src/panfrost/midgard/helpers.h

   1 /* Copyright (c) 2018-2019 Alyssa Rosenzweig (alyssa@rosenzweig.io)
   2  *
   3  * Permission is hereby granted, free of charge, to any person obtaining a copy
   4  * of this software and associated documentation files (the "Software"), to deal
   5  * in the Software without restriction, including without limitation the rights
   6  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
   7  * copies of the Software, and to permit persons to whom the Software is
   8  * furnished to do so, subject to the following conditions:
   9  *
  10  * The above copyright notice and this permission notice shall be included in
  11  * all copies or substantial portions of the Software.
  12  *
  13  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  14  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  15  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  16  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  17  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  18  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  19  * THE SOFTWARE.
  20  */
  21
  22 #ifndef __MDG_HELPERS_H
  23 #define __MDG_HELPERS_H
  24
  25 #include "util/macros.h"
  26 #include <stdio.h>
  27 #include <string.h>
  28
  29 #define OP_IS_LOAD_VARY_F(op) (\
  30                 op == midgard_op_ld_vary_16 || \
  31                 op == midgard_op_ld_vary_32 \
  32         )
  33
  34 #define OP_IS_PROJECTION(op) ( \
  35                 op == midgard_op_ldst_perspective_division_z || \
  36                 op == midgard_op_ldst_perspective_division_w \
  37         )
  38
  39 #define OP_IS_VEC4_ONLY(op) ( \
  40                 OP_IS_PROJECTION(op) || \
  41                 op == midgard_op_ld_cubemap_coords \
  42         )
  43
  44 #define OP_IS_MOVE(op) ( \
  45                 op == midgard_alu_op_fmov || \
  46                 op == midgard_alu_op_imov \
  47         )
  48
  49 #define OP_IS_UBO_READ(op) ( \
  50                 op == midgard_op_ld_ubo_char  || \
  51                 op == midgard_op_ld_ubo_char2  || \
  52                 op == midgard_op_ld_ubo_char4  || \
  53                 op == midgard_op_ld_ubo_short4  || \
  54                 op == midgard_op_ld_ubo_int4 \
  55         )
  56
  57 #define OP_IS_CSEL_V(op) ( \
  58                 op == midgard_alu_op_icsel_v || \
  59                 op == midgard_alu_op_fcsel_v \
  60         )
  61
  62 #define OP_IS_CSEL(op) ( \
  63                 OP_IS_CSEL_V(op) || \
  64                 op == midgard_alu_op_icsel || \
  65                 op == midgard_alu_op_fcsel \
  66         )
  67
  68 #define OP_IS_UNSIGNED_CMP(op) ( \
  69                 op == midgard_alu_op_ult || \
  70                 op == midgard_alu_op_ule \
  71         )
  72
  73 #define OP_IS_INTEGER_CMP(op) ( \
  74                 op == midgard_alu_op_ieq || \
  75                 op == midgard_alu_op_ine || \
  76                 op == midgard_alu_op_ilt || \
  77                 op == midgard_alu_op_ile || \
  78                 OP_IS_UNSIGNED_CMP(op) \
  79         )
  80
  81 /* ALU control words are single bit fields with a lot of space */
  82
  83 #define ALU_ENAB_VEC_MUL  (1 << 17)
  84 #define ALU_ENAB_SCAL_ADD  (1 << 19)
  85 #define ALU_ENAB_VEC_ADD  (1 << 21)
  86 #define ALU_ENAB_SCAL_MUL  (1 << 23)
  87 #define ALU_ENAB_VEC_LUT  (1 << 25)
  88 #define ALU_ENAB_BR_COMPACT (1 << 26)
  89 #define ALU_ENAB_BRANCH   (1 << 27)
  90
  91 /* Other opcode properties that don't conflict with the ALU_ENABs, non-ISA */
  92
  93 /* Denotes an opcode that takes a vector input with a fixed-number of
  94  * channels, but outputs to only a single output channel, like dot products.
  95  * For these, to determine the effective mask, this quirk can be set. We have
  96  * an intentional off-by-one (a la MALI_POSITIVE), since 0-channel makes no
  97  * sense but we need to fit 4 channels in 2-bits. Similarly, 1-channel doesn't
  98  * make sense (since then why are we quirked?), so that corresponds to "no
  99  * count set" */
 100
 101 #define OP_CHANNEL_COUNT(c) ((c - 1) << 0)
 102 #define GET_CHANNEL_COUNT(c) ((c & (0x3 << 0)) ? ((c & (0x3 << 0)) + 1) : 0)
 103
 104 /* For instructions that take a single argument, normally the first argument
 105  * slot is used for the argument and the second slot is a dummy #0 constant.
