Module: Mesa Branch: main Commit: d57f5c9cac8dce2460b246405d2176e1e2a1ba05 URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=d57f5c9cac8dce2460b246405d2176e1e2a1ba05
Author: Georg Lehmann <[email protected]> Date: Tue Oct 4 15:44:16 2022 +0200 radv,aco: Lower uclz in NIR. Signed-off-by: Georg Lehmann <[email protected]> Reviewed-by: Rhys Perry <[email protected]> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18951> --- src/amd/compiler/aco_instruction_selection.cpp | 13 ------------- src/amd/vulkan/radv_shader.c | 1 + 2 files changed, 1 insertion(+), 13 deletions(-) diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp index 74287e1609f..3a4268af5d4 100644 --- a/src/amd/compiler/aco_instruction_selection.cpp +++ b/src/amd/compiler/aco_instruction_selection.cpp @@ -1838,19 +1838,6 @@ visit_alu_instr(isel_context* ctx, nir_alu_instr* instr) } break; } - case nir_op_uclz: { - Temp src = get_alu_src(ctx, instr->src[0]); - if (src.regClass() == s1) { - Temp msb_rev = bld.sop1(aco_opcode::s_flbit_i32_b32, bld.def(s1), src); - bld.sop2(aco_opcode::s_min_u32, Definition(dst), Operand::c32(32u), msb_rev); - } else if (src.regClass() == v1) { - Temp msb_rev = bld.vop1(aco_opcode::v_ffbh_u32, bld.def(v1), src); - bld.vop2(aco_opcode::v_min_u32, Definition(dst), Operand::c32(32u), msb_rev); - } else { - isel_err(&instr->instr, "Unimplemented NIR instr bit size"); - } - break; - } case nir_op_bitfield_reverse: { if (dst.regClass() == s1) { bld.sop1(aco_opcode::s_brev_b32, Definition(dst), get_alu_src(ctx, instr->src[0])); diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c index 21f465de94e..81daff18988 100644 --- a/src/amd/vulkan/radv_shader.c +++ b/src/amd/vulkan/radv_shader.c @@ -91,6 +91,7 @@ get_nir_options_for_stage(struct radv_physical_device *device, gl_shader_stage s .lower_iadd_sat = device->rad_info.gfx_level <= GFX8, .lower_hadd = true, .lower_mul_32x16 = true, + .lower_uclz = true, .has_fsub = true, .has_isub = true, .has_sdot_4x8 = device->rad_info.has_accelerated_dot_product,
