From 2d2f49c91952e18f3362346e19b45c72b1f6db32 Mon Sep 17 00:00:00 2001 From: Andre Maasikas Date: Wed, 23 Sep 2009 14:20:59 +0300 Subject: r600: add support for CUBE textures, also TXP seems to work here ... --- src/mesa/drivers/dri/r600/r700_assembler.c | 306 ++++++++++++++++++++++++----- src/mesa/drivers/dri/r600/r700_assembler.h | 4 + 2 files changed, 263 insertions(+), 47 deletions(-) (limited to 'src/mesa/drivers/dri') diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c index 81269350e4..dc25f3b418 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.c +++ b/src/mesa/drivers/dri/r600/r700_assembler.c @@ -213,7 +213,7 @@ GLboolean is_reduction_opcode(PVSDWORD* dest) { if (dest->dst.op3 == 0) { - if ( (dest->dst.opcode == SQ_OP2_INST_DOT4 || dest->dst.opcode == SQ_OP2_INST_DOT4_IEEE) ) + if ( (dest->dst.opcode == SQ_OP2_INST_DOT4 || dest->dst.opcode == SQ_OP2_INST_DOT4_IEEE || dest->dst.opcode == SQ_OP2_INST_CUBE) ) { return GL_TRUE; } @@ -350,6 +350,7 @@ unsigned int r700GetNumOperands(r700_AssemblerBase* pAsm) case SQ_OP2_INST_PRED_SETNE: case SQ_OP2_INST_DOT4: case SQ_OP2_INST_DOT4_IEEE: + case SQ_OP2_INST_CUBE: return 2; case SQ_OP2_INST_MOV: @@ -469,6 +470,9 @@ int Init_r700_AssemblerBase(SHADER_PIPE_TYPE spt, r700_AssemblerBase* pAsm, R700 pAsm->number_of_inputs = 0; + pAsm->is_tex = GL_FALSE; + pAsm->need_tex_barrier = GL_FALSE; + return 0; } @@ -682,7 +686,7 @@ GLboolean add_tex_instruction(r700_AssemblerBase* pAsm, // If this clause constains any TEX instruction that is dependent on a previous instruction, // set the barrier bit - if( pAsm->pInstDeps[pAsm->uiCurInst].nDstDep > (-1) ) + if( pAsm->pInstDeps[pAsm->uiCurInst].nDstDep > (-1) || pAsm->need_tex_barrier == GL_TRUE ) { pAsm->cf_current_tex_clause_ptr->m_Word1.f.barrier = 0x1; } @@ -1279,42 +1283,48 @@ GLboolean tex_src(r700_AssemblerBase *pAsm) GLboolean bValidTexCoord = GL_FALSE; + if(pAsm->aArgSubst[1] >= 0) + { + bValidTexCoord = GL_TRUE; + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = pAsm->aArgSubst[1]; + } + else + { switch (pILInst->SrcReg[0].File) { - case PROGRAM_CONSTANT: - case PROGRAM_LOCAL_PARAM: - case PROGRAM_ENV_PARAM: - case PROGRAM_STATE_VAR: - bValidTexCoord = GL_TRUE; - setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); - pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; - pAsm->S[0].src.reg = pAsm->aArgSubst[1]; - break; - case PROGRAM_TEMPORARY: - bValidTexCoord = GL_TRUE; - pAsm->S[0].src.reg = pILInst->SrcReg[0].Index + - pAsm->starting_temp_register_number; - pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; - break; - case PROGRAM_INPUT: - switch (pILInst->SrcReg[0].Index) - { - case FRAG_ATTRIB_COL0: - case FRAG_ATTRIB_COL1: - case FRAG_ATTRIB_TEX0: - case FRAG_ATTRIB_TEX1: - case FRAG_ATTRIB_TEX2: - case FRAG_ATTRIB_TEX3: - case FRAG_ATTRIB_TEX4: - case FRAG_ATTRIB_TEX5: - case FRAG_ATTRIB_TEX6: - case FRAG_ATTRIB_TEX7: - bValidTexCoord = GL_TRUE; - pAsm->S[0].src.reg = - pAsm->uiFP_AttributeMap[pILInst->SrcReg[0].Index]; - pAsm->S[0].src.rtype = SRC_REG_INPUT; - break; - } - break; + case PROGRAM_CONSTANT: + case PROGRAM_LOCAL_PARAM: + case PROGRAM_ENV_PARAM: + case PROGRAM_STATE_VAR: + break; + case PROGRAM_TEMPORARY: + bValidTexCoord = GL_TRUE; + pAsm->S[0].src.reg = pILInst->SrcReg[0].Index + + pAsm->starting_temp_register_number; + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + break; + case PROGRAM_INPUT: + switch (pILInst->SrcReg[0].Index) + { + case FRAG_ATTRIB_COL0: + case FRAG_ATTRIB_COL1: + case FRAG_ATTRIB_TEX0: + case FRAG_ATTRIB_TEX1: + case FRAG_ATTRIB_TEX2: + case FRAG_ATTRIB_TEX3: + case FRAG_ATTRIB_TEX4: + case FRAG_ATTRIB_TEX5: + case FRAG_ATTRIB_TEX6: + case FRAG_ATTRIB_TEX7: + bValidTexCoord = GL_TRUE; + pAsm->S[0].src.reg = + pAsm->uiFP_AttributeMap[pILInst->SrcReg[0].Index]; + pAsm->S[0].src.rtype = SRC_REG_INPUT; + break; + } + break; + } } if(GL_TRUE == bValidTexCoord) @@ -2082,7 +2092,9 @@ GLboolean assemble_alu_instruction(r700_AssemblerBase *pAsm) { is_single_scalar_operation = GL_FALSE; number_of_scalar_operations = 4; - + +/* current assembler doesn't do more than 1 register per source */ +#if 0 /* check read port, only very preliminary algorithm, not count in src0/1 same comp case and prev slot repeat case; also not count relative addressing. TODO: improve performance. */ @@ -2117,6 +2129,7 @@ GLboolean assemble_alu_instruction(r700_AssemblerBase *pAsm) { bSplitInst = GL_TRUE; } +#endif } contiguous_slots_needed = 0; @@ -2337,9 +2350,7 @@ GLboolean next_ins(r700_AssemblerBase *pAsm) { struct prog_instruction *pILInst = &(pAsm->pILInst[pAsm->uiCurInst]); - if( GL_TRUE == IsTex(pILInst->Opcode) && - /* handle const moves to temp register */ - !(pAsm->D.dst.opcode == SQ_OP2_INST_MOV) ) + if( GL_TRUE == pAsm->is_tex ) { if (pILInst->TexSrcTarget == TEXTURE_RECT_INDEX) { if( GL_FALSE == assemble_tex_instruction(pAsm, GL_FALSE) ) @@ -2383,7 +2394,8 @@ GLboolean next_ins(r700_AssemblerBase *pAsm) pAsm->S[0].bits = 0; pAsm->S[1].bits = 0; pAsm->S[2].bits = 0; - + pAsm->is_tex = GL_FALSE; + pAsm->need_tex_barrier = GL_FALSE; return GL_TRUE; } @@ -3506,7 +3518,10 @@ GLboolean assemble_STP(r700_AssemblerBase *pAsm) GLboolean assemble_TEX(r700_AssemblerBase *pAsm) { GLboolean src_const; + GLboolean need_barrier = GL_FALSE; + checkop1(pAsm); + switch (pAsm->pILInst[pAsm->uiCurInst].SrcReg[0].File) { case PROGRAM_CONSTANT: @@ -3526,20 +3541,18 @@ GLboolean assemble_TEX(r700_AssemblerBase *pAsm) { if ( GL_FALSE == mov_temp(pAsm, 0) ) return GL_FALSE; + need_barrier = GL_TRUE; } switch (pAsm->pILInst[pAsm->uiCurInst].Opcode) { case OPCODE_TEX: - pAsm->D.dst.opcode = SQ_TEX_INST_SAMPLE; break; case OPCODE_TXB: radeon_error("do not support TXB yet\n"); return GL_FALSE; break; case OPCODE_TXP: - /* TODO : tex proj version : divid first 3 components by 4th */ - pAsm->D.dst.opcode = SQ_TEX_INST_SAMPLE; break; default: radeon_error("Internal error: