diff options
Diffstat (limited to 'src/gallium/auxiliary/tgsi/tgsi_exec.c')
| -rw-r--r-- | src/gallium/auxiliary/tgsi/tgsi_exec.c | 80 |
1 files changed, 63 insertions, 17 deletions
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c index f98b66dc0b..989b6eec27 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_exec.c +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c @@ -133,7 +133,7 @@ tgsi_exec_machine_bind_shader( struct tgsi_exec_machine *mach, const struct tgsi_token *tokens, uint numSamplers, - struct tgsi_sampler *samplers) + struct tgsi_sampler **samplers) { uint k; struct tgsi_parse_context parse; @@ -1045,11 +1045,28 @@ fetch_source( union tgsi_exec_channel index; uint swizzle; + /* We start with a direct index into a register file. + * + * file[1], + * where: + * file = SrcRegister.File + * [1] = SrcRegister.Index + */ index.i[0] = index.i[1] = index.i[2] = index.i[3] = reg->SrcRegister.Index; + /* There is an extra source register that indirectly subscripts + * a register file. The direct index now becomes an offset + * that is being added to the indirect register. + * + * file[ind[2].x+1], + * where: + * ind = SrcRegisterInd.File + * [2] = SrcRegisterInd.Index + * .x = SrcRegisterInd.SwizzleX + */ if (reg->SrcRegister.Indirect) { union tgsi_exec_channel index2; union tgsi_exec_channel indir_index; @@ -1086,19 +1103,31 @@ fetch_source( } } - if( reg->SrcRegister.Dimension ) { - switch( reg->SrcRegister.File ) { + /* There is an extra source register that is a second + * subscript to a register file. Effectively it means that + * the register file is actually a 2D array of registers. + * + * file[1][3] == file[1*sizeof(file[1])+3], + * where: + * [3] = SrcRegisterDim.Index + */ + if (reg->SrcRegister.Dimension) { + /* The size of the first-order array depends on the register file type. + * We need to multiply the index to the first array to get an effective, + * "flat" index that points to the beginning of the second-order array. + */ + switch (reg->SrcRegister.File) { case TGSI_FILE_INPUT: - index.i[0] *= 17; - index.i[1] *= 17; - index.i[2] *= 17; - index.i[3] *= 17; + index.i[0] *= TGSI_EXEC_MAX_INPUT_ATTRIBS; + index.i[1] *= TGSI_EXEC_MAX_INPUT_ATTRIBS; + index.i[2] *= TGSI_EXEC_MAX_INPUT_ATTRIBS; + index.i[3] *= TGSI_EXEC_MAX_INPUT_ATTRIBS; break; case TGSI_FILE_CONSTANT: - index.i[0] *= 4096; - index.i[1] *= 4096; - index.i[2] *= 4096; - index.i[3] *= 4096; + index.i[0] *= TGSI_EXEC_MAX_CONST_BUFFER; + index.i[1] *= TGSI_EXEC_MAX_CONST_BUFFER; + index.i[2] *= TGSI_EXEC_MAX_CONST_BUFFER; + index.i[3] *= TGSI_EXEC_MAX_CONST_BUFFER; break; default: assert( 0 ); @@ -1109,6 +1138,17 @@ fetch_source( index.i[2] += reg->SrcRegisterDim.Index; index.i[3] += reg->SrcRegisterDim.Index; + /* Again, the second subscript index can be addressed indirectly + * identically to the first one. + * Nothing stops us from indirectly addressing the indirect register, + * but there is no need for that, so we won't exercise it. + * + * file[1][ind[4].y+3], + * where: + * ind = SrcRegisterDimInd.File + * [4] = SrcRegisterDimInd.Index + * .y = SrcRegisterDimInd.SwizzleX + */ if (reg->SrcRegisterDim.Indirect) { union tgsi_exec_channel index2; union tgsi_exec_channel indir_index; @@ -1141,6 +1181,11 @@ fetch_source( index.i[i] = 0; } } + + /* If by any chance there was a need for a 3D array of register + * files, we would have to check whether SrcRegisterDim is followed + * by a dimension register and continue the saga. + */ } swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index ); @@ -1490,7 +1535,7 @@ exec_kilp(struct tgsi_exec_machine *mach, /* - * Fetch a texel using STR texture coordinates. + * Fetch a four texture samples using STR texture coordinates. */ static void fetch_texel( struct tgsi_sampler *sampler, @@ -1524,7 +1569,7 @@ exec_tex(struct tgsi_exec_machine *mach, boolean projected) { const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index; - union tgsi_exec_channel r[8]; + union tgsi_exec_channel r[4]; uint chan_index; float lodBias; @@ -1547,7 +1592,7 @@ exec_tex(struct tgsi_exec_machine *mach, else lodBias = 0.0; - fetch_texel(&mach->Samplers[unit], + fetch_texel(mach->Samplers[unit], &r[0], NULL, NULL, lodBias, /* S, T, P, BIAS */ &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ break; @@ -1573,7 +1618,7 @@ exec_tex(struct tgsi_exec_machine *mach, else lodBias = 0.0; - fetch_texel(&mach->Samplers[unit], + fetch_texel(mach->Samplers[unit], &r[0], &r[1], &r[2], lodBias, /* inputs */ &r[0], &r[1], &r[2], &r[3]); /* outputs */ break; @@ -1599,7 +1644,7 @@ exec_tex(struct tgsi_exec_machine *mach, else lodBias = 0.0; - fetch_texel(&mach->Samplers[unit], + fetch_texel(mach->Samplers[unit], &r[0], &r[1], &r[2], lodBias, &r[0], &r[1], &r[2], &r[3]); break; @@ -1709,6 +1754,7 @@ exec_declaration( break; default: + eval = NULL; assert( 0 ); } @@ -1751,7 +1797,7 @@ exec_instruction( case TGSI_OPCODE_ARL: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_trunc( &r[0], &r[0] ); + micro_flr( &r[0], &r[0] ); STORE( &r[0], 0, chan_index ); } break; |