<div dir="ltr"> <div class="gmail_quote"><div dir="ltr">On Tue, Dec 4, 2018 at 12:27 PM Karol Herbst <<a href="mailto:kherbst@redhat.com">kherbst@redhat.com</a>> wrote: </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Signed-off-by: Karol Herbst <<a href="mailto:kherbst@redhat.com" target="_blank">kherbst@redhat.com</a>> --- src/amd/vulkan/radv_meta_buffer.c | 8 ++-- src/amd/vulkan/radv_meta_bufimage.c | 28 +++++++------- src/amd/vulkan/radv_meta_clear.c | 4 +- src/amd/vulkan/radv_meta_fast_clear.c | 4 +- src/amd/vulkan/radv_meta_resolve_cs.c | 4 +- src/amd/vulkan/radv_query.c | 12 +++--- src/compiler/nir/nir_intrinsics.py | 12 +++--- src/compiler/nir/nir_lower_system_values.c | 43 +++++++++++++--------- 8 files changed, 61 insertions(+), 54 deletions(-) diff --git a/src/amd/vulkan/radv_meta_buffer.c b/src/amd/vulkan/radv_meta_buffer.c index 76854d7bbad..208988c3775 100644 --- a/src/amd/vulkan/radv_meta_buffer.c +++ b/src/amd/vulkan/radv_meta_buffer.c @@ -15,8 +15,8 @@ build_buffer_fill_shader(struct radv_device *dev) b.shader->info.cs.local_size[1] = 1; b.shader->info.cs.local_size[2] = 1; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], @@ -67,8 +67,8 @@ build_buffer_copy_shader(struct radv_device *dev) b.shader->info.cs.local_size[1] = 1; b.shader->info.cs.local_size[2] = 1; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], diff --git a/src/amd/vulkan/radv_meta_bufimage.c b/src/amd/vulkan/radv_meta_bufimage.c index 45df8438234..c8a733b3062 100644 --- a/src/amd/vulkan/radv_meta_bufimage.c +++ b/src/amd/vulkan/radv_meta_bufimage.c @@ -60,8 +60,8 @@ build_nir_itob_compute_shader(struct radv_device *dev, bool is_3d) output_img->data.descriptor_set = 0; output_img->data.binding = 1; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], @@ -289,8 +289,8 @@ build_nir_btoi_compute_shader(struct radv_device *dev, bool is_3d) output_img->data.descriptor_set = 0; output_img->data.binding = 1; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], @@ -511,8 +511,8 @@ build_nir_btoi_r32g32b32_compute_shader(struct radv_device *dev) output_img->data.descriptor_set = 0; output_img->data.binding = 1; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], @@ -719,8 +719,8 @@ build_nir_itoi_compute_shader(struct radv_device *dev, bool is_3d) output_img->data.descriptor_set = 0; output_img->data.binding = 1; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], @@ -932,8 +932,8 @@ build_nir_itoi_r32g32b32_compute_shader(struct radv_device *dev) output_img->data.descriptor_set = 0; output_img->data.binding = 1; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], @@ -1139,8 +1139,8 @@ build_nir_cleari_compute_shader(struct radv_device *dev, bool is_3d) output_img->data.descriptor_set = 0; output_img->data.binding = 0; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], @@ -1331,8 +1331,8 @@ build_nir_cleari_r32g32b32_compute_shader(struct radv_device *dev) output_img->data.descriptor_set = 0; output_img->data.binding = 0; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], diff --git a/src/amd/vulkan/radv_meta_clear.c b/src/amd/vulkan/radv_meta_clear.c index 5805d39c4b3..2e1274c929d 100644 --- a/src/amd/vulkan/radv_meta_clear.c +++ b/src/amd/vulkan/radv_meta_clear.c @@ -1025,8 +1025,8 @@ build_clear_htile_mask_shader() b.shader->info.cs.local_size[1] = 1; b.shader->info.cs.local_size[2] = 1; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], diff --git a/src/amd/vulkan/radv_meta_fast_clear.c b/src/amd/vulkan/radv_meta_fast_clear.c index f2f68961e2e..136b596aa4f 100644 --- a/src/amd/vulkan/radv_meta_fast_clear.c +++ b/src/amd/vulkan/radv_meta_fast_clear.c @@ -58,8 +58,8 @@ build_dcc_decompress_compute_shader(struct radv_device *dev) output_img->data.descriptor_set = 0; output_img->data.binding = 1; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], diff --git a/src/amd/vulkan/radv_meta_resolve_cs.c b/src/amd/vulkan/radv_meta_resolve_cs.c index e56df7f8a59..1ee8ce32ac0 100644 --- a/src/amd/vulkan/radv_meta_resolve_cs.c +++ b/src/amd/vulkan/radv_meta_resolve_cs.c @@ -99,8 +99,8 @@ build_resolve_compute_shader(struct radv_device *dev, bool is_integer, bool is_s img_type, "out_img"); output_img->data.descriptor_set = 0; output_img->data.binding = 1; - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], diff --git a/src/amd/vulkan/radv_query.c