src/gallium/auxiliary/gallivm/lp_bld_intr.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 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
  29 /**
  30  * @file
  31  * Helpers for emiting intrinsic calls.
  32  *
  33  * LLVM vanilla IR doesn't represent all basic arithmetic operations we care
  34  * about, and it is often necessary to resort target-specific intrinsics for
  35  * performance, convenience.
  36  *
  37  * Ideally we would like to stay away from target specific intrinsics and
  38  * move all the instruction selection logic into upstream LLVM where it belongs.
  39  *
  40  * These functions are also used for calling C functions provided by us from
  41  * generated LLVM code.
  42  *
  43  * @author Jose Fonseca <jfonseca@vmware.com>
  44  */
  45
  46
  47 #include "util/u_debug.h"
  48 #include "util/u_string.h"
  49 #include "util/bitscan.h"
  50
  51 #include "lp_bld_const.h"
  52 #include "lp_bld_intr.h"
  53 #include "lp_bld_type.h"
  54 #include "lp_bld_pack.h"
  55 #include "lp_bld_debug.h"
  56
  57
  58 void
  59 lp_format_intrinsic(char *name,
  60                     size_t size,
  61                     const char *name_root,
  62                     LLVMTypeRef type)
  63 {
  64    unsigned length = 0;
  65    unsigned width;
  66    char c;
  67
  68    LLVMTypeKind kind = LLVMGetTypeKind(type);
  69    if (kind == LLVMVectorTypeKind) {
  70       length = LLVMGetVectorSize(type);
  71       type = LLVMGetElementType(type);
  72       kind = LLVMGetTypeKind(type);
  73    }
  74
  75    switch (kind) {
  76    case LLVMIntegerTypeKind:
  77       c = 'i';
  78       width = LLVMGetIntTypeWidth(type);
  79       break;
  80    case LLVMFloatTypeKind:
  81       c = 'f';
  82       width = 32;
  83       break;
  84    case LLVMDoubleTypeKind:
  85       c = 'f';
  86       width = 64;
  87       break;
  88    default:
  89       unreachable("unexpected LLVMTypeKind");
  90    }
  91
  92    if (length) {
  93       snprintf(name, size, "%s.v%u%c%u", name_root, length, c, width);
  94    } else {
  95       snprintf(name, size, "%s.%c%u", name_root, c, width);
  96    }
  97 }
  98
  99
 100 LLVMValueRef
 101 lp_declare_intrinsic(LLVMModuleRef module,
 102                      const char *name,
 103                      LLVMTypeRef ret_type,
 104                      LLVMTypeRef *arg_types,
 105                      unsigned num_args)
 106 {
 107    LLVMTypeRef function_type;
 108    LLVMValueRef function;
 109
 110    assert(!LLVMGetNamedFunction(module, name));
 111
 112    function_type = LLVMFunctionType(ret_type, arg_types, num_args, 0);
 113    function = LLVMAddFunction(module, name, function_type);
 114
 115    LLVMSetFunctionCallConv(function, LLVMCCallConv);
 116    LLVMSetLinkage(function, LLVMExternalLinkage);
 117
 118    assert(LLVMIsDeclaration(function));
 119
 120    return function;
 121 }
 122
 123
 124 #if HAVE_LLVM < 0x0400
 125 static LLVMAttribute lp_attr_to_llvm_attr(enum lp_func_attr attr)
 126 {
 127    switch (attr) {
 128    case LP_FUNC_ATTR_ALWAYSINLINE: return LLVMAlwaysInlineAttribute;
 129    case LP_FUNC_ATTR_INREG: return LLVMInRegAttribute;
 130    case LP_FUNC_ATTR_NOALIAS: return LLVMNoAliasAttribute;
 131    case LP_FUNC_ATTR_NOUNWIND: return LLVMNoUnwindAttribute;
 132    case LP_FUNC_ATTR_READNONE: return LLVMReadNoneAttribute;
 133    case LP_FUNC_ATTR_READONLY: return LLVMReadOnlyAttribute;
 134    default:
 135       _debug_printf("Unhandled function attribute: %x\n", attr);
 136       return 0;
 137    }
 138 }
 139
 140 #else
 141
 142 static const char *attr_to_str(enum lp_func_attr attr)
 143 {
 144    switch (attr) {
 145    case LP_FUNC_ATTR_ALWAYSINLINE: return "alwaysinline";
 146    case LP_FUNC_ATTR_INREG: return "inreg";
