Module: Mesa Branch: main Commit: 1a1c81987da573ec3e8e625f9312a7bd04998481 URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=1a1c81987da573ec3e8e625f9312a7bd04998481
Author: M Henning <dra...@darkrefraction.com> Date: Thu Nov 23 13:55:53 2023 -0500 nak: Call nir_opt_load_store_vectorize Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26347> --- src/nouveau/compiler/nak_nir.c | 31 +++++++++++++++++++++++++++++++ 1 file changed, 31 insertions(+) diff --git a/src/nouveau/compiler/nak_nir.c b/src/nouveau/compiler/nak_nir.c index bbb355d1c38..74ed2c2c38a 100644 --- a/src/nouveau/compiler/nak_nir.c +++ b/src/nouveau/compiler/nak_nir.c @@ -965,6 +965,27 @@ nak_nir_lower_fs_outputs(nir_shader *nir) return true; } +static bool +nak_mem_vectorize_cb(unsigned align_mul, unsigned align_offset, + unsigned bit_size, unsigned num_components, + nir_intrinsic_instr *low, nir_intrinsic_instr *high, + void *cb_data) +{ + /* + * Since we legalize these later with nir_lower_mem_access_bit_sizes, + * we can optimistically combine anything that might be profitable + */ + assert(util_is_power_of_two_nonzero(align_mul)); + + unsigned max_bytes = 128u / 8u; + if (low->intrinsic == nir_intrinsic_load_ubo) + max_bytes = 64u / 8u; + + align_mul = MIN2(align_mul, max_bytes); + align_offset = align_offset % align_mul; + return align_offset + num_components * (bit_size / 8) <= align_mul; +} + static nir_mem_access_size_align nak_mem_access_size_align(nir_intrinsic_op intrin, uint8_t bytes, uint8_t bit_size, @@ -1049,8 +1070,18 @@ nak_postprocess_nir(nir_shader *nir, glsl_get_natural_size_align_bytes); OPT(nir, nir_lower_explicit_io, nir_var_function_temp, nir_address_format_32bit_offset); + nak_optimize_nir(nir, nak); } + nir_load_store_vectorize_options vectorize_opts = {}; + vectorize_opts.modes = nir_var_mem_global | + nir_var_mem_ssbo | + nir_var_mem_shared | + nir_var_shader_temp; + vectorize_opts.callback = nak_mem_vectorize_cb; + vectorize_opts.robust_modes = robust2_modes; + OPT(nir, nir_opt_load_store_vectorize, &vectorize_opts); + nir_lower_mem_access_bit_sizes_options mem_bit_size_options = { .modes = nir_var_mem_constant | nir_var_mem_ubo | nir_var_mem_generic, .callback = nak_mem_access_size_align,
