[Mesa-dev] [PATCH v2 2/2] radv/query: Use 1-bit booleans in query shaders
Bas Nieuwenhuizen
bas at basnieuwenhuizen.nl
Wed Dec 19 22:08:45 UTC 2018
Reviewed-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
On Wed, Dec 19, 2018 at 11:05 PM Jason Ekstrand <jason at jlekstrand.net> wrote:
>
> Fixes: 44227453ec03f "nir: Switch to using 1-bit Booleans for almost..."
> Reviewed-by: Rhys Perry <pendingchaos02 at gmail.com>
> ---
> src/amd/vulkan/radv_query.c | 42 ++++++++++++++++++-------------------
> 1 file changed, 21 insertions(+), 21 deletions(-)
>
> diff --git a/src/amd/vulkan/radv_query.c b/src/amd/vulkan/radv_query.c
> index 6e88de691d6..89e70224630 100644
> --- a/src/amd/vulkan/radv_query.c
> +++ b/src/amd/vulkan/radv_query.c
> @@ -54,7 +54,7 @@ static unsigned get_max_db(struct radv_device *device)
>
> static nir_ssa_def *nir_flag_set(nir_builder *b, nir_ssa_def *flags, uint32_t flag)
> {
> - return nir_iand(b, flags, nir_imm_int(b, flag));
> + return nir_i2b(b, nir_iand(b, flags, nir_imm_int(b, flag)));
> }
>
> static void radv_break_on_count(nir_builder *b, nir_variable *var, nir_ssa_def *count)
> @@ -138,7 +138,7 @@ build_occlusion_query_shader(struct radv_device *device) {
> nir_variable *outer_counter = nir_local_variable_create(b.impl, glsl_int_type(), "outer_counter");
> nir_variable *start = nir_local_variable_create(b.impl, glsl_uint64_t_type(), "start");
> nir_variable *end = nir_local_variable_create(b.impl, glsl_uint64_t_type(), "end");
> - nir_variable *available = nir_local_variable_create(b.impl, glsl_int_type(), "available");
> + nir_variable *available = nir_local_variable_create(b.impl, glsl_bool_type(), "available");
> unsigned db_count = get_max_db(device);
>
> nir_ssa_def *flags = radv_load_push_int(&b, 0, "flags");
> @@ -176,7 +176,7 @@ build_occlusion_query_shader(struct radv_device *device) {
>
> nir_store_var(&b, result, nir_imm_int64(&b, 0), 0x1);
> nir_store_var(&b, outer_counter, nir_imm_int(&b, 0), 0x1);
> - nir_store_var(&b, available, nir_imm_int(&b, 1), 0x1);
> + nir_store_var(&b, available, nir_imm_true(&b), 0x1);
>
> nir_loop *outer_loop = nir_loop_create(b.shader);
> nir_builder_cf_insert(&b, &outer_loop->cf_node);
> @@ -214,14 +214,14 @@ build_occlusion_query_shader(struct radv_device *device) {
>
> b.cursor = nir_after_cf_list(&update_if->else_list);
>
> - nir_store_var(&b, available, nir_imm_int(&b, 0), 0x1);
> + nir_store_var(&b, available, nir_imm_false(&b), 0x1);
>
> b.cursor = nir_after_cf_node(&outer_loop->cf_node);
>
> /* Store the result if complete or if partial results have been requested. */
>
> nir_ssa_def *result_is_64bit = nir_flag_set(&b, flags, VK_QUERY_RESULT_64_BIT);
> - nir_ssa_def *result_size = nir_b32csel(&b, result_is_64bit, nir_imm_int(&b, 8), nir_imm_int(&b, 4));
> + nir_ssa_def *result_size = nir_bcsel(&b, result_is_64bit, nir_imm_int(&b, 8), nir_imm_int(&b, 4));
>
> nir_if *store_if = nir_if_create(b.shader);
> store_if->condition = nir_src_for_ssa(nir_ior(&b, nir_flag_set(&b, flags, VK_QUERY_RESULT_PARTIAL_BIT), nir_load_var(&b, available)));
> @@ -264,7 +264,7 @@ build_occlusion_query_shader(struct radv_device *device) {
> b.cursor = nir_after_cf_list(&availability_if->then_list);
>
> store = nir_intrinsic_instr_create(b.shader, nir_intrinsic_store_ssbo);
> - store->src[0] = nir_src_for_ssa(nir_load_var(&b, available));
> + store->src[0] = nir_src_for_ssa(nir_b2i32(&b, nir_load_var(&b, available)));
> store->src[1] = nir_src_for_ssa(&dst_buf->dest.ssa);
> store->src[2] = nir_src_for_ssa(nir_iadd(&b, result_size, output_base));
> nir_intrinsic_set_write_mask(store, 0x1);
> @@ -296,11 +296,11 @@ build_pipeline_statistics_query_shader(struct radv_device *device) {
> * uint64_t dst_offset = dst_base;
> * uint32_t elem_size = flags & VK_QUERY_RESULT_64_BIT ? 8 : 4;
> * uint32_t elem_count = stats_mask >> 16;
> - * uint32_t available = src_buf[avail_offset + 4 * global_id.x];
> + * uint32_t available32 = src_buf[avail_offset + 4 * global_id.x];
> * if (flags & VK_QUERY_RESULT_WITH_AVAILABILITY_BIT) {
> - * dst_buf[dst_offset + elem_count * elem_size] = available;
> + * dst_buf[dst_offset + elem_count * elem_size] = available32;
> * }
> - * if (available) {
> + * if ((bool)available32) {
> * // repeat 11 times:
> * if (stats_mask & (1 << 0)) {