 147    case LP_FUNC_ATTR_NOALIAS: return "noalias";
 148    case LP_FUNC_ATTR_NOUNWIND: return "nounwind";
 149    case LP_FUNC_ATTR_READNONE: return "readnone";
 150    case LP_FUNC_ATTR_READONLY: return "readonly";
 151    case LP_FUNC_ATTR_WRITEONLY: return "writeonly";
 152    case LP_FUNC_ATTR_INACCESSIBLE_MEM_ONLY: return "inaccessiblememonly";
 153    case LP_FUNC_ATTR_CONVERGENT: return "convergent";
 154    default:
 155       _debug_printf("Unhandled function attribute: %x\n", attr);
 156       return 0;
 157    }
 158 }
 159
 160 #endif
 161
 162 void
 163 lp_add_function_attr(LLVMValueRef function_or_call,
 164                      int attr_idx, enum lp_func_attr attr)
 165 {
 166
 167 #if HAVE_LLVM < 0x0400
 168    LLVMAttribute llvm_attr = lp_attr_to_llvm_attr(attr);
 169    if (LLVMIsAFunction(function_or_call)) {
 170       if (attr_idx == -1) {
 171          LLVMAddFunctionAttr(function_or_call, llvm_attr);
 172       } else {
 173          LLVMAddAttribute(LLVMGetParam(function_or_call, attr_idx - 1), llvm_attr);
 174       }
 175    } else {
 176       LLVMAddInstrAttribute(function_or_call, attr_idx, llvm_attr);
 177    }
 178 #else
 179
 180    LLVMModuleRef module;
 181    if (LLVMIsAFunction(function_or_call)) {
 182       module = LLVMGetGlobalParent(function_or_call);
 183    } else {
 184       LLVMBasicBlockRef bb = LLVMGetInstructionParent(function_or_call);
 185       LLVMValueRef function = LLVMGetBasicBlockParent(bb);
 186       module = LLVMGetGlobalParent(function);
 187    }
 188    LLVMContextRef ctx = LLVMGetModuleContext(module);
 189
 190    const char *attr_name = attr_to_str(attr);
 191    unsigned kind_id = LLVMGetEnumAttributeKindForName(attr_name,
 192                                                       strlen(attr_name));
 193    LLVMAttributeRef llvm_attr = LLVMCreateEnumAttribute(ctx, kind_id, 0);
 194
 195    if (LLVMIsAFunction(function_or_call))
 196       LLVMAddAttributeAtIndex(function_or_call, attr_idx, llvm_attr);
 197    else
 198       LLVMAddCallSiteAttribute(function_or_call, attr_idx, llvm_attr);
 199 #endif
 200 }
 201
 202 static void
 203 lp_add_func_attributes(LLVMValueRef function, unsigned attrib_mask)
 204 {
 205    /* NoUnwind indicates that the intrinsic never raises a C++ exception.
 206     * Set it for all intrinsics.
 207     */
 208    attrib_mask |= LP_FUNC_ATTR_NOUNWIND;
 209    attrib_mask &= ~LP_FUNC_ATTR_LEGACY;
 210
 211    while (attrib_mask) {
 212       enum lp_func_attr attr = 1u << u_bit_scan(&attrib_mask);
 213       lp_add_function_attr(function, -1, attr);
 214    }
 215 }
 216
 217 LLVMValueRef
 218 lp_build_intrinsic(LLVMBuilderRef builder,
 219                    const char *name,
 220                    LLVMTypeRef ret_type,
 221                    LLVMValueRef *args,
 222                    unsigned num_args,
 223                    unsigned attr_mask)
 224 {
 225    LLVMModuleRef module = LLVMGetGlobalParent(LLVMGetBasicBlockParent(LLVMGetInsertBlock(builder)));
 226    LLVMValueRef function, call;
 227    bool set_callsite_attrs = HAVE_LLVM >= 0x0400 &&
 228                              !(attr_mask & LP_FUNC_ATTR_LEGACY);
 229
 230    function = LLVMGetNamedFunction(module, name);
 231    if(!function) {
 232       LLVMTypeRef arg_types[LP_MAX_FUNC_ARGS];
 233       unsigned i;
 234
 235       assert(num_args <= LP_MAX_FUNC_ARGS);
 236
 237       for(i = 0; i < num_args; ++i) {
 238          assert(args[i]);
 239          arg_types[i] = LLVMTypeOf(args[i]);
 240       }
 241
 242       function = lp_declare_intrinsic(module, name, ret_type, arg_types, num_args);
 243
 244       /*
 245        * If llvm removes an intrinsic we use, we'll hit this abort (rather
 246        * than a call to address zero in the jited code).
