/* * Copyright (C) 2005 Ben Skeggs. * * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining * a copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including * without limitation the rights to use, copy, modify, merge, publish, * distribute, sublicense, and/or sell copies of the Software, and to * permit persons to whom the Software is furnished to do so, subject to * the following conditions: * * The above copyright notice and this permission notice (including the * next paragraph) shall be included in all copies or substantial * portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * */ /** * \file * * Emit the r300_fragment_program_code that can be understood by the hardware. * Input is a pre-transformed radeon_program. * * \author Ben Skeggs * * \author Jerome Glisse * * \todo FogOption * * \todo Verify results of opcodes for accuracy, I've only checked them in * specific cases. */ #include "glheader.h" #include "macros.h" #include "enums.h" #include "shader/prog_instruction.h" #include "shader/prog_parameter.h" #include "shader/prog_print.h" #include "r300_context.h" #include "r300_fragprog.h" #include "r300_reg.h" #include "r300_state.h" /* Mapping Mesa registers to R300 temporaries */ struct reg_acc { int reg; /* Assigned hw temp */ unsigned int refcount; /* Number of uses by mesa program */ }; /** * Describe the current lifetime information for an R300 temporary */ struct reg_lifetime { /* Index of the first slot where this register is free in the sense that it can be used as a new destination register. This is -1 if the register has been assigned to a Mesa register and the last access to the register has not yet been emitted */ int free; /* Index of the first slot where this register is currently reserved. This is used to stop e.g. a scalar operation from being moved before the allocation time of a register that was first allocated for a vector operation. */ int reserved; /* Index of the first slot in which the register can be used as a source without losing the value that is written by the last emitted instruction that writes to the register */ int vector_valid; int scalar_valid; /* Index to the slot where the register was last read. This is also the first slot in which the register may be written again */ int vector_lastread; int scalar_lastread; }; /** * Store usage information about an ALU instruction slot during the * compilation of a fragment program. */ #define SLOT_SRC_VECTOR (1<<0) #define SLOT_SRC_SCALAR (1<<3) #define SLOT_SRC_BOTH (SLOT_SRC_VECTOR | SLOT_SRC_SCALAR) #define SLOT_OP_VECTOR (1<<16) #define SLOT_OP_SCALAR (1<<17) #define SLOT_OP_BOTH (SLOT_OP_VECTOR | SLOT_OP_SCALAR) struct r300_pfs_compile_slot { /* Bitmask indicating which parts of the slot are used, using SLOT_ constants defined above */ unsigned int used; /* Selected sources */ int vsrc[3]; int ssrc[3]; }; /** * Store information during compilation of fragment programs. */ struct r300_pfs_compile_state { struct r300_fragment_program_compiler *compiler; int nrslots; /* number of ALU slots used so far */ /* Track which (parts of) slots are already filled with instructions */ struct r300_pfs_compile_slot slot[PFS_MAX_ALU_INST]; /* Track the validity of R300 temporaries */ struct reg_lifetime hwtemps[PFS_NUM_TEMP_REGS]; /* Used to map Mesa's inputs/temps onto hardware temps */ int temp_in_use; struct reg_acc temps[PFS_NUM_TEMP_REGS]; struct reg_acc inputs[32]; /* don't actually need 32... */ /* Track usage of hardware temps, for register allocation, * indirection detection, etc. */ GLuint used_in_node; GLuint dest_in_node; }; /* * Usefull macros and values */ #define ERROR(fmt, args...) do { \ fprintf(stderr, "%s::%s(): " fmt "\n", \ __FILE__, __FUNCTION__, ##args); \ fp->error = GL_TRUE; \ } while(0) #define PFS_INVAL 0xFFFFFFFF #define COMPILE_STATE \ struct r300_fragment_program *fp = cs->compiler->fp; \ struct r300_fragment_program_code *code = cs->compiler->code; \ (void)code; (void)fp #define SWIZZLE_XYZ 0 #define SWIZZLE_XXX 1 #define SWIZZLE_YYY 2 #define SWIZZLE_ZZZ 3 #define SWIZZLE_WWW 4 #define SWIZZLE_YZX 5 #define SWIZZLE_ZXY 6 #define SWIZZLE_WZY 7 #define SWIZZLE_111 8 #define SWIZZLE_000 9 #define SWIZZLE_HHH 10 #define swizzle(r, x, y, z, w) do_swizzle(cs, r, \ ((SWIZZLE_##x<<0)| \ (SWIZZLE_##y<<3)| \ (SWIZZLE_##z<<6)| \ (SWIZZLE_##w<<9)), \ 0) #define REG_TYPE_INPUT 0 #define REG_TYPE_OUTPUT 1 #define REG_TYPE_TEMP 2 #define REG_TYPE_CONST 3 #define REG_TYPE_SHIFT 0 #define REG_INDEX_SHIFT 2 #define REG_VSWZ_SHIFT 8 #define REG_SSWZ_SHIFT 13 #define REG_NEGV_SHIFT 18 #define REG_NEGS_SHIFT 19 #define REG_ABS_SHIFT 20 #define REG_NO_USE_SHIFT 21 // Hack for refcounting #define REG_VALID_SHIFT 22 // Does the register contain a defined value? #define REG_BUILTIN_SHIFT 23 // Is it a builtin (like all zero/all one)? #define REG_TYPE_MASK (0x03 << REG_TYPE_SHIFT) #define REG_INDEX_MASK (0x3F << REG_INDEX_SHIFT) #define REG_VSWZ_MASK (0x1F << REG_VSWZ_SHIFT) #define REG_SSWZ_MASK (0x1F << REG_SSWZ_SHIFT) #define REG_NEGV_MASK (0x01 << REG_NEGV_SHIFT) #define REG_NEGS_MASK (0x01 << REG_NEGS_SHIFT) #define REG_ABS_MASK (0x01 << REG_ABS_SHIFT) #define REG_NO_USE_MASK (0x01 << REG_NO_USE_SHIFT) #define REG_VALID_MASK (0x01 << REG_VALID_SHIFT) #define REG_BUILTIN_MASK (0x01 << REG_BUILTIN_SHIFT) #define REG(type, index, vswz, sswz, nouse, valid, builtin) \ (((type << REG_TYPE_SHIFT) & REG_TYPE_MASK) | \ ((index << REG_INDEX_SHIFT) & REG_INDEX_MASK) | \ ((nouse << REG_NO_USE_SHIFT) & REG_NO_USE_MASK) | \ ((valid << REG_VALID_SHIFT) & REG_VALID_MASK) | \ ((builtin << REG_BUILTIN_SHIFT) & REG_BUILTIN_MASK) | \ ((vswz << REG_VSWZ_SHIFT) & REG_VSWZ_MASK) | \ ((sswz << REG_SSWZ_SHIFT) & REG_SSWZ_MASK)) #define REG_GET_TYPE(reg) \ ((reg & REG_TYPE_MASK) >> REG_TYPE_SHIFT) #define REG_GET_INDEX(reg) \ ((reg & REG_INDEX_MASK) >> REG_INDEX_SHIFT) #define REG_GET_VSWZ(reg) \ ((reg & REG_VSWZ_MASK) >> REG_VSWZ_SHIFT) #define REG_GET_SSWZ(reg) \ ((reg & REG_SSWZ_MASK) >> REG_SSWZ_SHIFT) #define REG_GET_NO_USE(reg) \ ((reg & REG_NO_USE_MASK) >> REG_NO_USE_SHIFT) #define REG_GET_VALID(reg) \ ((reg & REG_VALID_MASK) >> REG_VALID_SHIFT) #define REG_GET_BUILTIN(reg) \ ((reg & REG_BUILTIN_MASK) >> REG_BUILTIN_SHIFT) #define REG_SET_TYPE(reg, type) \ reg = ((reg & ~REG_TYPE_MASK) | \ ((type << REG_TYPE_SHIFT) & REG_TYPE_MASK)) #define REG_SET_INDEX(reg, index) \ reg = ((reg & ~REG_INDEX_MASK) | \ ((index << REG_INDEX_SHIFT) & REG_INDEX_MASK)) #define REG_SET_VSWZ(reg, vswz) \ reg = ((reg & ~REG_VSWZ_MASK) | \ ((vswz << REG_VSWZ_SHIFT) & REG_VSWZ_MASK)) #define REG_SET_SSWZ(reg, sswz) \ reg = ((reg & ~REG_SSWZ_MASK) | \ ((sswz << REG_SSWZ_SHIFT) & REG_SSWZ_MASK)) #define REG_SET_NO_USE(reg, nouse) \ reg = ((reg & ~REG_NO_USE_MASK) | \ ((nouse << REG_NO_USE_SHIFT) & REG_NO_USE_MASK)) #define REG_SET_VALID(reg, valid) \ reg = ((reg & ~REG_VALID_MASK) | \ ((valid << REG_VALID_SHIFT) & REG_VALID_MASK)) #define REG_SET_BUILTIN(reg, builtin) \ reg = ((reg & ~REG_BUILTIN_MASK) | \ ((builtin << REG_BUILTIN_SHIFT) & REG_BUILTIN_MASK)) #define REG_ABS(reg) \ reg = (reg | REG_ABS_MASK) #define REG_NEGV(reg) \ reg = (reg | REG_NEGV_MASK) #define REG_NEGS(reg) \ reg = (reg | REG_NEGS_MASK) #define NOP_INST0 ( \ (R300_ALU_OUTC_MAD) | \ (R300_ALU_ARGC_ZERO << R300_ALU_ARG0C_SHIFT) | \ (R300_ALU_ARGC_ZERO << R300_ALU_ARG1C_SHIFT) | \ (R300_ALU_ARGC_ZERO << R300_ALU_ARG2C_SHIFT)) #define NOP_INST1 ( \ ((0 | SRC_CONST) << R300_ALU_SRC0C_SHIFT) | \ ((0 | SRC_CONST) << R300_ALU_SRC1C_SHIFT) | \ ((0 | SRC_CONST) << R300_ALU_SRC2C_SHIFT)) #define NOP_INST2 ( \ (R300_ALU_OUTA_MAD) | \ (R300_ALU_ARGA_ZERO << R300_ALU_ARG0A_SHIFT) | \ (R300_ALU_ARGA_ZERO << R300_ALU_ARG1A_SHIFT) | \ (R300_ALU_ARGA_ZERO << R300_ALU_ARG2A_SHIFT)) #define NOP_INST3 ( \ ((0 | SRC_CONST) << R300_ALU_SRC0A_SHIFT) | \ ((0 | SRC_CONST) << R300_ALU_SRC1A_SHIFT) | \ ((0 | SRC_CONST) << R300_ALU_SRC2A_SHIFT)) /* * Datas structures for fragment program generation */ /* description of r300 native hw instructions */ static const struct { const char *name; int argc; int v_op; int s_op; } r300_fpop[] = { /* *INDENT-OFF* */ {"MAD", 3, R300_ALU_OUTC_MAD, R300_ALU_OUTA_MAD}, {"DP3", 2, R300_ALU_OUTC_DP3, R300_ALU_OUTA_DP4}, {"DP4", 2, R300_ALU_OUTC_DP4, R300_ALU_OUTA_DP4}, {"MIN", 2, R300_ALU_OUTC_MIN, R300_ALU_OUTA_MIN}, {"MAX", 2, R300_ALU_OUTC_MAX, R300_ALU_OUTA_MAX}, {"CMP", 3, R300_ALU_OUTC_CMP, R300_ALU_OUTA_CMP}, {"FRC", 1, R300_ALU_OUTC_FRC, R300_ALU_OUTA_FRC}, {"EX2", 1, R300_ALU_OUTC_REPL_ALPHA, R300_ALU_OUTA_EX2}, {"LG2", 1, R300_ALU_OUTC_REPL_ALPHA, R300_ALU_OUTA_LG2}, {"RCP", 1, R300_ALU_OUTC_REPL_ALPHA, R300_ALU_OUTA_RCP}, {"RSQ", 1, R300_ALU_OUTC_REPL_ALPHA, R300_ALU_OUTA_RSQ}, {"REPL_ALPHA", 1, R300_ALU_OUTC_REPL_ALPHA, PFS_INVAL}, {"CMPH", 3, R300_ALU_OUTC_CMPH, PFS_INVAL}, /* *INDENT-ON* */ }; /* vector swizzles r300 can support natively, with a couple of * cases we handle specially * * REG_VSWZ/REG_SSWZ is an index into this table */ /* mapping from SWIZZLE_* to r300 native values for scalar insns */ #define SWIZZLE_HALF 6 #define MAKE_SWZ3(x, y, z) (MAKE_SWIZZLE4(SWIZZLE_##x, \ SWIZZLE_##y, \ SWIZZLE_##z, \ SWIZZLE_ZERO)) /* native swizzles */ static const struct r300_pfs_swizzle { GLuint hash; /* swizzle value this matches */ GLuint base; /* base value for hw swizzle */ GLuint stride; /* difference in base between arg0/1/2 */ GLuint flags; } v_swiz[] = { /* *INDENT-OFF* */ {MAKE_SWZ3(X, Y, Z), R300_ALU_ARGC_SRC0C_XYZ, 4, SLOT_SRC_VECTOR}, {MAKE_SWZ3(X, X, X), R300_ALU_ARGC_SRC0C_XXX, 4, SLOT_SRC_VECTOR}, {MAKE_SWZ3(Y, Y, Y), R300_ALU_ARGC_SRC0C_YYY, 4, SLOT_SRC_VECTOR}, {MAKE_SWZ3(Z, Z, Z), R300_ALU_ARGC_SRC0C_ZZZ, 4, SLOT_SRC_VECTOR}, {MAKE_SWZ3(W, W, W), R300_ALU_ARGC_SRC0A, 1, SLOT_SRC_SCALAR}, {MAKE_SWZ3(Y, Z, X), R300_ALU_ARGC_SRC0C_YZX, 1, SLOT_SRC_VECTOR}, {MAKE_SWZ3(Z, X, Y), R300_ALU_ARGC_SRC0C_ZXY, 1, SLOT_SRC_VECTOR}, {MAKE_SWZ3(W, Z, Y), R300_ALU_ARGC_SRC0CA_WZY, 1, SLOT_SRC_BOTH}, {MAKE_SWZ3(ONE, ONE, ONE), R300_ALU_ARGC_ONE, 0, 0}, {MAKE_SWZ3(ZERO, ZERO, ZERO), R300_ALU_ARGC_ZERO, 0, 0}, {MAKE_SWZ3(HALF, HALF, HALF), R300_ALU_ARGC_HALF, 0, 0}, {PFS_INVAL, 0, 0, 0}, /* *INDENT-ON* */ }; /* used during matching of non-native swizzles */ #define SWZ_X_MASK (7 << 0) #define SWZ_Y_MASK (7 << 3) #define SWZ_Z_MASK (7 << 6) #define SWZ_W_MASK (7 << 9) static const struct { GLuint hash; /* used to mask matching swizzle components */ int mask; /* actual outmask */ int count; /* count of components matched */ } s_mask[] = { /* *INDENT-OFF* */ {SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK, 1 | 2 | 4, 3}, {SWZ_X_MASK | SWZ_Y_MASK, 1 | 2, 2}, {SWZ_X_MASK | SWZ_Z_MASK, 1 | 4, 2}, {SWZ_Y_MASK | SWZ_Z_MASK, 2 | 4, 2}, {SWZ_X_MASK, 1, 1}, {SWZ_Y_MASK, 2, 1}, {SWZ_Z_MASK, 4, 1}, {PFS_INVAL, PFS_INVAL, PFS_INVAL} /* *INDENT-ON* */ }; static const struct { int base; /* hw value of swizzle */ int stride; /* difference between SRC0/1/2 */ GLuint flags; } s_swiz[] = { /* *INDENT-OFF* */ {R300_ALU_ARGA_SRC0C_X, 3, SLOT_SRC_VECTOR}, {R300_ALU_ARGA_SRC0C_Y, 3, SLOT_SRC_VECTOR}, {R300_ALU_ARGA_SRC0C_Z, 3, SLOT_SRC_VECTOR}, {R300_ALU_ARGA_SRC0A, 1, SLOT_SRC_SCALAR}, {R300_ALU_ARGA_ZERO, 0, 0}, {R300_ALU_ARGA_ONE, 0, 0}, {R300_ALU_ARGA_HALF, 0, 0} /* *INDENT-ON* */ }; /* boiler-plate reg, for convenience */ static const GLuint undef = REG(REG_TYPE_TEMP, 0, SWIZZLE_XYZ, SWIZZLE_W, GL_FALSE, GL_FALSE, GL_FALSE); /* constant one source */ static const GLuint pfs_one = REG(REG_TYPE_CONST, 0, SWIZZLE_111, SWIZZLE_ONE, GL_FALSE, GL_TRUE, GL_TRUE); /* constant half source */ static const GLuint pfs_half = REG(REG_TYPE_CONST, 0, SWIZZLE_HHH, SWIZZLE_HALF, GL_FALSE, GL_TRUE, GL_TRUE); /* constant zero source */ static const GLuint pfs_zero = REG(REG_TYPE_CONST, 0, SWIZZLE_000, SWIZZLE_ZERO, GL_FALSE, GL_TRUE, GL_TRUE); /* * Common functions prototypes */ static void emit_arith(struct r300_pfs_compile_state *cs, int op, GLuint dest, int mask, GLuint src0, GLuint src1, GLuint src2, int flags); /** * Get an R300 temporary that can be written to in the given slot. */ static int get_hw_temp(struct r300_pfs_compile_state *cs, int slot) { COMPILE_STATE; int r; for (r = 0; r < PFS_NUM_TEMP_REGS; ++r) { if (cs->hwtemps[r].free >= 0 && cs->hwtemps[r].free <= slot) break; } if (r >= PFS_NUM_TEMP_REGS) { ERROR("Out of hardware temps\n"); return 0; } // Reserved is used to avoid the following scenario: // R300 temporary X is first assigned to Mesa temporary Y during vector ops // R300 temporary X is then assigned to Mesa temporary Z for further vector ops // Then scalar ops on Mesa temporary Z are emitted and move back in time // to overwrite the value of temporary Y. // End scenario. cs->hwtemps[r].reserved = cs->hwtemps[r].free; cs->hwtemps[r].free = -1; // Reset to some value that won't mess things