> * uint64_t start = src_buf[src_offset + 8 * indices[0]];
> @@ -372,10 +372,10 @@ build_pipeline_statistics_query_shader(struct radv_device *device) {
> nir_ssa_dest_init(&load->instr, &load->dest, 1, 32, NULL);
> load->num_components = 1;
> nir_builder_instr_insert(&b, &load->instr);
> - nir_ssa_def *available = &load->dest.ssa;
> + nir_ssa_def *available32 = &load->dest.ssa;
>
> nir_ssa_def *result_is_64bit = nir_flag_set(&b, flags, VK_QUERY_RESULT_64_BIT);
> - nir_ssa_def *elem_size = nir_b32csel(&b, result_is_64bit, nir_imm_int(&b, 8), nir_imm_int(&b, 4));
> + nir_ssa_def *elem_size = nir_bcsel(&b, result_is_64bit, nir_imm_int(&b, 8), nir_imm_int(&b, 4));
> nir_ssa_def *elem_count = nir_ushr(&b, stats_mask, nir_imm_int(&b, 16));
>
> /* Store the availability bit if requested. */
> @@ -387,7 +387,7 @@ build_pipeline_statistics_query_shader(struct radv_device *device) {
> b.cursor = nir_after_cf_list(&availability_if->then_list);
>
> nir_intrinsic_instr *store = nir_intrinsic_instr_create(b.shader, nir_intrinsic_store_ssbo);
> - store->src[0] = nir_src_for_ssa(available);
> + store->src[0] = nir_src_for_ssa(available32);
> store->src[1] = nir_src_for_ssa(&dst_buf->dest.ssa);
> store->src[2] = nir_src_for_ssa(nir_iadd(&b, output_base, nir_imul(&b, elem_count, elem_size)));
> nir_intrinsic_set_write_mask(store, 0x1);
> @@ -397,7 +397,7 @@ build_pipeline_statistics_query_shader(struct radv_device *device) {
> b.cursor = nir_after_cf_node(&availability_if->cf_node);
>
> nir_if *available_if = nir_if_create(b.shader);
> - available_if->condition = nir_src_for_ssa(available);
> + available_if->condition = nir_src_for_ssa(nir_i2b(&b, available32));
> nir_cf_node_insert(b.cursor, &available_if->cf_node);
>
> b.cursor = nir_after_cf_list(&available_if->then_list);
> @@ -566,12 +566,12 @@ build_tfb_query_shader(struct radv_device *device)
> glsl_vector_type(GLSL_TYPE_UINT64, 2),
> "result");
> nir_variable *available =
> - nir_local_variable_create(b.impl, glsl_int_type(), "available");
> + nir_local_variable_create(b.impl, glsl_bool_type(), "available");
>
> nir_store_var(&b, result,
> nir_vec2(&b, nir_imm_int64(&b, 0),
> nir_imm_int64(&b, 0)), 0x3);
> - nir_store_var(&b, available, nir_imm_int(&b, 0), 0x1);
> + nir_store_var(&b, available, nir_imm_false(&b), 0x1);
>
> nir_ssa_def *flags = radv_load_push_int(&b, 0, "flags");
>
> @@ -630,8 +630,8 @@ build_tfb_query_shader(struct radv_device *device)
> avails[1] = nir_iand(&b, nir_channel(&b, &load2->dest.ssa, 1),
> nir_channel(&b, &load2->dest.ssa, 3));
> nir_ssa_def *result_is_available =
> - nir_iand(&b, nir_iand(&b, avails[0], avails[1]),
> - nir_imm_int(&b, 0x80000000));
> + nir_i2b(&b, nir_iand(&b, nir_iand(&b, avails[0], avails[1]),
> + nir_imm_int(&b, 0x80000000)));
>
> /* Only compute result if available. */
> nir_if *available_if = nir_if_create(b.shader);
> @@ -664,7 +664,7 @@ build_tfb_query_shader(struct radv_device *device)
> nir_store_var(&b, result,
> nir_vec2(&b, num_primitive_written,
> primitive_storage_needed), 0x3);
> - nir_store_var(&b, available, nir_imm_int(&b, 1), 0x1);
> + nir_store_var(&b, available, nir_imm_true(&b), 0x1);
>
> b.cursor = nir_after_cf_node(&available_if->cf_node);
>
> @@ -672,8 +672,8 @@ build_tfb_query_shader(struct radv_device *device)
> nir_ssa_def *result_is_64bit =
> nir_flag_set(&b, flags, VK_QUERY_RESULT_64_BIT);
> nir_ssa_def *result_size =
> - nir_b32csel(&b, result_is_64bit, nir_imm_int(&b, 16),
> - nir_imm_int(&b, 8));
> + nir_bcsel(&b, result_is_64bit, nir_imm_int(&b, 16),
> + nir_imm_int(&b, 8));
>
> /* Store the result if complete or partial results have been requested. */
> nir_if *store_if = nir_if_create(b.shader);
> @@ -722,7 +722,7 @@ build_tfb_query_shader(struct radv_device *device)
> b.cursor = nir_after_cf_list(&availability_if->then_list);
>
> store = nir_intrinsic_instr_create(b.shader, nir_intrinsic_store_ssbo);
> - store->src[0] = nir_src_for_ssa(nir_load_var(&b, available));
> + store->src[0] = nir_src_for_ssa(nir_b2i32(&b, nir_load_var(&b, available)));
> store->src[1] = nir_src_for_ssa(&dst_buf->dest.ssa);
> store->src[2] = nir_src_for_ssa(nir_iadd(&b, result_size, output_base));
> nir_intrinsic_set_write_mask(store, 0x1);
> --
> 2.19.2
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
More information about the mesa-dev
mailing list