 247        */
 248       if (LLVMGetIntrinsicID(function) == 0) {
 249          _debug_printf("llvm (version 0x%x) found no intrinsic for %s, going to crash...\n",
 250                 HAVE_LLVM, name);
 251          abort();
 252       }
 253
 254       if (!set_callsite_attrs)
 255          lp_add_func_attributes(function, attr_mask);
 256
 257       if (gallivm_debug & GALLIVM_DEBUG_IR) {
 258          lp_debug_dump_value(function);
 259       }
 260    }
 261
 262    call = LLVMBuildCall(builder, function, args, num_args, "");
 263    if (set_callsite_attrs)
 264       lp_add_func_attributes(call, attr_mask);
 265    return call;
 266 }
 267
 268
 269 LLVMValueRef
 270 lp_build_intrinsic_unary(LLVMBuilderRef builder,
 271                          const char *name,
 272                          LLVMTypeRef ret_type,
 273                          LLVMValueRef a)
 274 {
 275    return lp_build_intrinsic(builder, name, ret_type, &a, 1, 0);
 276 }
 277
 278
 279 LLVMValueRef
 280 lp_build_intrinsic_binary(LLVMBuilderRef builder,
 281                           const char *name,
 282                           LLVMTypeRef ret_type,
 283                           LLVMValueRef a,
 284                           LLVMValueRef b)
 285 {
 286    LLVMValueRef args[2];
 287
 288    args[0] = a;
 289    args[1] = b;
 290
 291    return lp_build_intrinsic(builder, name, ret_type, args, 2, 0);
 292 }
 293
 294
 295 /**
 296  * Call intrinsic with arguments adapted to intrinsic vector length.
 297  *
 298  * Split vectors which are too large for the hw, or expand them if they
 299  * are too small, so a caller calling a function which might use intrinsics
 300  * doesn't need to do splitting/expansion on its own.
 301  * This only supports intrinsics where src and dst types match.
 302  */
 303 LLVMValueRef
 304 lp_build_intrinsic_binary_anylength(struct gallivm_state *gallivm,
 305                                     const char *name,
 306                                     struct lp_type src_type,
 307                                     unsigned intr_size,
 308                                     LLVMValueRef a,
 309                                     LLVMValueRef b)
 310 {
 311    unsigned i;
 312    struct lp_type intrin_type = src_type;
 313    LLVMBuilderRef builder = gallivm->builder;
 314    LLVMValueRef i32undef = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context));
 315    LLVMValueRef anative, bnative;
 316    unsigned intrin_length = intr_size / src_type.width;
 317
 318    intrin_type.length = intrin_length;
 319
 320    if (intrin_length > src_type.length) {
 321       LLVMValueRef elems[LP_MAX_VECTOR_LENGTH];
 322       LLVMValueRef constvec, tmp;
 323
 324       for (i = 0; i < src_type.length; i++) {
 325          elems[i] = lp_build_const_int32(gallivm, i);
 326       }
 327       for (; i < intrin_length; i++) {
 328          elems[i] = i32undef;
 329       }
 330       if (src_type.length == 1) {
 331          LLVMTypeRef elem_type = lp_build_elem_type(gallivm, intrin_type);
 332          a = LLVMBuildBitCast(builder, a, LLVMVectorType(elem_type, 1), "");
 333          b = LLVMBuildBitCast(builder, b, LLVMVectorType(elem_type, 1), "");
 334       }
 335       constvec = LLVMConstVector(elems, intrin_length);
 336       anative = LLVMBuildShuffleVector(builder, a, a, constvec, "");
 337       bnative = LLVMBuildShuffleVector(builder, b, b, constvec, "");
 338       tmp = lp_build_intrinsic_binary(builder, name,
 339                                       lp_build_vec_type(gallivm, intrin_type),
 340                                       anative, bnative);
 341       if (src_type.length > 1) {
 342          constvec = LLVMConstVector(elems, src_type.length);
 343          return LLVMBuildShuffleVector(builder, tmp, tmp, constvec, "");
 344       }
 345       else {
 346          return LLVMBuildExtractElement(builder, tmp, elems[0], "");
 347       }
 348    }
 349    else if (intrin_length < src_type.length) {
 350       unsigned num_vec = src_type.length / intrin_length;
 351       LLVMValueRef tmp[LP_MAX_VECTOR_LENGTH];
 352
 353       /* don't support arbitrary size here as this is so yuck */
 354       if (src_type.length % intrin_length) {
 355          /* FIXME: This is something which should be supported
 356           * but there doesn't seem to be any need for it currently
 357           * so crash and burn.
