1 /************************************************************************** 2 * 3 * Copyright 2012 VMware, Inc. 4 * All Rights Reserved. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the 8 * "Software"), to deal in the Software without restriction, including 9 * without limitation the rights to use, copy, modify, merge, publish, 10 * distribute, sub license, and/or sell copies of the Software, and to 11 * permit persons to whom the Software is furnished to do so, subject to 12 * the following conditions: 13 * 14 * The above copyright notice and this permission notice (including the 15 * next paragraph) shall be included in all copies or substantial portions 16 * of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR 22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 25 * 26 **************************************************************************/ 27 28 #include "pipe/p_state.h" 29 #include "util/u_debug.h" 30 31 #include "gallivm/lp_bld_type.h" 32 #include "gallivm/lp_bld_arit.h" 33 #include "gallivm/lp_bld_const.h" 34 #include "gallivm/lp_bld_logic.h" 35 #include "gallivm/lp_bld_swizzle.h" 36 #include "gallivm/lp_bld_flow.h" 37 #include "gallivm/lp_bld_debug.h" 38 39 #include "lp_bld_blend.h" 40 41 /** 42 * Is (a OP b) == (b OP a)? 43 */ 44 boolean 45 lp_build_blend_func_commutative(unsigned func) 46 { 47 switch (func) { 48 case PIPE_BLEND_ADD: 49 case PIPE_BLEND_MIN: 50 case PIPE_BLEND_MAX: 51 return TRUE; 52 case PIPE_BLEND_SUBTRACT: 53 case PIPE_BLEND_REVERSE_SUBTRACT: 54 return FALSE; 55 default: 56 assert(0); 57 return TRUE; 58 } 59 } 60 61 62 /** 63 * Whether the blending functions are the reverse of each other. 64 */ 65 boolean 66 lp_build_blend_func_reverse(unsigned rgb_func, unsigned alpha_func) 67 { 68 if(rgb_func == alpha_func) 69 return FALSE; 70 if(rgb_func == PIPE_BLEND_SUBTRACT && alpha_func == PIPE_BLEND_REVERSE_SUBTRACT) 71 return TRUE; 72 if(rgb_func == PIPE_BLEND_REVERSE_SUBTRACT && alpha_func == PIPE_BLEND_SUBTRACT) 73 return TRUE; 74 return FALSE; 75 } 76 77 78 /** 79 * Whether the blending factors are complementary of each other. 80 */ 81 static inline boolean 82 lp_build_blend_factor_complementary(unsigned src_factor, unsigned dst_factor) 83 { 84 return dst_factor == (src_factor ^ 0x10); 85 } 86 87 88 /** 89 * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendEquationSeparate.xml 90 */ 91 LLVMValueRef 92 lp_build_blend_func(struct lp_build_context *bld, 93 unsigned func, 94 LLVMValueRef term1, 95 LLVMValueRef term2) 96 { 97 switch (func) { 98 case PIPE_BLEND_ADD: 99 return lp_build_add(bld, term1, term2); 100 case PIPE_BLEND_SUBTRACT: 101 return lp_build_sub(bld, term1, term2); 102 case PIPE_BLEND_REVERSE_SUBTRACT: 103 return lp_build_sub(bld, term2, term1); 104 case PIPE_BLEND_MIN: 105 return lp_build_min(bld, term1, term2); 106 case PIPE_BLEND_MAX: 107 return lp_build_max(bld, term1, term2); 108 default: 109 assert(0); 110 return bld->zero; 111 } 112 } 113 114 115 /** 116 * Performs optimisations and blending independent of SoA/AoS 117 * 118 * @param func the blend function 119 * @param factor_src PIPE_BLENDFACTOR_xxx 120 * @param factor_dst PIPE_BLENDFACTOR_xxx 121 * @param src source rgba 122 * @param dst dest rgba 123 * @param src_factor src factor