 106  * However, there are exceptions: instructions like fmov store their argument
 107  * in the _second_ slot and store a dummy r24 in the first slot, designated by
 108  * QUIRK_FLIPPED_R24 */
 109
 110 #define QUIRK_FLIPPED_R24 (1 << 2)
 111
 112 /* Is the op commutative? */
 113 #define OP_COMMUTES (1 << 3)
 114
 115 /* Does the op convert types between int- and float- space (i2f/f2u/etc) */
 116 #define OP_TYPE_CONVERT (1 << 4)
 117
 118 /* Is this opcode the first in a f2x (rte, rtz, rtn, rtp) sequence? If so,
 119  * takes a roundmode argument in the IR. This has the semantic of rounding the
 120  * source (it's all fused in), which is why it doesn't necessarily make sense
 121  * for i2f (though folding there might be necessary for OpenCL reasons). Comes
 122  * up in format conversion, i.e. f2u_rte */
 123 #define MIDGARD_ROUNDS (1 << 5)
 124
 125 /* Vector-independant shorthands for the above; these numbers are arbitrary and
 126  * not from the ISA. Convert to the above with unit_enum_to_midgard */
 127
 128 #define UNIT_MUL 0
 129 #define UNIT_ADD 1
 130 #define UNIT_LUT 2
 131
 132 #define IS_ALU(tag) (tag >= TAG_ALU_4)
 133
 134 /* Special register aliases */
 135
 136 #define MAX_WORK_REGISTERS 16
 137
 138 /* Uniforms are begin at (REGISTER_UNIFORMS - uniform_count) */
 139 #define REGISTER_UNIFORMS 24
 140
 141 /* r24 and r25 are special registers that only exist during the pipeline,
 142  * by using them when we don't care about the register we skip a roundtrip
 143  * to the register file. */
 144 #define REGISTER_UNUSED 24
 145 #define REGISTER_CONSTANT 26
 146 #define REGISTER_LDST_BASE 26
 147 #define REGISTER_TEXTURE_BASE 28
 148 #define REGISTER_SELECT 31
 149
 150 /* SSA helper aliases to mimic the registers. */
 151
 152 #define SSA_FIXED_SHIFT 24
 153 #define SSA_FIXED_REGISTER(reg) (((1 + (reg)) << SSA_FIXED_SHIFT) | 1)
 154 #define SSA_REG_FROM_FIXED(reg) ((((reg) & ~1) >> SSA_FIXED_SHIFT) - 1)
 155 #define SSA_FIXED_MINIMUM SSA_FIXED_REGISTER(0)
 156
 157 #define COMPONENT_X 0x0
 158 #define COMPONENT_Y 0x1
 159 #define COMPONENT_Z 0x2
 160 #define COMPONENT_W 0x3
 161
 162 #define SWIZZLE_IDENTITY { \
 163         { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 }, \
 164         { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 }, \
 165         { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 }, \
 166         { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 } \
 167 }
 168
 169 #define SWIZZLE_IDENTITY_4 { \
 170         { 0, 1, 2, 3, 0, 0, 0, 0, 0, 0,  0,  0,  0,  0,  0,  0 }, \
 171         { 0, 1, 2, 3, 0, 0, 0, 0, 0, 0,  0,  0,  0,  0,  0,  0 }, \
 172         { 0, 1, 2, 3, 0, 0, 0, 0, 0, 0,  0,  0,  0,  0,  0,  0 }, \
 173         { 0, 1, 2, 3, 0, 0, 0, 0, 0, 0,  0,  0,  0,  0,  0,  0 }, \
 174 }
 175
 176 static inline unsigned
 177 mask_of(unsigned nr_comp)
 178 {
 179         return (1 << nr_comp) - 1;
 180 }
 181
 182 /* See ISA notes */
 183
 184 #define LDST_NOP (3)
 185
 186 /* There are five ALU units: VMUL, VADD, SMUL, SADD, LUT. A given opcode is
 187  * implemented on some subset of these units (or occassionally all of them).
