci = _mm_shufflehi_epi16(_mm_shufflelo_epi16(subtc, _MM_SHUFFLE(3, 1, 3, 1)), _MM_SHUFFLE(3, 1, 3, 1)), pix1, pix2, fracm;
+                                       __m128i tci = _mm_shuffle_epi32(_mm_shufflelo_epi16(subtc, _MM_SHUFFLE(3, 1, 3, 1)), _MM_SHUFFLE(1, 0, 1, 0)), pix1, pix2, pix3, pix4, fracm;
+                                       tci = _mm_min_epi16(_mm_max_epi16(_mm_add_epi16(tci, _mm_setr_epi32(0, 1, 0x10000, 0x10001)), _mm_setzero_si128()), tcmax);
+                                       tci = _mm_madd_epi16(tci, tcoffset);
+                                       pix1 = _mm_unpacklo_epi8(_mm_unpacklo_epi32(_mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(tci)]),
+                                                                                                                               _mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(1, 1, 1, 1)))])),
+                                                                                       _mm_setzero_si128());
+                                       pix2 = _mm_unpacklo_epi8(_mm_unpacklo_epi32(_mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(2, 2, 2, 2)))]),
+                                                                                                                               _mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(3, 3, 3, 3)))])),
+                                                                                       _mm_setzero_si128());
+                                       tci = _mm_shuffle_epi32(_mm_shufflehi_epi16(subtc, _MM_SHUFFLE(3, 1, 3, 1)), _MM_SHUFFLE(3, 2, 3, 2));
+                                       tci = _mm_and_si128(_mm_add_epi16(tci, _mm_setr_epi32(0, 1, 0x10000, 0x10001)), tcmax);
+                                       tci = _mm_madd_epi16(tci, tcoffset);
+                                       pix3 = _mm_unpacklo_epi8(_mm_unpacklo_epi32(_mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(tci)]),
+                                                                                                                               _mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(1, 1, 1, 1)))])),
+                                                                                       _mm_setzero_si128());
+                                       pix4 = _mm_unpacklo_epi8(_mm_unpacklo_epi32(_mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(2, 2, 2, 2)))]),
+                                                                                                                               _mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(3, 3, 3, 3)))])),
+                                                                                       _mm_setzero_si128());
+                                       fracm = _mm_srli_epi16(subtc, 1);
+                                       pix1 = _mm_add_epi16(pix1,
+                                                                                _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(pix2, pix1), 1),
+                                                                                                                _mm_shuffle_epi32(_mm_shufflelo_epi16(fracm, _MM_SHUFFLE(2, 2, 2, 2)), _MM_SHUFFLE(1, 0, 1, 0))));
+                                       pix3 = _mm_add_epi16(pix3,
+                                                                                _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(pix4, pix3), 1),
+                                                                                                                _mm_shuffle_epi32(_mm_shufflehi_epi16(fracm, _MM_SHUFFLE(2, 2, 2, 2)), _MM_SHUFFLE(3, 2, 3, 2))));
+                                       pix2 = _mm_unpacklo_epi64(pix1, pix3);
+                                       pix4 = _mm_unpackhi_epi64(pix1, pix3);
+                                       pix2 = _mm_add_epi16(pix2,
+                                                                                _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(pix4, pix2), 1),
+                                                                                                                _mm_shufflehi_epi16(_mm_shufflelo_epi16(fracm, _MM_SHUFFLE(0, 0, 0, 0)), _MM_SHUFFLE(0, 0, 0, 0))));
+                                       _mm_storel_epi64((__m128i *)&outi[x], _mm_packus_epi16(pix2, _mm_shufflelo_epi16(pix2, _MM_SHUFFLE(3, 2, 3, 2))));
+                               }
+                               if (x <= endsub)
+                               {
+                                       __m128i tci = _mm_shuffle_epi32(_mm_shufflelo_epi16(subtc, _MM_SHUFFLE(3, 1, 3, 1)), _MM_SHUFFLE(1, 0, 1, 0)), pix1, pix2, fracm;
                                        tci = _mm_min_epi16(_mm_max_epi16(_mm_add_epi16(tci, _mm_setr_epi32(0, 1, 0x10000, 0x10001)), _mm_setzero_si128()), tcmax);
                                        tci = _mm_madd_epi16(tci, tcoffset);
                                        pix1 = _mm_unpacklo_epi8(_mm_unpacklo_epi32(_mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(tci)]), 
@@ -1820,13 +1854,48 @@ void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(const DPSOFTRAST_State_Draw_Span
                                                                                 _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(pix2, pix1), 1),
                                                                                                                 _mm_shufflelo_epi16(fracm, _MM_SHUFFLE(0, 0, 0, 0))));
                                        outi[x] = _mm_cvtsi128_si32(_mm_packus_epi16(pix1, pix1));
+                                       x++;
                                }
                        }
                        else
                        {
-                               for (; x <= endsub; x++, subtc = _mm_add_epi32(subtc, substep))
+                               for (; x + 1 <= endsub; x += 2, subtc = _mm_add_epi32(subtc, substep))
                                {
-                                       __m128i tci = _mm_shufflehi_epi16(_mm_shufflelo_epi16(subtc, _MM_SHUFFLE(3, 1, 3, 1)), _MM_SHUFFLE(3, 1, 3, 1)), pix1, pix2, fracm;
+                                       __m128i tci = _mm_shuffle_epi32(_mm_shufflelo_epi16(subtc, _MM_SHUFFLE(3, 1, 3, 1)), _MM_SHUFFLE(1, 0, 1, 0)), pix1, pix2, pix3, pix4, fracm;
+                                       tci = _mm_and_si128(_mm_add_epi16(tci, _mm_setr_epi32(0, 1, 0x10000, 0x10001)), tcmax);