computed value 124 * @param dst_factor dst factor computed value 125 * @param not_alpha_dependent same factors accross all channels of src/dst 126 * 127 * not_alpha_dependent should be: 128 * SoA: always true as it is only one channel at a time 129 * AoS: rgb_src_factor == alpha_src_factor && rgb_dst_factor == alpha_dst_factor 130 * 131 * Note that pretty much every possible optimisation can only be done on non-unorm targets 132 * due to unorm values not going above 1.0 meaning factorisation can change results. 133 * e.g. (0.9 * 0.9) + (0.9 * 0.9) != 0.9 * (0.9 + 0.9) as result of + is always <= 1. 134 */ 135 LLVMValueRef 136 lp_build_blend(struct lp_build_context *bld, 137 unsigned func, 138 unsigned factor_src, 139 unsigned factor_dst, 140 LLVMValueRef src, 141 LLVMValueRef dst, 142 LLVMValueRef src_factor, 143 LLVMValueRef dst_factor, 144 boolean not_alpha_dependent, 145 boolean optimise_only) 146 { 147 LLVMValueRef result, src_term, dst_term; 148 149 /* If we are not alpha dependent we can mess with the src/dst factors */ 150 if (not_alpha_dependent) { 151 if (lp_build_blend_factor_complementary(factor_src, factor_dst)) { 152 if (func == PIPE_BLEND_ADD) { 153 if (factor_src < factor_dst) { 154 return lp_build_lerp(bld, src_factor, dst, src, 0); 155 } else { 156 return lp_build_lerp(bld, dst_factor, src, dst, 0); 157 } 158 } else if(bld->type.floating && func == PIPE_BLEND_SUBTRACT) { 159 result = lp_build_add(bld, src, dst); 160 161 if (factor_src < factor_dst) { 162 result = lp_build_mul(bld, result, src_factor); 163 return lp_build_sub(bld, result, dst); 164 } else { 165 result = lp_build_mul(bld, result, dst_factor); 166 return lp_build_sub(bld, src, result); 167 } 168 } else if(bld->type.floating && func == PIPE_BLEND_REVERSE_SUBTRACT) { 169 result = lp_build_add(bld, src, dst); 170 171 if (factor_src < factor_dst) { 172 result = lp_build_mul(bld, result, src_factor); 173 return lp_build_sub(bld, dst, result); 174 } else { 175 result = lp_build_mul(bld, result, dst_factor); 176 return lp_build_sub(bld, result, src); 177 } 178 } 179 } 180 181 if (bld->type.floating && factor_src == factor_dst) { 182 if (func == PIPE_BLEND_ADD || 183 func == PIPE_BLEND_SUBTRACT || 184 func == PIPE_BLEND_REVERSE_SUBTRACT) { 185 LLVMValueRef result; 186 result = lp_build_blend_func(bld, func, src, dst); 187 return lp_build_mul(bld, result, src_factor); 188 } 189 } 190 } 191 192 if (optimise_only) 193 return NULL; 194 195 src_term = lp_build_mul(bld, src, src_factor); 196 dst_term = lp_build_mul(bld, dst, dst_factor); 197 return lp_build_blend_func(bld, func, src_term, dst_term); 198 } 199 200 void 201 lp_build_alpha_to_coverage(struct gallivm_state *gallivm, 202 struct lp_type type, 203 struct lp_build_mask_context *mask, 204 LLVMValueRef alpha, 205 boolean do_branch) 206 { 207 struct lp_build_context bld; 208 LLVMValueRef test; 209 LLVMValueRef alpha_ref_value; 210 211 lp_build_context_init(&bld, gallivm, type); 212 213 alpha_ref_value = lp_build_const_vec(gallivm, type, 0.5); 214 215 test = lp_build_cmp(&bld, PIPE_FUNC_GREATER, alpha, alpha_ref_value); 216 217 lp_build_name(test, "alpha_to_coverage"); 218 219 lp_build_mask_update(mask, test); 220 221 if (do_branch) 222 lp_build_mask_check(mask); 223 } 224