fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-07 07:08:04 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

aco: flush denormals for 16-bit fmin/fmax on GFX8

Browse source 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26445>
This commit is contained in:
Rhys Perry 2023-11-28 17:57:25 +00:00 • committed by Marge Bot
parent a9a2eabea8
commit b7725b072b
1 changed files with 9 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
src/amd/compiler/aco_instruction_selection.cpp
View file

@ -879,8 +879,11 @@ emit_vop2_instruction(isel_context* ctx, nir_alu_instr* instr, aco_opcode opc, T
if (flush_denorms && ctx->program->gfx_level < GFX9) { if (flush_denorms && ctx->program->gfx_level < GFX9) {
assert(dst.size() == 1); assert(dst.size() == 1);
Temp tmp = bld.vop2(opc, bld.def(v1), op[0], op[1]); Temp tmp = bld.vop2(opc, bld.def(dst.regClass()), op[0], op[1]);
bld.vop2(aco_opcode::v_mul_f32, Definition(dst), Operand::c32(0x3f800000u), tmp); if (dst.bytes() == 2)
bld.vop2(aco_opcode::v_mul_f16, Definition(dst), Operand::c16(0x3c00), tmp);
else
bld.vop2(aco_opcode::v_mul_f32, Definition(dst), Operand::c32(0x3f800000u), tmp);
} else { } else {
if (nuw) { if (nuw) {
bld.nuw().vop2(opc, Definition(dst), op[0], op[1]); bld.nuw().vop2(opc, Definition(dst), op[0], op[1]);
@ -2447,8 +2450,8 @@ visit_alu_instr(isel_context* ctx, nir_alu_instr* instr)
} }
case nir_op_fmax: { case nir_op_fmax: {
if (dst.regClass() == v2b) { if (dst.regClass() == v2b) {
// TODO: check fp_mode.must_flush_denorms16_64 emit_vop2_instruction(ctx, instr, aco_opcode::v_max_f16, dst, true, false,
emit_vop2_instruction(ctx, instr, aco_opcode::v_max_f16, dst, true); ctx->block->fp_mode.must_flush_denorms16_64);
} else if (dst.regClass() == v1 && instr->def.bit_size == 16) { } else if (dst.regClass() == v1 && instr->def.bit_size == 16) {
emit_vop3p_instruction(ctx, instr, aco_opcode::v_pk_max_f16, dst); emit_vop3p_instruction(ctx, instr, aco_opcode::v_pk_max_f16, dst);
} else if (dst.regClass() == v1) { } else if (dst.regClass() == v1) {
@ -2464,8 +2467,8 @@ visit_alu_instr(isel_context* ctx, nir_alu_instr* instr)
} }
case nir_op_fmin: { case nir_op_fmin: {
if (dst.regClass() == v2b) { if (dst.regClass() == v2b) {
// TODO: check fp_mode.must_flush_denorms16_64 emit_vop2_instruction(ctx, instr, aco_opcode::v_min_f16, dst, true, false,
emit_vop2_instruction(ctx, instr, aco_opcode::v_min_f16, dst, true); ctx->block->fp_mode.must_flush_denorms16_64);
} else if (dst.regClass() == v1 && instr->def.bit_size == 16) { } else if (dst.regClass() == v1 && instr->def.bit_size == 16) {
emit_vop3p_instruction(ctx, instr, aco_opcode::v_pk_min_f16, dst, true); emit_vop3p_instruction(ctx, instr, aco_opcode::v_pk_min_f16, dst, true);
} else if (dst.regClass() == v1) { } else if (dst.regClass() == v1) {