 188  * This table encodes a bit mask of valid units for each opcode, so the
 189  * scheduler can figure where to plonk the instruction. */
 190
 191 /* Shorthands for each unit */
 192 #define UNIT_VMUL ALU_ENAB_VEC_MUL
 193 #define UNIT_SADD ALU_ENAB_SCAL_ADD
 194 #define UNIT_VADD ALU_ENAB_VEC_ADD
 195 #define UNIT_SMUL ALU_ENAB_SCAL_MUL
 196 #define UNIT_VLUT ALU_ENAB_VEC_LUT
 197
 198 /* Shorthands for usual combinations of units */
 199
 200 #define UNITS_MUL (UNIT_VMUL | UNIT_SMUL)
 201 #define UNITS_ADD (UNIT_VADD | UNIT_SADD)
 202 #define UNITS_MOST (UNITS_MUL | UNITS_ADD)
 203 #define UNITS_ALL (UNITS_MOST | UNIT_VLUT)
 204 #define UNITS_SCALAR (UNIT_SADD | UNIT_SMUL)
 205 #define UNITS_VECTOR (UNIT_VMUL | UNIT_VADD)
 206 #define UNITS_ANY_VECTOR (UNITS_VECTOR | UNIT_VLUT)
 207
 208 struct mir_op_props {
 209         const char *name;
 210         unsigned props;
 211 };
 212
 213 /* For load/store */
 214
 215 struct mir_ldst_op_props {
 216         const char *name;
 217         unsigned props;
 218 };
 219
 220 struct mir_tag_props {
 221         const char *name;
 222         unsigned size;
 223 };
 224
 225 /* Lower 2-bits are a midgard_reg_mode */
 226 #define GET_LDST_SIZE(c) (c & 3)
 227
 228 /* Store (so the primary register is a source, not a destination */
 229 #define LDST_STORE (1 << 2)
 230
 231 /* Mask has special meaning and should not be manipulated directly */
 232 #define LDST_SPECIAL_MASK (1 << 3)
 233
 234 /* Non-store operation has side effects and should not be eliminated even if
 235  * its mask is 0 */
 236 #define LDST_SIDE_FX (1 << 4)
 237
 238 /* Computes an address according to indirects/zext/shift/etc */
 239 #define LDST_ADDRESS (1 << 5)
 240
 241 /* This file is common, so don't define the tables themselves. #include
 242  * midgard_op.h if you need that, or edit midgard_ops.c directly */
 243
 244 /* Duplicate bits to convert a per-component to duplicated 8-bit format,
 245  * which is used for vector units */
 246
 247 static inline unsigned
 248 expand_writemask(unsigned mask, unsigned log2_channels)
 249 {
 250         unsigned o = 0;
 251         unsigned factor = 8 >> log2_channels;
 252         unsigned expanded = (1 << factor) - 1;
 253
 254         for (unsigned i = 0; i < (1 << log2_channels); ++i)
 255                 if (mask & (1 << i))
 256                         o |= (expanded << (factor * i));
 257
 258         return o;
 259 }
 260
 261 /* Coerce structs to integer */
 262
 263 static inline unsigned
 264 vector_alu_srco_unsigned(midgard_vector_alu_src src)
 265 {
 266         unsigned u;
 267         memcpy(&u, &src, sizeof(src));
 268         return u;
 269 }
 270
 271 static inline midgard_vector_alu_src
 272 vector_alu_from_unsigned(unsigned u)
 273 {
 274         midgard_vector_alu_src s;
 275         memcpy(&s, &u, sizeof(s));
 276         return s;
 277 }
 278
 279 static inline void
 280 mir_compose_swizzle(unsigned *left, unsigned *right, unsigned *final_out)
 281 {
 282         unsigned out[16];
 283
 284         for (unsigned c = 0; c < 16; ++c)
 285                 out[c] = right[left[c]];
 286
 287         memcpy(final_out, out, sizeof(out));
 288 }
 289
 290 /* Checks for an xyzw.. swizzle, given a mask */
 291
 292 static inline bool
 293 mir_is_simple_swizzle(unsigned *swizzle, unsigned mask)
 294 {
 295         for (unsigned i = 0; i < 16; ++i) {
 296                 if (!(mask & (1 << i))) continue;
 297
 298                 if (swizzle[i] != i)
 299                         return false;
 300         }
 301
 302         return true;
 303 }
 304
 305 /* Packs a load/store argument */
 306
 307 static inline uint8_t
 308 midgard_ldst_reg(unsigned reg, unsigned component, unsigned size)
 309 {
 310         assert((reg == REGISTER_LDST_BASE) || (reg == REGISTER_LDST_BASE + 1));
 311         assert(size == 16 || size == 32 || size == 64);
 312
 313         /* Shift so everything is in terms of 32-bit units */
 314         if (size == 64) {
 315                 assert(component < 2);
 316                 component <<= 1;
 317         } else if (size == 16) {
 318                 assert((component & 1) == 0);
 319                 component >>= 1;
 320         }
 321
 322         midgard_ldst_register_select sel = {
 323                 .component = component,
 324                 .select = reg - 26
 325         };
 326
 327         uint8_t packed;
 328         memcpy(&packed, &sel, sizeof(packed));
 329
 330         return packed;
 331 }
 332
 333 static inline bool
 334 midgard_is_branch_unit(unsigned unit)
 335 {
 336         return (unit == ALU_ENAB_BRANCH) || (unit == ALU_ENAB_BR_COMPACT);
 337 }
 338
 339 /* Packs ALU mod argument */
 340 struct midgard_instruction;
 341 unsigned mir_pack_mod(struct midgard_instruction *ins, unsigned i, bool scalar);
 342
 343 void
 344 mir_print_constant_component(FILE *fp, const midgard_constants *consts,
 345                              unsigned c, midgard_reg_mode reg_mode, bool half,
 346                              unsigned mod, midgard_alu_op op);
 347
 348 #endif