up when the user // tries to read from a temporary that hasn't been assigned a value yet. // In the normal case, vector_valid and scalar_valid should be set to // a sane value by the first emit that writes to this temporary. cs->hwtemps[r].vector_valid = 0; cs->hwtemps[r].scalar_valid = 0; if (r > code->max_temp_idx) code->max_temp_idx = r; return r; } /** * Get an R300 temporary that will act as a TEX destination register. */ static int get_hw_temp_tex(struct r300_pfs_compile_state *cs) { COMPILE_STATE; int r; for (r = 0; r < PFS_NUM_TEMP_REGS; ++r) { if (cs->used_in_node & (1 << r)) continue; // Note: Be very careful here if (cs->hwtemps[r].free >= 0 && cs->hwtemps[r].free <= 0) break; } if (r >= PFS_NUM_TEMP_REGS) return get_hw_temp(cs, 0); /* Will cause an indirection */ cs->hwtemps[r].reserved = cs->hwtemps[r].free; cs->hwtemps[r].free = -1; // Reset to some value that won't mess things up when the user // tries to read from a temporary that hasn't been assigned a value yet. // In the normal case, vector_valid and scalar_valid should be set to // a sane value by the first emit that writes to this temporary. cs->hwtemps[r].vector_valid = cs->nrslots; cs->hwtemps[r].scalar_valid = cs->nrslots; if (r > code->max_temp_idx) code->max_temp_idx = r; return r; } /** * Mark the given hardware register as free. */ static void free_hw_temp(struct r300_pfs_compile_state *cs, int idx) { // Be very careful here. Consider sequences like // MAD r0, r1,r2,r3 // TEX r4, ... // The TEX instruction may be moved in front of the MAD instruction // due to the way nodes work. We don't want to alias r1 and r4 in // this case. // I'm certain the register allocation could be further sanitized, // but it's tricky because of stuff that can happen inside emit_tex // and emit_arith. cs->hwtemps[idx].free = cs->nrslots + 1; } /** * Create a new Mesa temporary register. */ static GLuint get_temp_reg(struct r300_pfs_compile_state *cs) { COMPILE_STATE; GLuint r = undef; GLuint index; index = ffs(~cs->temp_in_use); if (!index) { ERROR("Out of program temps\n"); return r; } cs->temp_in_use |= (1 << --index); cs->temps[index].refcount = 0xFFFFFFFF; cs->temps[index].reg = -1; REG_SET_TYPE(r, REG_TYPE_TEMP); REG_SET_INDEX(r, index); REG_SET_VALID(r, GL_TRUE); return r; } /** * Free a Mesa temporary and the associated R300 temporary. */ static void free_temp(struct r300_pfs_compile_state *cs, GLuint r) { GLuint index = REG_GET_INDEX(r); if (!(cs->temp_in_use & (1 << index))) return; if (REG_GET_TYPE(r) == REG_TYPE_TEMP) { free_hw_temp(cs, cs->temps[index].reg); cs->temps[index].reg = -1; cs->temp_in_use &= ~(1 << index); } else if (REG_GET_TYPE(r) == REG_TYPE_INPUT) { free_hw_temp(cs, cs->inputs[index].reg); cs->inputs[index].reg = -1; } } /** * Emit a hardware constant/parameter. * * \p cp Stable pointer to an array of 4 floats. * The pointer must be stable in the sense that it remains to be valid * and hold the contents of the constant/parameter throughout the lifetime * of the fragment program (actually, up until the next time the fragment * program is translated). */ static GLuint emit_const4fv(struct r300_pfs_compile_state *cs, const GLfloat * cp) { COMPILE_STATE; GLuint reg = undef; int index; for (index = 0; index < code->const_nr; ++index) { if (code->constant[index] == cp) break; } if (index >= code->const_nr) { if (index >= PFS_NUM_CONST_REGS) { ERROR("Out of hw constants!\n"); return reg; } code->const_nr++; code->constant[index] = cp; } REG_SET_TYPE(reg, REG_TYPE_CONST); REG_SET_INDEX(reg, index); REG_SET_VALID(reg, GL_TRUE); return reg; } static INLINE GLuint negate(GLuint r) { REG_NEGS(r); REG_NEGV(r); return r; } /* Hack, to prevent clobbering sources used multiple times when * emulating non-native instructions */ static INLINE GLuint keep(GLuint r) { REG_SET_NO_USE(r, GL_TRUE); return r; } static INLINE GLuint absolute(GLuint r) { REG_ABS(r); return r; } static int swz_native(struct r300_pfs_compile_state *cs, GLuint src, GLuint * r, GLuint arbneg) { COMPILE_STATE; /* Native swizzle, handle negation */ src = (src & ~REG_NEGS_MASK) | (((arbneg >> 3) & 1) << REG_NEGS_SHIFT); if ((arbneg & 0x7) == 0x0) { src = src & ~REG_NEGV_MASK; *r = src; } else if ((arbneg & 0x7) == 0x7) { src |= REG_NEGV_MASK; *r = src; } else { if (!REG_GET_VALID(*r)) *r = get_temp_reg(cs); src |= REG_NEGV_MASK; emit_arith(cs, PFS_OP_MAD, *r, arbneg & 0x7, keep(src), pfs_one, pfs_zero, 0); src = src & ~REG_NEGV_MASK; emit_arith(cs, PFS_OP_MAD, *r, (arbneg ^ 0x7) | WRITEMASK_W, src, pfs_one, pfs_zero, 0); } return 3; } static int swz_emit_partial(struct r300_pfs_compile_state *cs, GLuint src, GLuint * r, int mask, int mc, GLuint arbneg) { COMPILE_STATE; GLuint tmp; GLuint wmask = 0; if (!REG_GET_VALID(*r)) *r = get_temp_reg(cs); /* A partial match, VSWZ/mask define what parts of the * desired swizzle we match */ if (mc + s_mask[mask].count == 3) { wmask = WRITEMASK_W; src |= ((arbneg >> 3) & 1) << REG_NEGS_SHIFT; } tmp = arbneg & s_mask[mask].mask; if (tmp) { tmp = tmp ^ s_mask[mask].mask; if (tmp) { emit_arith(cs, PFS_OP_MAD, *r, arbneg & s_mask[mask].mask, keep(src) | REG_NEGV_MASK, pfs_one, pfs_zero, 0); if (!wmask) { REG_SET_NO_USE(src, GL_TRUE); } else { REG_SET_NO_USE(src, GL_FALSE); } emit_arith(cs, PFS_OP_MAD, *r, tmp | wmask, src, pfs_one, pfs_zero, 0); } else { if (!wmask) { REG_SET_NO_USE(src, GL_TRUE); } else { REG_SET_NO_USE(src, GL_FALSE); } emit_arith(cs, PFS_OP_MAD, *r, (arbneg & s_mask[mask].mask) | wmask, src | REG_NEGV_MASK, pfs_one, pfs_zero, 0); } } else { if (!wmask) { REG_SET_NO_USE(src, GL_TRUE); } else { REG_SET_NO_USE(src, GL_FALSE); } emit_arith(cs, PFS_OP_MAD, *r, s_mask[mask].mask | wmask, src, pfs_one, pfs_zero, 0); } return s_mask[mask].count; } static GLuint do_swizzle(struct r300_pfs_compile_state *cs, GLuint src, GLuint arbswz, GLuint arbneg) { COMPILE_STATE; GLuint r = undef; GLuint vswz; int c_mask = 0; int v_match = 0; /* If swizzling from something without an XYZW native swizzle, * emit result to a temp, and do new swizzle from the temp. */ #if 0 if (REG_GET_VSWZ(src) != SWIZZLE_XYZ || REG_GET_SSWZ(src) != SWIZZLE_W) { GLuint temp = get_temp_reg(fp); emit_arith(fp, PFS_OP_MAD, temp, WRITEMASK_XYZW, src, pfs_one, pfs_zero, 0); src = temp; } #endif if (REG_GET_VSWZ(src) != SWIZZLE_XYZ || REG_GET_SSWZ(src) != SWIZZLE_W) { GLuint vsrcswz = (v_swiz[REG_GET_VSWZ(src)]. hash & (SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK)) | REG_GET_SSWZ(src) << 9; GLint i; GLuint newswz = 0; GLuint offset; for (i = 0; i < 4; ++i) { offset = GET_SWZ(arbswz, i); newswz |= (offset <= 3) ? GET_SWZ(vsrcswz, offset) << i * 3 : offset << i * 3; } arbswz = newswz & (SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK); REG_SET_SSWZ(src, GET_SWZ(newswz, 3)); } else { /* set scalar swizzling */ REG_SET_SSWZ(src, GET_SWZ(arbswz, 3)); } do { vswz = REG_GET_VSWZ(src); do { int chash; REG_SET_VSWZ(src, vswz); chash = v_swiz[REG_GET_VSWZ(src)].hash & s_mask[c_mask].hash; if (chash == (arbswz & s_mask[c_mask].hash)) { if (s_mask[c_mask].count == 3) { v_match += swz_native(cs, src, &r, arbneg); } else { v_match += swz_emit_partial(cs, src, &r, c_mask, v_match, arbneg); } if (v_match == 3) return r; /* Fill with something invalid.. all 0's was * wrong before, matched SWIZZLE_X. So all * 1's will be okay for now */ arbswz |= (PFS_INVAL & s_mask[c_mask].hash); } } while (v_swiz[++vswz].hash != PFS_INVAL); REG_SET_VSWZ(src, SWIZZLE_XYZ); } while (s_mask[++c_mask].hash != PFS_INVAL); ERROR("should NEVER get here\n"); return r; } static GLuint t_src(struct r300_pfs_compile_state *cs, struct prog_src_register fpsrc) { COMPILE_STATE; GLuint r = undef; switch (fpsrc.File) { case PROGRAM_TEMPORARY: REG_SET_INDEX(r, fpsrc.Index); REG_SET_VALID(r, GL_TRUE); REG_SET_TYPE(r, REG_TYPE_TEMP); break; case PROGRAM_INPUT: REG_SET_INDEX(r, fpsrc.Index); REG_SET_VALID(r, GL_TRUE); REG_SET_TYPE(r, REG_TYPE_INPUT); break; case PROGRAM_LOCAL_PARAM: r = emit_const4fv(cs, fp->mesa_program.Base.LocalParams[fpsrc. Index]); break; case PROGRAM_ENV_PARAM: r = emit_const4fv(cs, cs->compiler->r300->radeon.glCtx->FragmentProgram.Parameters[fpsrc.Index]); break; case PROGRAM_STATE_VAR: case PROGRAM_NAMED_PARAM: case PROGRAM_CONSTANT: r = emit_const4fv(cs, fp->mesa_program.Base.Parameters-> ParameterValues[fpsrc.Index]); break; case PROGRAM_BUILTIN: switch(fpsrc.Swizzle) { case SWIZZLE_1111: r = pfs_one; break; case SWIZZLE_0000: r = pfs_zero; break; default: ERROR("bad PROGRAM_BUILTIN swizzle %u\n", fpsrc.Swizzle); break; } break; default: ERROR("unknown SrcReg->File %x\n", fpsrc.File); return r; } /* no point swizzling ONE/ZERO/HALF constants... */ if (REG_GET_VSWZ(r) < SWIZZLE_111 || REG_GET_SSWZ(r) < SWIZZLE_ZERO) r = do_swizzle(cs, r, fpsrc.Swizzle, fpsrc.NegateBase); if (fpsrc.Abs) r = absolute(r); if (fpsrc.NegateAbs) r = negate(r); return r; } static GLuint t_scalar_src(struct r300_pfs_compile_state *cs, struct prog_src_register fpsrc) { struct prog_src_register src = fpsrc; int sc = GET_SWZ(fpsrc.Swizzle, 0); /* X */ src.Swizzle = ((sc << 0) | (sc << 3) | (sc << 6) | (sc << 9)); return t_src(cs, src); } static GLuint t_dst(struct r300_pfs_compile_state *cs, struct prog_dst_register dest) { COMPILE_STATE; GLuint r = undef; switch (dest.File) { case PROGRAM_TEMPORARY: REG_SET_INDEX(r, dest.Index); REG_SET_VALID(r, GL_TRUE); REG_SET_TYPE(r, REG_TYPE_TEMP); return r; case PROGRAM_OUTPUT: REG_SET_TYPE(r, REG_TYPE_OUTPUT); switch (dest.Index) { case FRAG_RESULT_COLR: case FRAG_RESULT_DEPR: REG_SET_INDEX(r, dest.Index); REG_SET_VALID(r, GL_TRUE); return r; default: ERROR("Bad DstReg->Index 0x%x\n", dest.Index); return r; } default: ERROR("Bad DstReg->File 0x%x\n", dest.File); return r; } } static int t_hw_src(struct r300_pfs_compile_state *cs, GLuint src, GLboolean tex) { COMPILE_STATE; int idx; int index = REG_GET_INDEX(src); switch (REG_GET_TYPE(src)) { case REG_TYPE_TEMP: /* NOTE: if reg==-1 here, a source is being read that * hasn't been written to. Undefined results. */ if (cs->temps[index].reg == -1) cs->temps[index].reg = get_hw_temp(cs, cs->nrslots); idx = cs->temps[index].reg; if (!REG_GET_NO_USE(src) && (--cs->temps[index].refcount == 0)) free_temp(cs, src); break; case REG_TYPE_INPUT: idx = cs->inputs[index].reg; if (!REG_GET_NO_USE(src) && (--cs->inputs[index].refcount == 0)) free_hw_temp(cs, cs->inputs[index].reg); break; case REG_TYPE_CONST: return (index | SRC_CONST); default: ERROR("Invalid type for source reg\n"); return (0 | SRC_CONST); } if (!tex) cs->used_in_node |= (1 << idx); return idx; } static int t_hw_dst(struct r300_pfs_compile_state *cs, GLuint dest, GLboolean tex, int slot) { COMPILE_STATE; int idx; GLuint index = REG_GET_INDEX(dest); assert(REG_GET_VALID(dest)); switch (REG_GET_TYPE(dest)) { case REG_TYPE_TEMP: if (cs->temps[REG_GET_INDEX(dest)].reg == -1) { if (!tex) { cs->temps[index].reg = get_hw_temp(cs, slot); } else { cs->temps[index].reg = get_hw_temp_tex(cs); } } idx = cs->temps[index].reg; if (!REG_GET_NO_USE(dest) && (--cs->temps[index].refcount == 0)) free_temp(cs, dest); cs->dest_in_node |= (1 << idx); cs->used_in_node |= (1 << idx); break; case REG_TYPE_OUTPUT: switch (index) { case FRAG_RESULT_COLR: code->node[code->cur_node].flags |= R300_RGBA_OUT; break; case FRAG_RESULT_DEPR: fp->WritesDepth = GL_TRUE; code->node[code->cur_node].flags |= R300_W_OUT; break; } return index; break; default: ERROR("invalid dest reg type %d\n", REG_GET_TYPE(dest)); return 0; } return idx; } static void emit_nop(struct r300_pfs_compile_state *cs) { COMPILE_STATE; if (cs->nrslots >= PFS_MAX_ALU_INST) { ERROR("Out of ALU instruction slots\n"); return; } code->alu.inst[cs->nrslots].inst0 = NOP_INST0; code->alu.inst[cs->nrslots].inst1 = NOP_INST1; code->alu.inst[cs->nrslots].inst2 = NOP_INST2; code->alu.inst[cs->nrslots].inst3 = NOP_INST3; cs->nrslots++; } static void emit_tex(struct r300_pfs_compile_state *cs, struct prog_instruction *fpi, int opcode) { COMPILE_STATE; GLuint coord = t_src(cs, fpi->SrcReg[0]); GLuint dest = undef; GLuint din, uin; int unit = fpi->TexSrcUnit; int hwsrc, hwdest; /* Ensure correct node indirection */ uin = cs->used_in_node; din = cs->dest_in_node; /* Resolve source/dest to hardware registers */ hwsrc = t_hw_src(cs, coord, GL_TRUE); if (opcode != R300_TEX_OP_KIL) { dest = t_dst(cs, fpi->DstReg); hwdest = t_hw_dst(cs, dest, GL_TRUE, code->node[code->cur_node].alu_offset); /* Use a temp that hasn't been used in this node, rather * than causing an indirection */ if (uin & (1 << hwdest)) { free_hw_temp(cs, hwdest); hwdest = get_hw_temp_tex(cs); cs->temps[REG_GET_INDEX(dest)].reg = hwdest; } } else { hwdest = 0; unit = 0; } /* Indirection if source has been written in this node, or if the * dest has been read/written in this node */ if ((REG_GET_TYPE(coord) != REG_TYPE_CONST && (din & (1 << hwsrc))) || (uin & (1 << hwdest))) { /* Finish off current node */ if (code->node[code->cur_node].alu_offset == cs->nrslots) emit_nop(cs); code->node[code->cur_node].alu_end = cs->nrslots - code->node[code->cur_node].alu_offset - 1; assert(code->node[code->cur_node].alu_end >= 0); if (++code->cur_node >= PFS_MAX_TEX_INDIRECT) { ERROR("too many levels of texture indirection\n"); return; } /* Start new node */ code->node[code->cur_node].tex_offset = code->tex.length; code->node[code->cur_node].alu_offset = cs->nrslots; code->node[code->cur_node].tex_end = -1; code->node[code->cur_node].alu_end = -1; code->node[code->cur_node].flags = 0; cs->used_in_node = 0; cs->dest_in_node = 0; } if (code->cur_node == 0) code->first_node_has_tex = 1; code->tex.inst[code->tex.length++] = 0 | (hwsrc << R300_SRC_ADDR_SHIFT) | (hwdest << R300_DST_ADDR_SHIFT) | (unit << R300_TEX_ID_SHIFT) | (opcode << R300_TEX_INST_SHIFT); cs->dest_in_node |= (1 << hwdest); if (REG_GET_TYPE(coord) != REG_TYPE_CONST) cs->used_in_node |= (1 << hwsrc); code->node[code->cur_node].tex_end++; } /** * Returns the first slot where we could possibly allow writing to dest, * according to register allocation. */ static int get_earliest_allowed_write(struct r300_pfs_compile_state *cs, GLuint dest, int mask) { COMPILE_STATE; int idx; int pos; GLuint index = REG_GET_INDEX(dest); assert(REG_GET_VALID(dest)); switch (REG_GET_TYPE(dest)) { case REG_TYPE_TEMP: if (cs->temps[index].reg == -1) return 0; idx = cs->temps[index].reg; break; case REG_TYPE_OUTPUT: return 0; default: ERROR("invalid dest reg type %d\n", REG_GET_TYPE(dest)); return 0; } pos = cs->hwtemps[idx].reserved; if (mask & WRITEMASK_XYZ) { if (pos < cs->hwtemps[idx].vector_lastread) pos = cs->hwtemps[idx].vector_lastread; } if (mask & WRITEMASK_W) { if (pos < cs->hwtemps[idx].scalar_lastread) pos = cs->hwtemps[idx].scalar_lastread; } return pos; } /** * Allocates a slot for an ALU instruction that can consist of * a vertex part or a scalar part or both. * * Sources from src (src[0] to src[argc-1]) are added to the slot in the * appropriate position (vector and/or scalar), and their positions are * recorded in the srcpos array. * * This function emits instruction code for the source fetch and the * argument selection. It does not emit instruction code for the * opcode or the destination selection. * * @return the index of the slot */ static int find_and_prepare_slot(struct r300_pfs_compile_state *cs, GLboolean emit_vop, GLboolean emit_sop, int argc, GLuint * src, GLuint dest, int mask) { COMPILE_STATE; int hwsrc[3]; int srcpos[3]; unsigned int used; int tempused; int tempvsrc[3]; int tempssrc[3]; int pos; int regnr; int i, j; // Determine instruction slots, whether sources are required on // vector or scalar side, and the smallest slot number where // all source registers are available used = 0; if (emit_vop) used |= SLOT_OP_VECTOR; if (emit_sop) used |= SLOT_OP_SCALAR; pos = get_earliest_allowed_write(cs, dest, mask); if (code->node[code->cur_node].alu_offset > pos) pos = code->node[code->cur_node].alu_offset; for (i = 0; i < argc; ++i) { if (!REG_GET_BUILTIN(src[i])) { if (emit_vop) used |= v_swiz[REG_GET_VSWZ(src[i])].flags << i; if (emit_sop) used |= s_swiz[REG_GET_SSWZ(src[i])].flags << i; } hwsrc[i] = t_hw_src(cs, src[i], GL_FALSE); /* Note: sideeffects wrt refcounting! */ regnr = hwsrc[i] & 31; if (REG_GET_TYPE(src[i]) == REG_TYPE_TEMP) { if (used & (SLOT_SRC_VECTOR << i)) { if (cs->hwtemps[regnr].vector_valid > pos) pos = cs->hwtemps[regnr].vector_valid; } if (used & (SLOT_SRC_SCALAR << i)) { if (cs->hwtemps[regnr].scalar_valid > pos) pos = cs->hwtemps[regnr].scalar_valid; } } } // Find a slot that fits for (;; ++pos) { if (cs->slot[pos].used & used & SLOT_OP_BOTH) continue; if (pos >= cs->nrslots) { if (cs->nrslots >= PFS_MAX_ALU_INST) { ERROR("Out of ALU instruction slots\n"); return -1; } code->alu.inst[pos].inst0 = NOP_INST0; code->alu.inst[pos].inst1 = NOP_INST1; code->alu.inst[pos].inst2 = NOP_INST2; code->alu.inst[pos].inst3 = NOP_INST3; cs->nrslots++; } // Note: When we need both parts (vector and scalar) of a source, // we always try to put them into the same position. This makes the // code easier to read, and it is optimal (i.e. one doesn't gain // anything by splitting the parts). // It also avoids headaches with swizzles that access both parts (i.e WXY) tempused = cs->slot[pos].used; for (i = 0; i < 3; ++i) { tempvsrc[i] = cs->slot[pos].vsrc[i]; tempssrc[i] = cs->slot[pos].ssrc[i]; } for (i = 0; i < argc; ++i) { int flags = (used >> i) & SLOT_SRC_BOTH; if (!flags) { srcpos[i] = 0; continue; } for (j = 0; j < 3; ++j) { if ((tempused >> j) & flags & SLOT_SRC_VECTOR) { if (tempvsrc[j] != hwsrc[i]) continue; } if ((tempused >> j) & flags & SLOT_SRC_SCALAR) { if (tempssrc[j] != hwsrc[i]) continue; } break; } if (j == 3) break; srcpos[i] = j; tempused |= flags << j; if (flags & SLOT_SRC_VECTOR) tempvsrc[j] = hwsrc[i]; if (flags & SLOT_SRC_SCALAR) tempssrc[j] = hwsrc[i]; } if (i == argc) break; } // Found a slot, reserve it cs->slot[pos].used = tempused | (used & SLOT_OP_BOTH); for (i = 0; i < 3; ++i) { cs->slot[pos].vsrc[i] = tempvsrc[i]; cs->slot[pos].ssrc[i] = tempssrc[i]; } for (i = 0; i < argc; ++i) { if (REG_GET_TYPE(src[i]) == REG_TYPE_TEMP) { int regnr = hwsrc[i] & 31; if (used & (SLOT_SRC_VECTOR << i)) { if (cs->hwtemps[regnr].vector_lastread < pos) cs->hwtemps[regnr].vector_lastread = pos; } if (used & (SLOT_SRC_SCALAR << i)) { if (cs->hwtemps[regnr].scalar_lastread < pos) cs->hwtemps[regnr].scalar_lastread = pos; } } } // Emit the source fetch code code->alu.inst[pos].inst1 &= ~R300_ALU_SRC_MASK; code->alu.inst[pos].inst1 |= ((cs->slot[pos].vsrc[0] << R300_ALU_SRC0C_SHIFT) | (cs->slot[pos].vsrc[1] << R300_ALU_SRC1C_SHIFT) | (cs->slot[pos].vsrc[2] << R300_ALU_SRC2C_SHIFT)); code->alu.inst[pos].inst3 &= ~R300_ALU_SRC_MASK; code->alu.inst[pos].inst3 |= ((cs->slot[pos].ssrc[0] << R300_ALU_SRC0A_SHIFT) | (cs->slot[pos].ssrc[1] << R300_ALU_SRC1A_SHIFT) | (cs->slot[pos].ssrc[2] << R300_ALU_SRC2A_SHIFT)); // Emit the argument selection code if (emit_vop) { int swz[3]; for (i = 0; i < 3; ++i) { if (i < argc) { swz[i] = (v_swiz[REG_GET_VSWZ(src[i])].base + (srcpos[i] * v_swiz[REG_GET_VSWZ(src[i])]. stride)) | ((src[i] & REG_NEGV_MASK) ? ARG_NEG : 0) | ((src[i] & REG_ABS_MASK) ? ARG_ABS : 