From 2cd466f6c3192015ea1794afc57eb453d7f13818 Mon Sep 17 00:00:00 2001 From: Francisco Jerez Date: Sat, 18 Jul 2015 17:09:37 +0300 Subject: i965/fs: Implement lowering of logical texturing opcodes on Gen4. Unlike its Gen5 and Gen7 counterparts this patch isn't a plain refactor of the previous Gen4 texturing code, it's more of a rewrite largely based on emit_texture_gen4_simd16(). The reason is that on the one hand the original emit_texture_gen4() code didn't seem easily fixable to be SIMD width-invariant and had plenty of clutter to support SIMD-width workarounds which are no longer required. On the other hand emit_texture_gen4_simd16() was missing a number of SIMD8-only opcodes. This should generalize both and roughly match their current behaviour where there is overlap. Incidentally this will fix the following piglits on Gen4: arb_shader_texture_lod.execution.arb_shader_texture_lod-texgrad arb_shader_texture_lod.execution.tex-miplevel-selection *gradarb 2d arb_shader_texture_lod.execution.tex-miplevel-selection *gradarb 3d arb_shader_texture_lod.execution.tex-miplevel-selection *projgradarb 2d arb_shader_texture_lod.execution.tex-miplevel-selection *projgradarb 2d_projvec4 arb_shader_texture_lod.execution.tex-miplevel-selection *projgradarb 3d Acked-by: Jason Ekstrand Reviewed-by: Kenneth Graunke --- src/mesa/drivers/dri/i965/brw_fs.cpp | 108 ++++++++++++++++++++++++++++++++++- 1 file changed, 107 insertions(+), 1 deletion(-) diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp index f36933d485d..77d4a7d7b66 100644 --- a/src/mesa/drivers/dri/i965/brw_fs.cpp +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp @@ -3386,6 +3386,110 @@ lower_fb_write_logical_send(const fs_builder &bld, fs_inst *inst, inst->header_size = header_size; } +static void +lower_sampler_logical_send_gen4(const fs_builder &bld, fs_inst *inst, opcode op, + const fs_reg &coordinate, + const fs_reg &shadow_c, + const fs_reg &lod, const fs_reg &lod2, + const fs_reg &sampler, + unsigned coord_components, + unsigned grad_components) +{ + const bool has_lod = (op == SHADER_OPCODE_TXL || op == FS_OPCODE_TXB || + op == SHADER_OPCODE_TXF || op == SHADER_OPCODE_TXS); + fs_reg msg_begin(MRF, 1, BRW_REGISTER_TYPE_F); + fs_reg msg_end = msg_begin; + + /* g0 header. */ + msg_end = offset(msg_end, bld.group(8, 0), 1); + + for (unsigned i = 0; i < coord_components; i++) + bld.MOV(retype(offset(msg_end, bld, i), coordinate.type), + offset(coordinate, bld, i)); + + msg_end = offset(msg_end, bld, coord_components); + + /* Messages other than SAMPLE and RESINFO in SIMD16 and TXD in SIMD8 + * require all three components to be present and zero if they are unused. + */ + if (coord_components > 0 && + (has_lod || shadow_c.file != BAD_FILE || + (op == SHADER_OPCODE_TEX && bld.dispatch_width() == 8))) { + for (unsigned i = coord_components; i < 3; i++) + bld.MOV(offset(msg_end, bld, i), fs_reg(0.0f)); + + msg_end = offset(msg_end, bld, 3 - coord_components); + } + + if (op == SHADER_OPCODE_TXD) { + /* TXD unsupported in SIMD16 mode. */ + assert(bld.dispatch_width() == 8); + + /* the slots for u and v are always present, but r is optional */ + if (coord_components < 2) + msg_end = offset(msg_end, bld, 2 - coord_components); + + /* P = u, v, r + * dPdx = dudx, dvdx, drdx + * dPdy = dudy, dvdy, drdy + * + * 1-arg: Does not exist. + * + * 2-arg: dudx dvdx dudy dvdy + * dPdx.x dPdx.y dPdy.x dPdy.y + * m4 m5 m6 m7 + * + * 3-arg: dudx dvdx drdx dudy dvdy drdy + * dPdx.x dPdx.y dPdx.z dPdy.x dPdy.y dPdy.z + * m5 m6 m7 m8 m9 m10 + */ + for (unsigned i = 0; i < grad_components; i++) + bld.MOV(offset(msg_end, bld, i), offset(lod, bld, i)); + + msg_end = offset(msg_end, bld, MAX2(grad_components, 2)); + + for (unsigned i = 0; i < grad_components; i++) + bld.MOV(offset(msg_end, bld, i), offset(lod2, bld, i)); + + msg_end = offset(msg_end, bld, MAX2(grad_components, 2)); + } + + if (has_lod) { + /* Bias/LOD with shadow comparitor is unsupported in SIMD16 -- *Without* + * shadow comparitor (including RESINFO) it's unsupported in SIMD8 mode. + */ + assert(shadow_c.file != BAD_FILE ? bld.dispatch_width() == 8 : + bld.dispatch_width() == 16); + + const brw_reg_type type = + (op == SHADER_OPCODE_TXF || op == SHADER_OPCODE_TXS ? + BRW_REGISTER_TYPE_UD : BRW_REGISTER_TYPE_F); + bld.MOV(retype(msg_end, type), lod); + msg_end = offset(msg_end, bld, 1); + } + + if (shadow_c.file != BAD_FILE) { + if (op == SHADER_OPCODE_TEX && bld.dispatch_width() == 8) { + /* There's no plain shadow compare message, so we use shadow + * compare with a bias of 0.0. + */ + bld.MOV(msg_end, fs_reg(0.0f)); + msg_end = offset(msg_end, bld, 1); + } + + bld.MOV(msg_end, shadow_c); + msg_end = offset(msg_end, bld, 1); + } + + inst->opcode = op; + inst->src[0] = reg_undef; + inst->src[1] = sampler; + inst->resize_sources(2); + inst->base_mrf = msg_begin.reg; + inst->mlen = msg_end.reg - msg_begin.reg; + inst->header_size = 1; +} + static void lower_sampler_logical_send_gen5(const fs_builder &bld, fs_inst *inst, opcode op, fs_reg coordinate, @@ -3725,7 +3829,9 @@ lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst, opcode op) sampler, offset_value, coord_components, grad_components); } else { - assert(!"Not implemented"); + lower_sampler_logical_send_gen4(bld, inst, op, coordinate, + shadow_c, lod, lod2, sampler, + coord_components, grad_components); } } -- cgit v1.2.3