diff options
author | Samuel Pitoiset <samuel.pitoiset@gmail.com> | 2020-12-10 21:47:23 +0100 |
---|---|---|
committer | Marge Bot <eric+marge@anholt.net> | 2020-12-14 16:22:39 +0000 |
commit | 77343576ebd1880d59f66d5b76befd24defe37c0 (patch) | |
tree | fa9f485a0d4ce3af7b8d3789b05c0c499eb4609a /src/amd/compiler | |
parent | 45524afe95dbbc517677807b7abb941031d4cd4d (diff) |
aco: implement a workaround for gl_FragCoord.z with VRS on GFX10.3
Without it, FragCoord.z will have the value of one of the fine pixels
instead of the center of the coarse pixel.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7837>
Diffstat (limited to 'src/amd/compiler')
-rw-r--r-- | src/amd/compiler/aco_instruction_selection.cpp | 28 | ||||
-rw-r--r-- | src/amd/compiler/aco_instruction_selection_setup.cpp | 9 |
2 files changed, 36 insertions, 1 deletions
diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp index 83eb230c034..bfb268a045f 100644 --- a/src/amd/compiler/aco_instruction_selection.cpp +++ b/src/amd/compiler/aco_instruction_selection.cpp @@ -4563,15 +4563,41 @@ void emit_interp_instr(isel_context *ctx, unsigned idx, unsigned component, Temp void emit_load_frag_coord(isel_context *ctx, Temp dst, unsigned num_components) { + Builder bld(ctx->program, ctx->block); + aco_ptr<Pseudo_instruction> vec(create_instruction<Pseudo_instruction>(aco_opcode::p_create_vector, Format::PSEUDO, num_components, 1)); for (unsigned i = 0; i < num_components; i++) vec->operands[i] = Operand(get_arg(ctx, ctx->args->ac.frag_pos[i])); if (G_0286CC_POS_W_FLOAT_ENA(ctx->program->config->spi_ps_input_ena)) { assert(num_components == 4); - Builder bld(ctx->program, ctx->block); vec->operands[3] = bld.vop1(aco_opcode::v_rcp_f32, bld.def(v1), get_arg(ctx, ctx->args->ac.frag_pos[3])); } + if (ctx->options->adjust_frag_coord_z && + G_0286CC_POS_Z_FLOAT_ENA(ctx->program->config->spi_ps_input_ena)) { + /* Adjust gl_FragCoord.z for VRS due to a hw bug on some GFX10.3 chips. */ + Operand frag_z = vec->operands[2]; + Temp adjusted_frag_z = bld.tmp(v1); + Temp tmp; + + /* dFdx fine */ + Temp tl = bld.vop1_dpp(aco_opcode::v_mov_b32, bld.def(v1), frag_z, dpp_quad_perm(0, 0, 2, 2)); + tmp = bld.vop2_dpp(aco_opcode::v_sub_f32, bld.def(v1), frag_z, tl, dpp_quad_perm(1, 1, 3, 3)); + emit_wqm(ctx, tmp, adjusted_frag_z, true); + + /* adjusted_frag_z * 0.0625 + frag_z */ + adjusted_frag_z = bld.vop3(aco_opcode::v_fma_f32, bld.def(v1), adjusted_frag_z, + Operand(0x3d800000u /* 0.0625 */), frag_z); + + /* VRS Rate X = Ancillary[2:3] */ + Temp x_rate = bld.vop3(aco_opcode::v_bfe_u32, bld.def(v1), + get_arg(ctx, ctx->args->ac.ancillary), Operand(2u), Operand(2u)); + + /* xRate = xRate == 0x1 ? adjusted_frag_z : frag_z. */ + Temp cond = bld.vopc(aco_opcode::v_cmp_eq_i32, bld.def(bld.lm), Operand(1u), Operand(x_rate)); + vec->operands[2] = bld.vop2(aco_opcode::v_cndmask_b32, bld.def(v1), frag_z, adjusted_frag_z, cond); + } + for (Operand& op : vec->operands) op = op.isUndefined() ? Operand(0u) : op; diff --git a/src/amd/compiler/aco_instruction_selection_setup.cpp b/src/amd/compiler/aco_instruction_selection_setup.cpp index 3d2dd2a4e40..f6f888429c0 100644 --- a/src/amd/compiler/aco_instruction_selection_setup.cpp +++ b/src/amd/compiler/aco_instruction_selection_setup.cpp @@ -909,6 +909,15 @@ void init_context(isel_context *ctx, nir_shader *shader) spi_ps_inputs |= S_0286CC_POS_X_FLOAT_ENA(1) << i; } + + if (ctx->options->adjust_frag_coord_z && + intrinsic->intrinsic == nir_intrinsic_load_frag_coord && + G_0286CC_POS_Z_FLOAT_ENA(spi_ps_inputs)) { + /* Enable ancillary for adjusting gl_FragCoord.z for + * VRS due to a hw bug on some GFX10.3 chips. + */ + spi_ps_inputs |= S_0286CC_ANCILLARY_ENA(1); + } break; } case nir_intrinsic_load_sample_id: |