 358           */
 359          debug_printf("%s: should handle arbitrary vector size\n",
 360                       __FUNCTION__);
 361          assert(0);
 362          return NULL;
 363       }
 364
 365       for (i = 0; i < num_vec; i++) {
 366          anative = lp_build_extract_range(gallivm, a, i*intrin_length,
 367                                         intrin_length);
 368          bnative = lp_build_extract_range(gallivm, b, i*intrin_length,
 369                                         intrin_length);
 370          tmp[i] = lp_build_intrinsic_binary(builder, name,
 371                                             lp_build_vec_type(gallivm, intrin_type),
 372                                             anative, bnative);
 373       }
 374       return lp_build_concat(gallivm, tmp, intrin_type, num_vec);
 375    }
 376    else {
 377       return lp_build_intrinsic_binary(builder, name,
 378                                        lp_build_vec_type(gallivm, src_type),
 379                                        a, b);
 380    }
 381 }
 382
 383
 384 LLVMValueRef
 385 lp_build_intrinsic_map(struct gallivm_state *gallivm,
 386                        const char *name,
 387                        LLVMTypeRef ret_type,
 388                        LLVMValueRef *args,
 389                        unsigned num_args)
 390 {
 391    LLVMBuilderRef builder = gallivm->builder;
 392    LLVMTypeRef ret_elem_type = LLVMGetElementType(ret_type);
 393    unsigned n = LLVMGetVectorSize(ret_type);
 394    unsigned i, j;
 395    LLVMValueRef res;
 396
 397    assert(num_args <= LP_MAX_FUNC_ARGS);
 398
 399    res = LLVMGetUndef(ret_type);
 400    for(i = 0; i < n; ++i) {
 401       LLVMValueRef index = lp_build_const_int32(gallivm, i);
 402       LLVMValueRef arg_elems[LP_MAX_FUNC_ARGS];
 403       LLVMValueRef res_elem;
 404       for(j = 0; j < num_args; ++j)
 405          arg_elems[j] = LLVMBuildExtractElement(builder, args[j], index, "");
 406       res_elem = lp_build_intrinsic(builder, name, ret_elem_type, arg_elems, num_args, 0);
 407       res = LLVMBuildInsertElement(builder, res, res_elem, index, "");
 408    }
 409
 410    return res;
 411 }
 412
 413
 414 LLVMValueRef
 415 lp_build_intrinsic_map_unary(struct gallivm_state *gallivm,
 416                              const char *name,
 417                              LLVMTypeRef ret_type,
 418                              LLVMValueRef a)
 419 {
 420    return lp_build_intrinsic_map(gallivm, name, ret_type, &a, 1);
 421 }
 422
 423
 424 LLVMValueRef
 425 lp_build_intrinsic_map_binary(struct gallivm_state *gallivm,
 426                               const char *name,
 427                               LLVMTypeRef ret_type,
 428                               LLVMValueRef a,
 429                               LLVMValueRef b)
 430 {
 431    LLVMValueRef args[2];
 432
 433    args[0] = a;
 434    args[1] = b;
 435
 436    return lp_build_intrinsic_map(gallivm, name, ret_type, args, 2);
 437 }
 438
 439