/************************************************************************** * * Copyright 2003 VMware, Inc. * 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, sub license, 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 NON-INFRINGEMENT. * IN NO EVENT SHALL VMWARE 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. * **************************************************************************/ #include #include "main/glheader.h" #include "main/macros.h" #include "main/enums.h" #include "tnl/t_context.h" #include "intel_batchbuffer.h" #include "i915_reg.h" #include "i915_context.h" #include "i915_program.h" #define A0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT) #define D0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT) #define T0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT) #define A0_SRC0( reg ) (((reg)&UREG_MASK)>>UREG_A0_SRC0_SHIFT_LEFT) #define A1_SRC0( reg ) (((reg)&UREG_MASK)<>UREG_A1_SRC1_SHIFT_LEFT) #define A2_SRC1( reg ) (((reg)&UREG_MASK)<>UREG_A2_SRC2_SHIFT_LEFT) /* These are special, and don't have swizzle/negate bits. */ #define T0_SAMPLER( reg ) (GET_UREG_NR(reg)<temp_flag); if (!bit) { fprintf(stderr, "%s: out of temporaries\n", __FILE__); exit(1); } p->temp_flag |= 1 << (bit - 1); return UREG(REG_TYPE_R, (bit - 1)); } GLuint i915_get_utemp(struct i915_fragment_program * p) { int bit = ffs(~p->utemp_flag); if (!bit) { fprintf(stderr, "%s: out of temporaries\n", __FILE__); exit(1); } p->utemp_flag |= 1 << (bit - 1); return UREG(REG_TYPE_U, (bit - 1)); } void i915_release_utemps(struct i915_fragment_program *p) { p->utemp_flag = ~0x7; } GLuint i915_emit_decl(struct i915_fragment_program *p, GLuint type, GLuint nr, GLuint d0_flags) { GLuint reg = UREG(type, nr); if (type == REG_TYPE_T) { if (p->decl_t & (1 << nr)) return reg; p->decl_t |= (1 << nr); } else if (type == REG_TYPE_S) { if (p->decl_s & (1 << nr)) return reg; p->decl_s |= (1 << nr); } else return reg; *(p->decl++) = (D0_DCL | D0_DEST(reg) | d0_flags); *(p->decl++) = D1_MBZ; *(p->decl++) = D2_MBZ; assert(p->decl <= p->declarations + ARRAY_SIZE(p->declarations)); p->nr_decl_insn++; return reg; } GLuint i915_emit_arith(struct i915_fragment_program * p, GLuint op, GLuint dest, GLuint mask, GLuint saturate, GLuint src0, GLuint src1, GLuint src2) { GLuint c[3]; GLuint nr_const = 0; assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST); dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)); assert(dest); if (GET_UREG_TYPE(src0) == REG_TYPE_CONST) c[nr_const++] = 0; if (GET_UREG_TYPE(src1) == REG_TYPE_CONST) c[nr_const++] = 1; if (GET_UREG_TYPE(src2) == REG_TYPE_CONST) c[nr_const++] = 2; /* Recursively call this function to MOV additional const values * into temporary registers. Use utemp registers for this - * currently shouldn't be possible to run out, but keep an eye on * this. */ if (nr_const > 1) { GLuint s[3], first, i, old_utemp_flag; s[0] = src0; s[1] = src1; s[2] = src2; old_utemp_flag = p->utemp_flag; first = GET_UREG_NR(s[c[0]]); for (i = 1; i < nr_const; i++) { if (GET_UREG_NR(s[c[i]]) != first) { GLuint tmp = i915_get_utemp(p); i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0, s[c[i]], 0, 0); s[c[i]] = tmp; } } src0 = s[0]; src1 = s[1]; src2 = s[2]; p->utemp_flag = old_utemp_flag; /* restore */ } if (p->csr >= p->program + ARRAY_SIZE(p->program)) { i915_program_error(p, "Program contains too many instructions"); return UREG_BAD; } *(p->csr++) = (op | A0_DEST(dest) | mask | saturate | A0_SRC0(src0)); *(p->csr++) = (A1_SRC0(src0) | A1_SRC1(src1)); *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2)); if (GET_UREG_TYPE(dest) == REG_TYPE_R) p->register_phases[GET_UREG_NR(dest)] = p->nr_tex_indirect; p->nr_alu_insn++; return dest; } static GLuint get_free_rreg (struct i915_fragment_program *p, GLuint live_regs) { int bit = ffs(~live_regs); if (!bit) { i915_program_error(p, "Can't find free R reg"); return UREG_BAD; } return UREG(REG_TYPE_R, bit - 1); } GLuint i915_emit_texld( struct i915_fragment_program *p, GLuint live_regs, GLuint dest, GLuint destmask, GLuint sampler, GLuint coord, GLuint op ) { if (coord != UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord))) { /* With the help of the "needed registers" table created earlier, pick * a register we can MOV the swizzled TC to (since TEX doesn't support * swizzled sources) */ GLuint swizCoord = get_free_rreg(p, live_regs); if (swizCoord == UREG_BAD) return 0; i915_emit_arith( p, A0_MOV, swizCoord, A0_DEST_CHANNEL_ALL, 0, coord, 0, 0 ); coord = swizCoord; } /* Don't worry about saturate as we only support texture formats * that are always in the 0..1 range. */ if (destmask != A0_DEST_CHANNEL_ALL) { GLuint tmp = i915_get_utemp(p); i915_emit_texld( p, 0, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op ); i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 ); return dest; } else { assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST); assert(dest == UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest))); /* Can't use unsaved temps for coords, as the phase boundary would result * in the contents becoming undefined. */ assert(GET_UREG_TYPE(coord) != REG_TYPE_U); if ((GET_UREG_TYPE(coord) != REG_TYPE_R) && (GET_UREG_TYPE(coord) != REG_TYPE_OC) && (GET_UREG_TYPE(coord) != REG_TYPE_OD) && (GET_UREG_TYPE(coord) != REG_TYPE_T)) { GLuint tmpCoord = get_free_rreg(p, live_regs); if (tmpCoord == UREG_BAD) return 0; i915_emit_arith(p, A0_MOV, tmpCoord, A0_DEST_CHANNEL_ALL, 0, coord, 0, 0); coord = tmpCoord; } /* Output register being oC or oD defines a phase boundary */ if (GET_UREG_TYPE(dest) == REG_TYPE_OC || GET_UREG_TYPE(dest) == REG_TYPE_OD) p->nr_tex_indirect++; /* Reading from an r# register whose contents depend on output of the * current phase defines a phase boundary. */ if (GET_UREG_TYPE(coord) == REG_TYPE_R && p->register_phases[GET_UREG_NR(coord)] == p->nr_tex_indirect) p->nr_tex_indirect++; if (p->csr >= p->program + ARRAY_SIZE(p->program)) { i915_program_error(p, "Program contains too many instructions"); return UREG_BAD; } *(p->csr++) = (op | T0_DEST( dest ) | T0_SAMPLER( sampler )); *(p->csr++) = T1_ADDRESS_REG( coord ); *(p->csr++) = T2_MBZ; if (GET_UREG_TYPE(dest) == REG_TYPE_R) p->register_phases[GET_UREG_NR(dest)] = p->nr_tex_indirect; p->nr_tex_insn++; return dest; } } GLuint i915_emit_const1f(struct i915_fragment_program * p, GLfloat c0) { GLint reg, idx; if (c0 == 0.0) return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO); if (c0 == 1.0) return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE); for (reg = 0; reg < I915_MAX_CONSTANT; reg++) { if (p->constant_flags[reg] == I915_CONSTFLAG_PARAM) continue; for (idx = 0; idx < 4; idx++) { if (!