fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-05 05:18:08 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

r600/sfn: support nir_op_mulz and legazy math rules

Browse source 
v2: Handle nir_op_ffmaz as well (Georg Lehmann)

Closes: #6390
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17076>
This commit is contained in:
Gert Wollny 2022-06-16 09:37:56 +02:00 • committed by Marge Bot
parent 9437cbb76d
commit 983385d319
4 changed files with 22 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
src/gallium/drivers/r600/r600_pipe_common.c
View file

@ -1367,7 +1367,6 @@ bool r600_common_screen_init(struct r600_common_screen *rscreen,
rscreen->nir_options = nir_options;
if (rscreen->info.family < CHIP_CEDAR)
rscreen->nir_options.force_indirect_unrolling_sampler = true;
@ -1411,7 +1410,9 @@ bool r600_common_screen_init(struct r600_common_screen *rscreen,
/* TGSI's ifind is reversed from ours, keep it the TGSI way. */
rscreen->nir_options.lower_find_msb_to_reverse = false;
}
} else {
rscreen->nir_options.has_fmulz = true;
}
rscreen->nir_options_fs = rscreen->nir_options;
rscreen->nir_options_fs.lower_all_io_to_temps = true;

18
src/gallium/drivers/r600/sfn/sfn_instr_alu.cpp
View file

@ -1263,7 +1263,13 @@ bool AluInstr::from_nir(nir_alu_instr *alu, Shader& shader)
case nir_op_flt: return emit_alu_op2(*alu, op2_setgt_dx10, shader, op2_opt_reverse);
case nir_op_fmax: return emit_alu_op2(*alu, op2_max_dx10, shader);
case nir_op_fmin: return emit_alu_op2(*alu, op2_min_dx10, shader);
case nir_op_fmul: return emit_alu_op2(*alu, op2_mul_ieee, shader);
case nir_op_fmul:
if (!shader.has_flag(Shader::sh_legacy_math_rules))
return emit_alu_op2(*alu, op2_mul_ieee, shader);
FALLTHROUGH;
case nir_op_fmulz: return emit_alu_op2(*alu, op2_mul, shader);
case nir_op_fneg: return emit_alu_op1(*alu, op1_mov, shader, {1 << alu_src0_neg});
case nir_op_fneu32: return emit_alu_op2(*alu, op2_setne_dx10, shader);
case nir_op_fneu: return emit_alu_op2(*alu, op2_setne_dx10, shader);
@ -1319,8 +1325,12 @@ bool AluInstr::from_nir(nir_alu_instr *alu, Shader& shader)
case nir_op_unpack_half_2x16_split_x: return emit_unpack_32_2x16_split_x(*alu, shader);
case nir_op_unpack_half_2x16_split_y: return emit_unpack_32_2x16_split_y(*alu, shader);
case nir_op_ffma:
if (!shader.has_flag(Shader::sh_legacy_math_rules))
return emit_alu_op3(*alu, op3_muladd_ieee, shader);
FALLTHROUGH;
case nir_op_ffmaz: return emit_alu_op3(*alu, op3_muladd, shader);
case nir_op_ffma: return emit_alu_op3(*alu, op3_muladd_ieee, shader);
case nir_op_mov: return emit_alu_op1(*alu, op1_mov, shader);
case nir_op_f2i32: return emit_alu_op1(*alu, op1_flt_to_int, shader);
case nir_op_vec2: return emit_create_vec(*alu, 2, shader);
@ -2034,7 +2044,9 @@ static bool emit_dot(const nir_alu_instr& alu, int n, Shader& shader)
srcs[2 * i + 1] = value_factory.zero();
}
AluInstr *ir = new AluInstr(op2_dot4_ieee, dest, srcs, AluInstr::last_write, 4);
auto op = unlikely(shader.has_flag(Shader::sh_legacy_math_rules)) ?
op2_dot4 : op2_dot4_ieee;
AluInstr *ir = new AluInstr(op, dest, srcs, AluInstr::last_write, 4);
if (src0.negate) ir->set_alu_flag(alu_src0_neg);
if (src0.abs) ir->set_alu_flag(alu_src0_abs);

3
src/gallium/drivers/r600/sfn/sfn_shader.cpp
View file

@ -430,6 +430,9 @@ bool Shader::process(nir_shader *nir)
{
m_ssbo_image_offset = nir->info.num_images;
if (nir->info.use_legacy_math_rules)
set_flag(sh_legacy_math_rules);
nir_foreach_uniform_variable(var, nir)
scan_uniforms(var);

1
src/gallium/drivers/r600/sfn/sfn_shader.h
View file

@ -199,6 +199,7 @@ public:
sh_txs_cube_array_comp,
sh_indirect_atomic,
sh_mem_barrier,
sh_legacy_math_rules,
sh_flags_count
};