bad texture op (not TEX)\n"); @@ -3547,6 +3560,190 @@ GLboolean assemble_TEX(r700_AssemblerBase *pAsm) break; } + if (pAsm->pILInst[pAsm->uiCurInst].Opcode == OPCODE_TXP) + { + GLuint tmp = gethelpr(pAsm); + pAsm->D.dst.opcode = SQ_OP2_INST_RECIP_IEEE; + pAsm->D.dst.math = 1; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp; + pAsm->D.dst.writew = 1; + + if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + { + return GL_FALSE; + } + swizzleagain_PVSSRC(&(pAsm->S[0].src), SQ_SEL_W, SQ_SEL_W, SQ_SEL_W, SQ_SEL_W); + if( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + pAsm->D.dst.opcode = SQ_OP2_INST_MUL; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp; + pAsm->D.dst.writex = 1; + pAsm->D.dst.writey = 1; + pAsm->D.dst.writez = 1; + pAsm->D.dst.writew = 0; + + if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + { + return GL_FALSE; + } + setaddrmode_PVSSRC(&(pAsm->S[1].src), ADDR_ABSOLUTE); + pAsm->S[1].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[1].src.reg = tmp; + setswizzle_PVSSRC(&(pAsm->S[1].src), SQ_SEL_W); + + if( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + pAsm->aArgSubst[1] = tmp; + need_barrier = GL_TRUE; + } + + if (pAsm->pILInst[pAsm->uiCurInst].TexSrcTarget == TEXTURE_CUBE_INDEX ) + { + GLuint tmp1 = gethelpr(pAsm); + GLuint tmp2 = gethelpr(pAsm); + + /* tmp1.xyzw = CUBE(R0.zzxy, R0.yxzz) */ + pAsm->D.dst.opcode = SQ_OP2_INST_CUBE; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp1; + nomask_PVSDST(&(pAsm->D.dst)); + + if( GL_FALSE == assemble_src(pAsm, 0, -1) ) + { + return GL_FALSE; + } + + if( GL_FALSE == assemble_src(pAsm, 0, 1) ) + { + return GL_FALSE; + } + + swizzleagain_PVSSRC(&(pAsm->S[0].src), SQ_SEL_Z, SQ_SEL_Z, SQ_SEL_X, SQ_SEL_Y); + swizzleagain_PVSSRC(&(pAsm->S[1].src), SQ_SEL_Y, SQ_SEL_X, SQ_SEL_Z, SQ_SEL_Z); + + if( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + /* tmp1.z = ABS(tmp1.z) dont have abs support in assembler currently + * have to do explicit instruction + */ + pAsm->D.dst.opcode = SQ_OP2_INST_MAX; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp1; + pAsm->D.dst.writez = 1; + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp1; + noswizzle_PVSSRC(&(pAsm->S[0].src)); + pAsm->S[1].bits = pAsm->S[0].bits; + flipneg_PVSSRC(&(pAsm->S[1].src)); + + next_ins(pAsm); + + /* tmp1.z = RCP_e(|tmp1.z|) */ + pAsm->D.dst.opcode = SQ_OP2_INST_RECIP_IEEE; + pAsm->D.dst.math = 1; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp1; + pAsm->D.dst.writez = 1; + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp1; + pAsm->S[0].src.swizzlex = SQ_SEL_Z; + + next_ins(pAsm); + + /* MULADD R0.x, R0.x, PS1, (0x3FC00000, 1.5f).x + * MULADD R0.y, R0.y, PS1, (0x3FC00000, 1.5f).x + * muladd has no writemask, have to use another temp + * also no support for imm constants, so add 1 here + */ + pAsm->D.dst.opcode = SQ_OP3_INST_MULADD; + pAsm->D.dst.op3 = 1; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp2; + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp1; + noswizzle_PVSSRC(&(pAsm->S[0].src)); + setaddrmode_PVSSRC(&(pAsm->S[1].src), ADDR_ABSOLUTE); + pAsm->S[1].