(p->constant_flags[reg] & (1 << idx)) || p->constant[reg][idx] == c0) { p->constant[reg][idx] = c0; p->constant_flags[reg] |= 1 << idx; if (reg + 1 > p->nr_constants) p->nr_constants = reg + 1; return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE); } } } fprintf(stderr, "%s: out of constants\n", __func__); p->error = 1; return 0; } GLuint i915_emit_const2f(struct i915_fragment_program * p, GLfloat c0, GLfloat c1) { GLint reg, idx; if (c0 == 0.0) return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W); if (c0 == 1.0) return swizzle(i915_emit_const1f(p, c1), ONE, X, Z, W); if (c1 == 0.0) return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W); if (c1 == 1.0) return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W); for (reg = 0; reg < I915_MAX_CONSTANT; reg++) { if (p->constant_flags[reg] == 0xf || p->constant_flags[reg] == I915_CONSTFLAG_PARAM) continue; for (idx = 0; idx < 3; idx++) { if (!(p->constant_flags[reg] & (3 << idx))) { p->constant[reg][idx] = c0; p->constant[reg][idx + 1] = c1; p->constant_flags[reg] |= 3 << idx; if (reg + 1 > p->nr_constants) p->nr_constants = reg + 1; return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO, ONE); } } } fprintf(stderr, "%s: out of constants\n", __func__); p->error = 1; return 0; } GLuint i915_emit_const4f(struct i915_fragment_program * p, GLfloat c0, GLfloat c1, GLfloat c2, GLfloat c3) { GLint reg; for (reg = 0; reg < I915_MAX_CONSTANT; reg++) { if (p->constant_flags[reg] == 0xf && p->constant[reg][0] == c0 && p->constant[reg][1] == c1 && p->constant[reg][2] == c2 && p->constant[reg][3] == c3) { return UREG(REG_TYPE_CONST, reg); } else if (p->constant_flags[reg] == 0) { p->constant[reg][0] = c0; p->constant[reg][1] = c1; p->constant[reg][2] = c2; p->constant[reg][3] = c3; p->constant_flags[reg] = 0xf; if (reg + 1 > p->nr_constants) p->nr_constants = reg + 1; return UREG(REG_TYPE_CONST, reg); } } fprintf(stderr, "%s: out of constants\n", __func__); p->error = 1; return 0; } GLuint i915_emit_const4fv(struct i915_fragment_program * p, const GLfloat * c) { return i915_emit_const4f(p, c[0], c[1], c[2], c[3]); } GLuint i915_emit_param4fv(struct i915_fragment_program * p, const GLfloat * values) { GLint reg, i; for (i = 0; i < p->nr_params; i++) { if (p->param[i].values == values) return UREG(REG_TYPE_CONST, p->param[i].reg); } for (reg = 0; reg < I915_MAX_CONSTANT; reg++) { if (p->constant_flags[reg] == 0) { p->constant_flags[reg] = I915_CONSTFLAG_PARAM; i = p->nr_params++; p->param[i].values = values; p->param[i].reg = reg; p->params_uptodate = 0; if (reg + 1 > p->nr_constants) p->nr_constants = reg + 1; return UREG(REG_TYPE_CONST, reg); } } fprintf(stderr, "%s: out of constants\n", __func__); p->error = 1; return 0; } /* Warning the user about program errors seems to be quite valuable, from * our bug reports. It unfortunately means piglit reporting errors * when we fall back to software due to an unsupportable program, though. */ void i915_program_error(struct i915_fragment_program *p, const char *fmt, ...) { if (unlikely((INTEL_DEBUG & (DEBUG_WM | DEBUG_PERF)) != 0)) { va_list args; fprintf(stderr, "i915_program_error: "); va_start(args, fmt); vfprintf(stderr, fmt, args); va_end(args); fprintf(stderr, "\n"); } p->error = 1; } void i915_init_program(struct i915_context *i915, struct i915_fragment_program *p) { struct gl_context *ctx = &i915->intel.ctx; p->translated = 0; p->params_uptodate = 0; p->on_hardware = 0; p->error = 0; memset(&p->register_phases, 0, sizeof(p->register_phases)); p->nr_tex_indirect = 1; p->nr_tex_insn = 0; p->nr_alu_insn = 0; p->nr_decl_insn = 0; p->ctx = ctx; memset(p->constant_flags, 0, sizeof(p->constant_flags)); p->nr_constants = 0; p->csr = p->program; p->decl = p->declarations; p->decl_s = 0; p->decl_t = 0; p->temp_flag = 0xffff000; p->utemp_flag = ~0x7; p->wpos_tex = -1; p->depth_written = 0; p->nr_params = 0; *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM; } void i915_fini_program(struct i915_fragment_program *p) { GLuint program_size = p->csr - p->program; GLuint decl_size = p->decl - p->declarations; if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT) { i915_program_error(p, "Exceeded max nr indirect texture lookups " "(%d out of %d)", p->nr_tex_indirect, I915_MAX_TEX_INDIRECT); } if (p->nr_tex_insn > I915_MAX_TEX_INSN) { i915_program_error(p, "Exceeded max TEX instructions (%d out of %d)", p->nr_tex_insn, I915_MAX_TEX_INSN); } if (p->nr_alu_insn > I915_MAX_ALU_INSN) i915_program_error(p, "Exceeded max ALU instructions (%d out of %d)", p->nr_alu_insn, I915_MAX_ALU_INSN); if (p->nr_decl_insn > I915_MAX_DECL_INSN) { i915_program_error(p, "Exceeded max DECL instructions (%d out of %d)", p->nr_decl_insn, I915_MAX_DECL_INSN); } if (p->error) { p->FragProg.NumNativeInstructions = 0; p->FragProg.NumNativeAluInstructions = 0; p->FragProg.NumNativeTexInstructions = 0; p->FragProg.NumNativeTexIndirections = 0; } else { p->FragProg.NumNativeInstructions = (p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn); p->FragProg.NumNativeAluInstructions = p->nr_alu_insn; p->FragProg.NumNativeTexInstructions = p->nr_tex_insn; p->FragProg.NumNativeTexIndirections = p->nr_tex_indirect; } p->declarations[0] |= program_size + decl_size - 2; } void i915_upload_program(struct i915_context *i915, struct i915_fragment_program *p) { GLuint program_size = p->csr - p->program; GLuint decl_size = p->decl - p->declarations; if (p->error) return; /* Could just go straight to the batchbuffer from here: */ if (i915->state.ProgramSize != (program_size + decl_size) || memcmp(i915->state.Program + decl_size, p->program, program_size * sizeof(int)) != 0) { I915_STATECHANGE(i915, I915_UPLOAD_PROGRAM); memcpy(i915->state.Program, p->declarations, decl_size * sizeof(int)); memcpy(i915->state.Program + decl_size, p->program, program_size * sizeof(int)); i915->state.ProgramSize = decl_size + program_size; } /* Always seemed to get a failure if I used memcmp() to * shortcircuit this state upload. Needs further investigation? */ if (p->nr_constants) { GLuint nr = p->nr_constants; I915_ACTIVESTATE(i915, I915_UPLOAD_CONSTANTS, 1); I915_STATECHANGE(i915, I915_UPLOAD_CONSTANTS); i915->state.Constant[0] = _3DSTATE_PIXEL_SHADER_CONSTANTS | ((nr) * 4); i915->state.Constant[1] = (1 << (nr - 1)) | ((1 << (nr - 1)) - 1); memcpy(&i915->state.Constant[2], p->constant, 4 * sizeof(int) * (nr)); i915->state.ConstantSize = 2 + (nr) * 4; if (0) { GLuint i; for (i = 0; i < nr; i++) { fprintf(stderr, "const[%d]: %f %f %f %f\n", i, p->constant[i][0], p->constant[i][1], p->constant[i][2], p->constant[i][3]); } } } else { I915_ACTIVESTATE(i915, I915_UPLOAD_CONSTANTS, 0); } p->on_hardware = 1; }