+                                       tci = _mm_madd_epi16(tci, tcoffset);
+                                       pix1 = _mm_unpacklo_epi8(_mm_unpacklo_epi32(_mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(tci)]),
+                                                                                                                               _mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(1, 1, 1, 1)))])),
+                                                                                       _mm_setzero_si128());
+                                       pix2 = _mm_unpacklo_epi8(_mm_unpacklo_epi32(_mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(2, 2, 2, 2)))]),
+                                                                                                                               _mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(3, 3, 3, 3)))])),
+                                                                                       _mm_setzero_si128());
+                                       tci = _mm_shuffle_epi32(_mm_shufflehi_epi16(subtc, _MM_SHUFFLE(3, 1, 3, 1)), _MM_SHUFFLE(3, 2, 3, 2));
+                                       tci = _mm_and_si128(_mm_add_epi16(tci, _mm_setr_epi32(0, 1, 0x10000, 0x10001)), tcmax);
+                                       tci = _mm_madd_epi16(tci, tcoffset);
+                                       pix3 = _mm_unpacklo_epi8(_mm_unpacklo_epi32(_mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(tci)]),
+                                                                                                                               _mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(1, 1, 1, 1)))])),
+                                                                                       _mm_setzero_si128());
+                                       pix4 = _mm_unpacklo_epi8(_mm_unpacklo_epi32(_mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(2, 2, 2, 2)))]),
+                                                                                                                               _mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(_mm_shuffle_epi32(tci, _MM_SHUFFLE(3, 3, 3, 3)))])),
+                                                                                       _mm_setzero_si128());
+                                       fracm = _mm_srli_epi16(subtc, 1);
+                                       pix1 = _mm_add_epi16(pix1,
+                                                                                _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(pix2, pix1), 1),
+                                                                                                                _mm_shuffle_epi32(_mm_shufflelo_epi16(fracm, _MM_SHUFFLE(2, 2, 2, 2)), _MM_SHUFFLE(1, 0, 1, 0))));
+                                       pix3 = _mm_add_epi16(pix3,
+                                                                                _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(pix4, pix3), 1),
+                                                                                                                _mm_shuffle_epi32(_mm_shufflehi_epi16(fracm, _MM_SHUFFLE(2, 2, 2, 2)), _MM_SHUFFLE(3, 2, 3, 2))));
+                                       pix2 = _mm_unpacklo_epi64(pix1, pix3);
+                                       pix4 = _mm_unpackhi_epi64(pix1, pix3);
+                                       pix2 = _mm_add_epi16(pix2,
+                                                                                _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(pix4, pix2), 1),
+                                                                                                                _mm_shufflehi_epi16(_mm_shufflelo_epi16(fracm, _MM_SHUFFLE(0, 0, 0, 0)), _MM_SHUFFLE(0, 0, 0, 0))));
+                                       _mm_storel_epi64((__m128i *)&outi[x], _mm_packus_epi16(pix2, _mm_shufflelo_epi16(pix2, _MM_SHUFFLE(3, 2, 3, 2))));
+                               }
+                               if (x <= endsub)
+                               {
+                                       __m128i tci = _mm_shuffle_epi32(_mm_shufflelo_epi16(subtc, _MM_SHUFFLE(3, 1, 3, 1)), _MM_SHUFFLE(1, 0, 1, 0)), pix1, pix2, fracm;
                                        tci = _mm_and_si128(_mm_add_epi16(tci, _mm_setr_epi32(0, 1, 0x10000, 0x10001)), tcmax);
                                        tci = _mm_madd_epi16(tci, tcoffset);
                                        pix1 = _mm_unpacklo_epi8(_mm_unpacklo_epi32(_mm_cvtsi32_si128(*(const int *)&pixelbase[_mm_cvtsi128_si32(tci)]),                                                                                        
@@ -1844,13 +1913,12 @@ void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(const DPSOFTRAST_State_Draw_Span
                                                                                 _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(pix2, pix1), 1),
                                                                                                                 _mm_shufflelo_epi16(fracm, _MM_SHUFFLE(0, 0, 0, 0))));
                                        outi[x] = _mm_cvtsi128_si32(_mm_packus_epi16(pix1, pix1));
+                                       x++;
                                }
                        }
                }
                else
                {
-                       subtc = _mm_unpacklo_epi64(subtc, _mm_add_epi32(subtc, substep));
-                       substep = _mm_slli_epi32(substep, 1);
                        if (flags & DPSOFTRAST_TEXTURE_FLAG_CLAMPTOEDGE)
                        {
                                for (; x + 1 <= endsub; x += 2, subtc = _mm_add_epi32(subtc, substep))
@@ -3413,7 +3481,7 @@ void DPSOFTRAST_Draw_ProcessSpans(void)
                        if (dpsoftrast.fb_colorpixels[0] && dpsoftrast.fb_colormask)
                                DPSOFTRAST_ShaderModeTable[dpsoftrast.shader_mode].Span(span);
                        if (dpsoftrast.user.depthmask)
-                               for (x = 0, d = depth;x < span->length;x++, d += depthslope)
+                               for (x = startx, d = depth + depthslope*startx;x < endx;x++, d += depthslope)
                                        if (pixelmask[x])
                                                depthpixel[x] = d;
                }