0); } else { swz[i] = R300_ALU_ARGC_ZERO; } } code->alu.inst[pos].inst0 &= ~(R300_ALU_ARG0C_MASK | R300_ALU_ARG1C_MASK | R300_ALU_ARG2C_MASK); code->alu.inst[pos].inst0 |= (swz[0] << R300_ALU_ARG0C_SHIFT) | (swz[1] << R300_ALU_ARG1C_SHIFT) | (swz[2] << R300_ALU_ARG2C_SHIFT); } if (emit_sop) { int swz[3]; for (i = 0; i < 3; ++i) { if (i < argc) { swz[i] = (s_swiz[REG_GET_SSWZ(src[i])].base + (srcpos[i] * s_swiz[REG_GET_SSWZ(src[i])]. stride)) | ((src[i] & REG_NEGS_MASK) ? ARG_NEG : 0) | ((src[i] & REG_ABS_MASK) ? ARG_ABS : 0); } else { swz[i] = R300_ALU_ARGA_ZERO; } } code->alu.inst[pos].inst2 &= ~(R300_ALU_ARG0A_MASK | R300_ALU_ARG1A_MASK | R300_ALU_ARG2A_MASK); code->alu.inst[pos].inst2 |= (swz[0] << R300_ALU_ARG0A_SHIFT) | (swz[1] << R300_ALU_ARG1A_SHIFT) | (swz[2] << R300_ALU_ARG2A_SHIFT); } return pos; } /** * Append an ALU instruction to the instruction list. */ static void emit_arith(struct r300_pfs_compile_state *cs, int op, GLuint dest, int mask, GLuint src0, GLuint src1, GLuint src2, int flags) { COMPILE_STATE; GLuint src[3] = { src0, src1, src2 }; int hwdest; GLboolean emit_vop, emit_sop; int vop, sop, argc; int pos; vop = r300_fpop[op].v_op; sop = r300_fpop[op].s_op; argc = r300_fpop[op].argc; if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT && REG_GET_INDEX(dest) == FRAG_RESULT_DEPR) { if (mask & WRITEMASK_Z) { mask = WRITEMASK_W; } else { return; } } emit_vop = GL_FALSE; emit_sop = GL_FALSE; if ((mask & WRITEMASK_XYZ) || vop == R300_ALU_OUTC_DP3) emit_vop = GL_TRUE; if ((mask & WRITEMASK_W) || vop == R300_ALU_OUTC_REPL_ALPHA) emit_sop = GL_TRUE; pos = find_and_prepare_slot(cs, emit_vop, emit_sop, argc, src, dest, mask); if (pos < 0) return; hwdest = t_hw_dst(cs, dest, GL_FALSE, pos); /* Note: Side effects wrt register allocation */ if (flags & PFS_FLAG_SAT) { vop |= R300_ALU_OUTC_CLAMP; sop |= R300_ALU_OUTA_CLAMP; } /* Throw the pieces together and get ALU/1 */ if (emit_vop) { code->alu.inst[pos].inst0 |= vop; code->alu.inst[pos].inst1 |= hwdest << R300_ALU_DSTC_SHIFT; if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) { if (REG_GET_INDEX(dest) == FRAG_RESULT_COLR) { code->alu.inst[pos].inst1 |= (mask & WRITEMASK_XYZ) << R300_ALU_DSTC_OUTPUT_MASK_SHIFT; } else assert(0); } else { code->alu.inst[pos].inst1 |= (mask & WRITEMASK_XYZ) << R300_ALU_DSTC_REG_MASK_SHIFT; cs->hwtemps[hwdest].vector_valid = pos + 1; } } /* And now ALU/3 */ if (emit_sop) { code->alu.inst[pos].inst2 |= sop; if (mask & WRITEMASK_W) { if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) { if (REG_GET_INDEX(dest) == FRAG_RESULT_COLR) { code->alu.inst[pos].inst3 |= (hwdest << R300_ALU_DSTA_SHIFT) | R300_ALU_DSTA_OUTPUT; } else if (REG_GET_INDEX(dest) == FRAG_RESULT_DEPR) { code->alu.inst[pos].inst3 |= R300_ALU_DSTA_DEPTH; } else assert(0); } else { code->alu.inst[pos].inst3 |= (hwdest << R300_ALU_DSTA_SHIFT) | R300_ALU_DSTA_REG; cs->hwtemps[hwdest].scalar_valid = pos + 1; } } } return; } static GLfloat SinCosConsts[2][4] = { { 1.273239545, // 4/PI -0.405284735, // -4/(PI*PI) 3.141592654, // PI 0.2225 // weight }, { 0.75, 0.0, 0.159154943, // 1/(2*PI) 6.283185307 // 2*PI } }; /** * Emit a LIT instruction. * \p flags may be PFS_FLAG_SAT * * Definition of LIT (from ARB_fragment_program): * tmp = VectorLoad(op0); * if (tmp.x < 0) tmp.x = 0; * if (tmp.y < 0) tmp.y = 0; * if (tmp.w < -(128.0-epsilon)) tmp.w = -(128.0-epsilon); * else if (tmp.w > 128-epsilon) tmp.w = 128-epsilon; * result.x = 1.0; * result.y = tmp.x; * result.z = (tmp.x > 0) ? RoughApproxPower(tmp.y, tmp.w) : 0.0; * result.w = 1.0; * * The longest path of computation is the one leading to result.z, * consisting of 5 operations. This implementation of LIT takes * 5 slots. So unless there's some special undocumented opcode, * this implementation is potentially optimal. Unfortunately, * emit_arith is a bit too conservative because it doesn't understand * partial writes to the vector component. */ static const GLfloat LitConst[4] = { 127.999999, 127.999999, 127.999999, -127.999999 }; static void emit_lit(struct r300_pfs_compile_state *cs, GLuint dest, int mask, GLuint src, int flags) { COMPILE_STATE; GLuint cnst; int needTemporary; GLuint temp; cnst = emit_const4fv(cs, LitConst); needTemporary = 0; if ((mask & WRITEMASK_XYZW) != WRITEMASK_XYZW) { needTemporary = 1; } else if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) { // LIT is typically followed by DP3/DP4, so there's no point // in creating special code for this case needTemporary = 1; } if (needTemporary) { temp = keep(get_temp_reg(cs)); } else { temp = keep(dest); } // Note: The order of emit_arith inside the slots is relevant, // because emit_arith only looks at scalar vs. vector when resolving // dependencies, and it does not consider individual vector components, // so swizzling between the two parts can create fake dependencies. // First slot emit_arith(cs, PFS_OP_MAX, temp, WRITEMASK_XY, keep(src), pfs_zero, undef, 0); emit_arith(cs, PFS_OP_MAX, temp, WRITEMASK_W, src, cnst, undef, 0); // Second slot emit_arith(cs, PFS_OP_MIN, temp, WRITEMASK_Z, swizzle(temp, W, W, W, W), cnst, undef, 0); emit_arith(cs, PFS_OP_LG2, temp, WRITEMASK_W, swizzle(temp, Y, Y, Y, Y), undef, undef, 0); // Third slot // If desired, we saturate the y result here. // This does not affect the use as a condition variable in the CMP later emit_arith(cs, PFS_OP_MAD, temp, WRITEMASK_W, temp, swizzle(temp, Z, Z, Z, Z), pfs_zero, 0); emit_arith(cs, PFS_OP_MAD, temp, WRITEMASK_Y, swizzle(temp, X, X, X, X), pfs_one, pfs_zero, flags); // Fourth slot emit_arith(cs, PFS_OP_MAD, temp, WRITEMASK_X, pfs_one, pfs_one, pfs_zero, 0); emit_arith(cs, PFS_OP_EX2, temp, WRITEMASK_W, temp, undef, undef, 0); // Fifth slot emit_arith(cs, PFS_OP_CMP, temp, WRITEMASK_Z, pfs_zero, swizzle(temp, W, W, W, W), negate(swizzle(temp, Y, Y, Y, Y)), flags); emit_arith(cs, PFS_OP_MAD, temp, WRITEMASK_W, pfs_one, pfs_one, pfs_zero, 0); if (needTemporary) { emit_arith(cs, PFS_OP_MAD, dest, mask, temp, pfs_one, pfs_zero, flags); free_temp(cs, temp); } else { // Decrease refcount of the destination t_hw_dst(cs, dest, GL_FALSE, cs->nrslots); } } static void emit_instruction(struct r300_pfs_compile_state *cs, struct prog_instruction *fpi) { COMPILE_STATE; GLuint src[3], dest, temp[2]; int flags, mask = 0; int const_sin[2]; if (fpi->SaturateMode == SATURATE_ZERO_ONE) flags = PFS_FLAG_SAT; else flags = 0; if (fpi->Opcode != OPCODE_KIL) { dest = t_dst(cs, fpi->DstReg); mask = fpi->DstReg.WriteMask; } switch (fpi->Opcode) { case OPCODE_ADD: src[0] = t_src(cs, fpi->SrcReg[0]); src[1] = t_src(cs, fpi->SrcReg[1]); emit_arith(cs, PFS_OP_MAD, dest, mask, src[0], pfs_one, src[1], flags); break; case OPCODE_CMP: src[0] = t_src(cs, fpi->SrcReg[0]); src[1] = t_src(cs, fpi->SrcReg[1]); src[2] = t_src(cs, fpi->SrcReg[2]); /* ARB_f_p - if src0.c < 0.0 ? src1.c : src2.c * r300 - if src2.c < 0.0 ? src1.c : src0.c */ emit_arith(cs, PFS_OP_CMP, dest, mask, src[2], src[1], src[0], flags); break; case OPCODE_COS: /* * cos using a parabola (see SIN): * cos(x): * x = (x/(2*PI))+0.75 * x = frac(x) * x = (x*2*PI)-PI * result = sin(x) */ temp[0] = get_temp_reg(cs); const_sin[0] = emit_const4fv(cs, SinCosConsts[0]); const_sin[1] = emit_const4fv(cs, SinCosConsts[1]); src[0] = t_scalar_src(cs, fpi->SrcReg[0]); /* add 0.5*PI and do range reduction */ emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X, swizzle(src[0], X, X, X, X), swizzle(const_sin[1], Z, Z, Z, Z), swizzle(const_sin[1], X, X, X, X), 0); emit_arith(cs, PFS_OP_FRC, temp[0], WRITEMASK_X, swizzle(temp[0], X, X, X, X), undef, undef, 0); emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(temp[0], X, X, X, X), swizzle(const_sin[1], W, W, W, W), //2*PI negate(swizzle(const_sin[0], Z, Z, Z, Z)), //-PI 0); /* SIN */ emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0], Z, Z, Z, Z), const_sin[0], pfs_zero, 0); emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X, swizzle(temp[0], Y, Y, Y, Y), absolute(swizzle(temp[0], Z, Z, Z, Z)), swizzle(temp[0], X, X, X, X), 0); emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_Y, swizzle(temp[0], X, X, X, X), absolute(swizzle(temp[0], X, X, X, X)), negate(swizzle(temp[0], X, X, X, X)), 0); emit_arith(cs, PFS_OP_MAD, dest, mask, swizzle(temp[0], Y, Y, Y, Y), swizzle(const_sin[0], W, W, W, W), swizzle(temp[0], X, X, X, X), flags); free_temp(cs, temp[0]); break; case OPCODE_DP3: src[0] = t_src(cs, fpi->SrcReg[0]); src[1] = t_src(cs, fpi->SrcReg[1]); emit_arith(cs, PFS_OP_DP3, dest, mask, src[0], src[1], undef, flags); break; case OPCODE_DP4: src[0] = t_src(cs, fpi->SrcReg[0]); src[1] = t_src(cs, fpi->SrcReg[1]); emit_arith(cs, PFS_OP_DP4, dest, mask, src[0], src[1], undef, flags); break; case OPCODE_DST: src[0] = t_src(cs, fpi->SrcReg[0]); src[1] = t_src(cs, fpi->SrcReg[1]); /* dest.y = src0.y * src1.y */ if (mask & WRITEMASK_Y) emit_arith(cs, PFS_OP_MAD, dest, WRITEMASK_Y, keep(src[0]), keep(src[1]), pfs_zero, flags); /* dest.z = src0.z */ if (mask & WRITEMASK_Z) emit_arith(cs, PFS_OP_MAD, dest, WRITEMASK_Z, src[0], pfs_one, pfs_zero, flags); /* result.x = 1.0 * result.w = src1.w */ if (mask & WRITEMASK_XW) { REG_SET_VSWZ(src[1], SWIZZLE_111); /*Cheat */ emit_arith(cs, PFS_OP_MAD, dest, mask & WRITEMASK_XW, src[1], pfs_one, pfs_zero, flags); } break; case OPCODE_EX2: src[0] = t_scalar_src(cs, fpi->SrcReg[0]); emit_arith(cs, PFS_OP_EX2, dest, mask, src[0], undef, undef, flags); break; case OPCODE_FRC: src[0] = t_src(cs, fpi->SrcReg[0]); emit_arith(cs, PFS_OP_FRC, dest, mask, src[0], undef, undef, flags); break; case OPCODE_KIL: emit_tex(cs, fpi, R300_TEX_OP_KIL); break; case OPCODE_LG2: src[0] = t_scalar_src(cs, fpi->SrcReg[0]); emit_arith(cs, PFS_OP_LG2, dest, mask, src[0], undef, undef, flags); break; case OPCODE_LIT: src[0] = t_src(cs, fpi->SrcReg[0]); emit_lit(cs, dest, mask, src[0], flags); break; case OPCODE_LRP: src[0] = t_src(cs, fpi->SrcReg[0]); src[1] = t_src(cs, fpi->SrcReg[1]); src[2] = t_src(cs, fpi->SrcReg[2]); /* result = tmp0tmp1 + (1 - tmp0)tmp2 * = tmp0tmp1 + tmp2 + (-tmp0)tmp2 * MAD temp, -tmp0, tmp2, tmp2 * MAD result, tmp0, tmp1, temp */ temp[0] = get_temp_reg(cs); emit_arith(cs, PFS_OP_MAD, temp[0], mask, negate(keep(src[0])), keep(src[2]), src[2], 0); emit_arith(cs, PFS_OP_MAD, dest, mask, src[0], src[1], temp[0], flags); free_temp(cs, temp[0]); break; case OPCODE_MAD: src[0] = t_src(cs, fpi->SrcReg[0]); src[1] = t_src(cs, fpi->SrcReg[1]); src[2] = t_src(cs, fpi->SrcReg[2]); emit_arith(cs, PFS_OP_MAD, dest, mask, src[0], src[1], src[2], flags); break; case OPCODE_MAX: src[0] = t_src(cs, fpi->SrcReg[0]); src[1] = t_src(cs, fpi->SrcReg[1]); emit_arith(cs, PFS_OP_MAX, dest, mask, src[0], src[1], undef, flags); break; case OPCODE_MIN: src[0] = t_src(cs, fpi->SrcReg[0]); src[1] = t_src(cs, fpi->SrcReg[1]); emit_arith(cs, PFS_OP_MIN, dest, mask, src[0], src[1], undef, flags); break; case OPCODE_MOV: src[0] = t_src(cs, fpi->SrcReg[0]); emit_arith(cs, PFS_OP_MAD, dest, mask, src[0], pfs_one, pfs_zero, flags); break; case OPCODE_MUL: src[0] = t_src(cs, fpi->SrcReg[0]); src[1] = t_src(cs, fpi->SrcReg[1]); emit_arith(cs, PFS_OP_MAD, dest, mask, src[0], src[1], pfs_zero, flags); break; case OPCODE_RCP: src[0] = t_scalar_src(cs, fpi->SrcReg[0]); emit_arith(cs, PFS_OP_RCP, dest, mask, src[0], undef, undef, flags); break; case OPCODE_RSQ: src[0] = t_scalar_src(cs, fpi->SrcReg[0]); emit_arith(cs, PFS_OP_RSQ, dest, mask, absolute(src[0]), pfs_zero, pfs_zero, flags); break; case OPCODE_SCS: /* * scs using a parabola : * scs(x): * result.x = sin(-abs(x)+0.5*PI) (cos) * result.y = sin(x) (sin) * */ temp[0] = get_temp_reg(cs); temp[1] = get_temp_reg(cs); const_sin[0] = emit_const4fv(cs, SinCosConsts[0]); const_sin[1] = emit_const4fv(cs, SinCosConsts[1]); src[0] = t_scalar_src(cs, fpi->SrcReg[0]); /* x = -abs(x)+0.5*PI */ emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(const_sin[0], Z, Z, Z, Z), //PI pfs_half, negate(abs (swizzle(keep(src[0]), X, X, X, X))), 0); /* C*x (sin) */ emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_W, swizzle(const_sin[0], Y, Y, Y, Y), swizzle(keep(src[0]), X, X, X, X), pfs_zero, 0); /* B*x, C*x (cos) */ emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0], Z, Z, Z, Z), const_sin[0], pfs_zero, 0); /* B*x (sin) */ emit_arith(cs, PFS_OP_MAD, temp[1], WRITEMASK_W, swizzle(const_sin[0], X, X, X, X), keep(src[0]), pfs_zero, 0); /* y = B*x + C*x*abs(x) (sin) */ emit_arith(cs, PFS_OP_MAD, temp[1], WRITEMASK_Z, absolute(src[0]), swizzle(temp[0], W, W, W, W), swizzle(temp[1], W, W, W, W), 0); /* y = B*x + C*x*abs(x) (cos) */ emit_arith(cs, PFS_OP_MAD, temp[1], WRITEMASK_W, swizzle(temp[0], Y, Y, Y, Y), absolute(swizzle(temp[0], Z, Z, Z, Z)), swizzle(temp[0], X, X, X, X), 0); /* y*abs(y) - y (cos), y*abs(y) - y (sin) */ emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X | WRITEMASK_Y, swizzle(temp[1], W, Z, Y, X), absolute(swizzle(temp[1], W, Z, Y, X)), negate(swizzle(temp[1], W, Z, Y, X)), 0); /* dest.xy = mad(temp.xy, P, temp2.wz) */ emit_arith(cs, PFS_OP_MAD, dest, mask & (WRITEMASK_X | WRITEMASK_Y), temp[0], swizzle(const_sin[0], W, W, W, W), swizzle(temp[1], W, Z, Y, X), flags); free_temp(cs, temp[0]); free_temp(cs, temp[1]); break; case OPCODE_SIN: /* * using a parabola: * sin(x) = 4/pi * x + -4/(pi*pi) * x * abs(x) * extra precision is obtained by weighting against * itself squared. */ temp[0] = get_temp_reg(cs); const_sin[0] = emit_const4fv(cs, SinCosConsts[0]); const_sin[1] = emit_const4fv(cs, SinCosConsts[1]); src[0] = t_scalar_src(cs, fpi->SrcReg[0]); /* do range reduction */ emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X, swizzle(keep(src[0]), X, X, X, X), swizzle(const_sin[1], Z, Z, Z, Z), pfs_half, 0); emit_arith(cs, PFS_OP_FRC, temp[0], WRITEMASK_X, swizzle(temp[0], X, X, X, X), undef, undef, 0); emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(temp[0], X, X, X, X), swizzle(const_sin[1], W, W, W, W), //2*PI negate(swizzle(const_sin[0], Z, Z, Z, Z)), //PI 0); /* SIN */ emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0], Z, Z, Z, Z), const_sin[0], pfs_zero, 0); emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X, swizzle(temp[0], Y, Y, Y, Y), absolute(swizzle(temp[0], Z, Z, Z, Z)), swizzle(temp[0], X, X, X, X), 0); emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_Y, swizzle(temp[0], X, X, X, X), absolute(swizzle(temp[0], X, X, X, X)), negate(swizzle(temp[0], X, X, X, X)), 0); emit_arith(cs, PFS_OP_MAD, dest, mask, swizzle(temp[0], Y, Y, Y, Y), swizzle(const_sin[0], W, W, W, W), swizzle(temp[0], X, X, X, X), flags); free_temp(cs, temp[0]); break; case OPCODE_TEX: emit_tex(cs, fpi, R300_TEX_OP_LD); break; case OPCODE_TXB: emit_tex(cs, fpi, R300_TEX_OP_TXB); break; case OPCODE_TXP: emit_tex(cs, fpi, R300_TEX_OP_TXP); break; default: ERROR("unknown fpi->Opcode %d\n", fpi->Opcode); break; } } static GLboolean parse_program(struct r300_pfs_compile_state *cs) { COMPILE_STATE; int clauseidx; for (clauseidx = 0; clauseidx < cs->compiler->compiler.NumClauses; ++clauseidx) { struct radeon_clause* clause = &cs->compiler->compiler.Clauses[clauseidx]; int ip; for(ip = 0; ip < clause->NumInstructions; ++ip) { emit_instruction(cs, clause->Instructions + ip); if (fp->error) return GL_FALSE; } } return GL_TRUE; } /* - Init structures * - Determine what hwregs each input corresponds to */ static void init_program(struct r300_pfs_compile_state *cs) { COMPILE_STATE; struct gl_fragment_program *mp = &fp->mesa_program; GLuint InputsRead = mp->Base.InputsRead; GLuint temps_used = 0; /* for fp->temps[] */ int i, j; /* New compile, reset tracking data */ fp->optimization = driQueryOptioni(&cs->compiler->r300->radeon.optionCache, "fp_optimization"); fp->translated = GL_FALSE; fp->error = GL_FALSE; fp->WritesDepth = GL_FALSE; code->tex.length = 0; code->cur_node = 0; code->first_node_has_tex = 0; code->const_nr = 0; code->max_temp_idx = 0; code->node[0].alu_end = -1; code->node[0].tex_end = -1; for (i = 0; i < PFS_MAX_ALU_INST; i++) { for (j = 0; j < 3; j++) { cs->slot[i].vsrc[j] = SRC_CONST; cs->slot[i].ssrc[j] = SRC_CONST; } } /* Work out what temps the Mesa inputs correspond to, this must match * what setup_rs_unit does, which shouldn't be a problem as rs_unit * configures itself based on the fragprog's InputsRead * * NOTE: this depends on get_hw_temp() allocating registers in order, * starting from register 0. */ /* Texcoords come first */ for (i = 0; i < cs->compiler->r300->radeon.glCtx->Const.MaxTextureUnits; i++) { if (InputsRead & (FRAG_BIT_TEX0 << i)) { cs->inputs[FRAG_ATTRIB_TEX0 + i].refcount = 0; cs->inputs[FRAG_ATTRIB_TEX0 + i].reg = get_hw_temp(cs, 0); } } InputsRead &= ~FRAG_BITS_TEX_ANY; /* fragment position treated as a texcoord */ if (InputsRead & FRAG_BIT_WPOS) { cs->inputs[FRAG_ATTRIB_WPOS].refcount = 0; cs->inputs[FRAG_ATTRIB_WPOS].reg = get_hw_temp(cs, 0); } InputsRead &= ~FRAG_BIT_WPOS; /* Then primary colour */ if (InputsRead & FRAG_BIT_COL0) { cs->inputs[FRAG_ATTRIB_COL0].refcount = 0; cs->inputs[FRAG_ATTRIB_COL0].reg = get_hw_temp(cs, 0); } InputsRead &= ~FRAG_BIT_COL0; /* Secondary color */ if (InputsRead & FRAG_BIT_COL1) { cs->inputs[FRAG_ATTRIB_COL1].refcount = 0; cs->inputs[FRAG_ATTRIB_COL1].reg = get_hw_temp(cs, 0); } InputsRead &= ~FRAG_BIT_COL1; /* Anything else */ if (InputsRead) { WARN_ONCE("Don't know how to handle inputs 0x%x\n", InputsRead); /* force read from hwreg 0 for now */ for (i = 0; i < 32; i++) if (InputsRead & (1 << i)) cs->inputs[i].reg = 0; } /* Pre-parse the program, grabbing refcounts on input/temp regs. * That way, we can free up the reg when it's no longer needed */ for (i = 0; i < cs->compiler->compiler.Clauses[0].NumInstructions; ++i) { struct prog_instruction *fpi = cs->compiler->compiler.Clauses[0].Instructions + i; int idx; for (j = 0; j < 3; j++) { idx = fpi->SrcReg[j].Index; switch (fpi->SrcReg[j].File) { case PROGRAM_TEMPORARY: if (!(temps_used & (1 << idx))) { cs->temps[idx].reg = -1; cs->temps[idx].refcount = 1; temps_used |= (1 << idx); } else cs->temps[idx].refcount++; break; case PROGRAM_INPUT: cs->inputs[idx].refcount++; break; default: break; } } idx = fpi->DstReg.Index; if (fpi->DstReg.File == PROGRAM_TEMPORARY) { if (!(temps_used & (1 << idx))) { cs->temps[idx].reg = -1; cs->temps[idx].refcount = 1; temps_used |= (1 << idx); } else cs->temps[idx].refcount++; } } cs->temp_in_use = temps_used; } /** * Final compilation step: Turn the intermediate radeon_program into * machine-readable instructions. */ GLboolean r300FragmentProgramEmit(struct r300_fragment_program_compiler *compiler) { struct r300_pfs_compile_state cs; struct r300_fragment_program_code *code = compiler->code; _mesa_memset(&cs, 0, sizeof(cs)); cs.compiler = compiler; init_program(&cs); if (!parse_program(&cs)) return GL_FALSE; /* Finish off */ code->node[code->cur_node].alu_end = cs.nrslots - code->node[code->cur_node].alu_offset - 1; if (code->node[code->cur_node].tex_end < 0) code->node[code->cur_node].tex_end = 0; code->alu_offset = 0; code->alu_end = cs.nrslots - 1; code->tex_offset = 0; code->tex_end = code->tex.length ? code->tex.length - 1 : 0; assert(code->node[code->cur_node].alu_end >= 0); assert(code->alu_end >= 0); return GL_TRUE; }