1 /* 2 * Copyright (C) 2012 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17 18 #include "rsCpuIntrinsic.h" 19 #include "rsCpuIntrinsicInlines.h" 20 21 using namespace android; 22 using namespace android::renderscript; 23 24 namespace android { 25 namespace renderscript { 26 27 28 class RsdCpuScriptIntrinsic3DLUT : public RsdCpuScriptIntrinsic { 29 public: 30 virtual void populateScript(Script *); 31 virtual void invokeFreeChildren(); 32 33 virtual void setGlobalObj(uint32_t slot, ObjectBase *data); 34 35 virtual ~RsdCpuScriptIntrinsic3DLUT(); 36 RsdCpuScriptIntrinsic3DLUT(RsdCpuReferenceImpl *ctx, const Script *s, const Element *e); 37 38 protected: 39 ObjectBaseRef<Allocation> mLUT; 40 41 static void kernel(const RsForEachStubParamStruct *p, 42 uint32_t xstart, uint32_t xend, 43 uint32_t instep, uint32_t outstep); 44 }; 45 46 } 47 } 48 49 50 void RsdCpuScriptIntrinsic3DLUT::setGlobalObj(uint32_t slot, ObjectBase *data) { 51 rsAssert(slot == 0); 52 mLUT.set(static_cast<Allocation *>(data)); 53 } 54 55 extern "C" void rsdIntrinsic3DLUT_K(void *dst, const void *src, const void *lut, 56 size_t lut_stride_y, size_t lut_stride_z, 57 uint32_t count, const void *constants); 58 59 60 void RsdCpuScriptIntrinsic3DLUT::kernel(const RsForEachStubParamStruct *p, 61 uint32_t xstart, uint32_t xend, 62 uint32_t instep, uint32_t outstep) { 63 RsdCpuScriptIntrinsic3DLUT *cp = (RsdCpuScriptIntrinsic3DLUT *)p->usr; 64 65 uchar4 *out = (uchar4 *)p->out; 66 uchar4 *in = (uchar4 *)p->in; 67 uint32_t x1 = xstart; 68 uint32_t x2 = xend; 69 70 const uchar *bp = (const uchar *)cp->mLUT->mHal.drvState.lod[0].mallocPtr; 71 72 int4 dims = { 73 cp->mLUT->mHal.drvState.lod[0].dimX - 1, 74 cp->mLUT->mHal.drvState.lod[0].dimY - 1, 75 cp->mLUT->mHal.drvState.lod[0].dimZ - 1, 76 -1 77 }; 78 const float4 m = (float4)(1.f / 255.f) * convert_float4(dims); 79 const int4 coordMul = convert_int4(m * (float4)0x8000); 80 const size_t stride_y = cp->mLUT->mHal.drvState.lod[0].stride; 81 const size_t stride_z = stride_y * cp->mLUT->mHal.drvState.lod[0].dimY; 82 83 //ALOGE("strides %zu %zu", stride_y, stride_z); 84 85 while (x1 < x2) { 86 #if defined(ARCH_ARM_HAVE_VFP) 87 if (gArchUseSIMD) { 88 int32_t len = (x2 - x1 - 1) >> 1; 89 if(len > 0) { 90 const short neon_constants[] = { 91 coordMul.x, coordMul.y, coordMul.z, 0, 92 0, 0, 0, 0xffff, 93 94 }; 95 96 rsdIntrinsic3DLUT_K(out, in, bp, stride_y, stride_z, len, neon_constants); 97 x1 += len << 1; 98 out += len << 1; 99 in += len << 1; 100 } 101 } 102 103 #endif 104 105 int4 baseCoord = convert_int4(*in) * coordMul; 106 int4 coord1 = baseCoord >> (int4)15; 107 //int4 coord2 = min(coord1 + 1, gDims - 1); 108 109 int4 weight2 = baseCoord & 0x7fff; 110 int4 weight1 = (int4)0x8000 - weight2; 111 112 //ALOGE("coord1 %08x %08x %08x %08x", coord1.x, coord1.y, coord1.z, coord1.w); 113 const uchar *bp2 = bp + (coord1.x * 4) + (coord1.y * stride_y) + (coord1.z * stride_z); 114 const uchar4 *pt_00 = (const uchar4 *)&bp2[0]; 115 const uchar4 *pt_10 = (const uchar4 *)&bp2[stride_y]; 116 const uchar4 *pt_01 = (const uchar4 *)&bp2[stride_z]; 117 const uchar4 *pt_11 = (const uchar4 *)&bp2[stride_y + stride_z]; 118 119 uint4 v000 = convert_uint4(pt_00[0]); 120 uint4 v100 = convert_uint4(pt_00[1]); 121 uint4 v010 = convert_uint4(pt_10[0]); 122 uint4 v110 = convert_uint4(pt_10[1]); 123 uint4 v001 = convert_uint4(pt_01[0]); 124 uint4 v101 = convert_uint4(pt_01[1]); 125 uint4 v011 = convert_uint4(pt_11[0]); 126 uint4 v111 = convert_uint4(pt_11[1]); 127 128 uint4 yz00 = ((v000 * weight1.x) + (v100 * weight2.x)) >> (int4)7; 129 uint4 yz10 = ((v010 * weight1.x) + (v110 * weight2.x)) >> (int4)7; 130 uint4 yz01 = ((v001 * weight1.x) + (v101 * weight2.x)) >> (int4)7; 131 uint4 yz11 = ((v011 * weight1.x) + (v111 * weight2.x)) >> (int4)7; 132 133 uint4 z0 = ((yz00 * weight1.y) + (yz10 * weight2.y)) >> (int4)15; 134 uint4 z1 = ((yz01 * weight1.y) + (yz11 * weight2.y)) >> (int4)15; 135 136 uint4 v = ((z0 * weight1.z) + (z1 * weight2.z)) >> (int4)15; 137 uint4 v2 = (v + 0x7f) >> (int4)8; 138 139 uchar4 ret = convert_uchar4(v2); 140 ret.w = in->w; 141 142 #if 0 143 if (!x1) { 144 ALOGE("in %08x %08x %08x %08x", in->r, in->g, in->b, in->a); 145 ALOGE("baseCoord %08x %08x %08x %08x", baseCoord.x, baseCoord.y, baseCoord.z, baseCoord.w); 146 ALOGE("coord1 %08x %08x %08x %08x", coord1.x, coord1.y, coord1.z, coord1.w); 147 ALOGE("weight1 %08x %08x %08x %08x", weight1.x, weight1.y, weight1.z, weight1.w); 148 ALOGE("weight2 %08x %08x %08x %08x", weight2.x, weight2.y, weight2.z, weight2.w); 149 150 ALOGE("v000 %08x %08x %08x %08x", v000.x, v000.y, v000.z, v000.w); 151 ALOGE("v100 %08x %08x %08x %08x", v100.x, v100.y, v100.z, v100.w); 152 ALOGE("yz00 %08x %08x %08x %08x", yz00.x, yz00.y, yz00.z, yz00.w); 153 ALOGE("z0 %08x %08x %08x %08x", z0.x, z0.y, z0.z, z0.w); 154 155 ALOGE("v %08x %08x %08x %08x", v.x, v.y, v.z, v.w); 156 ALOGE("v2 %08x %08x %08x %08x", v2.x, v2.y, v2.z, v2.w); 157 } 158 #endif 159 *out = ret; 160 161 162 in++; 163 out++; 164 x1++; 165 } 166 } 167 168 RsdCpuScriptIntrinsic3DLUT::RsdCpuScriptIntrinsic3DLUT(RsdCpuReferenceImpl *ctx, 169 const Script *s, const Element *e) 170 : RsdCpuScriptIntrinsic(ctx, s, e, RS_SCRIPT_INTRINSIC_ID_3DLUT) { 171 172 mRootPtr = &kernel; 173 } 174 175 RsdCpuScriptIntrinsic3DLUT::~RsdCpuScriptIntrinsic3DLUT() { 176 } 177 178 void RsdCpuScriptIntrinsic3DLUT::populateScript(Script *s) { 179 s->mHal.info.exportedVariableCount = 1; 180 } 181 182 void RsdCpuScriptIntrinsic3DLUT::invokeFreeChildren() { 183 mLUT.clear(); 184 } 185 186 187 RsdCpuScriptImpl * rsdIntrinsic_3DLUT(RsdCpuReferenceImpl *ctx, 188 const Script *s, const Element *e) { 189 190 return new RsdCpuScriptIntrinsic3DLUT(ctx, s, e); 191 } 192 193 194