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[1].src.reg = tmp1; + setswizzle_PVSSRC(&(pAsm->S[1].src), SQ_SEL_Z); + setaddrmode_PVSSRC(&(pAsm->S[2].src), ADDR_ABSOLUTE); + pAsm->S[2].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[2].src.reg = tmp1; + setswizzle_PVSSRC(&(pAsm->S[2].src), SQ_SEL_1); + + next_ins(pAsm); + + /* ADD the remaining .5 */ + pAsm->D.dst.opcode = SQ_OP2_INST_ADD; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp2; + pAsm->D.dst.writex = 1; + pAsm->D.dst.writey = 1; + pAsm->D.dst.writez = 0; + pAsm->D.dst.writew = 0; + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp2; + noswizzle_PVSSRC(&(pAsm->S[0].src)); + setaddrmode_PVSSRC(&(pAsm->S[1].src), ADDR_ABSOLUTE); + pAsm->S[1].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[1].src.reg = 252; // SQ_ALU_SRC_0_5 + noswizzle_PVSSRC(&(pAsm->S[1].src)); + + next_ins(pAsm); + + /* tmp1.xy = temp2.xy */ + pAsm->D.dst.opcode = SQ_OP2_INST_MOV; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp1; + pAsm->D.dst.writex = 1; + pAsm->D.dst.writey = 1; + pAsm->D.dst.writez = 0; + pAsm->D.dst.writew = 0; + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp2; + noswizzle_PVSSRC(&(pAsm->S[0].src)); + + next_ins(pAsm); + pAsm->aArgSubst[1] = tmp1; + need_barrier = GL_TRUE; + + } + + pAsm->D.dst.opcode = SQ_TEX_INST_SAMPLE; + pAsm->is_tex = GL_TRUE; + if ( GL_TRUE == need_barrier ) + { + pAsm->need_tex_barrier = GL_TRUE; + } // Set src1 to tex unit id pAsm->S[1].src.reg = pAsm->pILInst[pAsm->uiCurInst].TexSrcUnit; pAsm->S[1].src.rtype = SRC_REG_TEMPORARY; @@ -3567,10 +3764,25 @@ GLboolean assemble_TEX(r700_AssemblerBase *pAsm) return GL_FALSE; } - if ( GL_FALSE == next_ins(pAsm) ) + if(pAsm->pILInst[pAsm->uiCurInst].Opcode == OPCODE_TXP) { - return GL_FALSE; + /* hopefully did swizzles before */ + noswizzle_PVSSRC(&(pAsm->S[0].src)); } + + if(pAsm->pILInst[pAsm->uiCurInst].TexSrcTarget == TEXTURE_CUBE_INDEX) + { + /* SAMPLE dst, tmp.yxwy, CUBE */ + pAsm->S[0].src.swizzlex = SQ_SEL_Y; + pAsm->S[0].src.swizzley = SQ_SEL_X; + pAsm->S[0].src.swizzlez = SQ_SEL_W; + pAsm->S[0].src.swizzlew = SQ_SEL_Y; + } + + if ( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } return GL_TRUE; } diff --git a/src/mesa/drivers/dri/r600/r700_assembler.h b/src/mesa/drivers/dri/r600/r700_assembler.h index 4e6e20011a..bd6df94ff9 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.h +++ b/src/mesa/drivers/dri/r600/r700_assembler.h @@ -374,6 +374,10 @@ typedef struct r700_AssemblerBase struct prog_instruction * pILInst; GLuint uiCurInst; GLboolean bR6xx; + /* helper to decide which type of instruction to assemble */ + GLboolean is_tex; + /* we inserted helper intructions and need barrier on next TEX ins */ + GLboolean need_tex_barrier; } r700_AssemblerBase; //Internal use -- cgit v1.2.3