b/src/amd/vulkan/radv_query.c index d3baf2357ff..a7d53b938a7 100644 --- a/src/amd/vulkan/radv_query.c +++ b/src/amd/vulkan/radv_query.c @@ -153,8 +153,8 @@ build_occlusion_query_shader(struct radv_device *device) { nir_ssa_dest_init(&src_buf->instr, &src_buf->dest, 1, 32, NULL); nir_builder_instr_insert(&b, &src_buf->instr); - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], @@ -343,8 +343,8 @@ build_pipeline_statistics_query_shader(struct radv_device *device) { nir_ssa_dest_init(&src_buf->instr, &src_buf->dest, 1, 32, NULL); nir_builder_instr_insert(&b, &src_buf->instr); - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], @@ -590,8 +590,8 @@ build_tfb_query_shader(struct radv_device *device) nir_builder_instr_insert(&b, &src_buf->instr); /* Compute global ID. */ - nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); - nir_ssa_def *wg_id = nir_load_work_group_id(&b); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b, 32); + nir_ssa_def *wg_id = nir_load_work_group_id(&b, 32); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], diff --git a/src/compiler/nir/nir_intrinsics.py b/src/compiler/nir/nir_intrinsics.py index 830c406b450..746ddd22d92 100644 --- a/src/compiler/nir/nir_intrinsics.py +++ b/src/compiler/nir/nir_intrinsics.py @@ -479,11 +479,11 @@ system_value("tess_coord", 3) system_value("tess_level_outer", 4) system_value("tess_level_inner", 2) system_value("patch_vertices_in", 1) -system_value("local_invocation_id", 3) -system_value("local_invocation_index", 1) -system_value("work_group_id", 3) +system_value("local_invocation_id", 3, bit_sizes=[32, 64]) +system_value("local_invocation_index", 1, bit_sizes=[32, 64]) +system_value("work_group_id", 3, bit_sizes=[32, 64]) system_value("user_clip_plane", 4, indices=[UCP_ID]) -system_value("num_work_groups", 3) +system_value("num_work_groups", 3, bit_sizes=[32, 64]) system_value("helper_invocation", 1) system_value("alpha_ref_float", 1) system_value("layer_id", 1) @@ -497,8 +497,8 @@ system_value("subgroup_le_mask", 0, bit_sizes=[32, 64]) system_value("subgroup_lt_mask", 0, bit_sizes=[32, 64]) system_value("num_subgroups", 1) system_value("subgroup_id", 1) -system_value("local_group_size", 3) -system_value("global_invocation_id", 3) +system_value("local_group_size", 3, bit_sizes=[32, 64]) +system_value("global_invocation_id", 3, bit_sizes=[32, 64]) system_value("work_dim", 1) # Blend constant color values. Float values are clamped.# diff --git a/src/compiler/nir/nir_lower_system_values.c b/src/compiler/nir/nir_lower_system_values.c index 68b0ea89c8d..3cb9f224ecd 100644 --- a/src/compiler/nir/nir_lower_system_values.c +++ b/src/compiler/nir/nir_lower_system_values.c @@ -29,7 +29,7 @@ #include "nir_builder.h" static nir_ssa_def* -build_local_group_size(nir_builder *b) +build_local_group_size(nir_builder *b, unsigned bit_size) { nir_ssa_def *local_size; @@ -38,21 +38,27 @@ build_local_group_size(nir_builder *b) * point, but its intrinsic can still be used. */ if (b->shader->info.cs.local_size_variable) { - local_size = nir_load_local_group_size(b); + local_size = nir_load_local_group_size(b, bit_size); } else { nir_const_value local_size_const; memset(&local_size_const, 0, sizeof(local_size_const)); - local_size_const.u32[0] = b->shader->info.cs.local_size[0]; - local_size_const.u32[1] = b->shader->info.cs.local_size[1]; - local_size_const.u32[2] = b->shader->info.cs.local_size[2]; - local_size = nir_build_imm(b, 3, 32, local_size_const); + if (bit_size == 32) { + local_size_const.u32[0] = b->shader->info.cs.local_size[0]; + local_size_const.u32[1] = b->shader->info.cs.local_size[1]; + local_size_const.u32[2] = b->shader->info.cs.local_size[2]; + } else { </blockquote><div> </div><div>assert(bit_size == 64); </div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> + local_size_const.u64[0] = b->shader->info.cs.local_size[0]; + local_size_const.u64[1] = b->shader->info.cs.local_size[1]; + local_size_const.u64[2] = b->shader->info.cs.local_size[2]; + } + local_size = nir_build_imm(b, 3, bit_size, local_size_const); } return local_size; } static nir_ssa_def * -build_local_invocation_id(nir_builder *b) +build_local_invocation_id(nir_builder *b, unsigned bit_size) { if (b->shader->options->lower_cs_local_id_from_index) { /* We lower gl_LocalInvocationID from gl_LocalInvocationIndex based @@ -72,8 +78,8 @@ build_local_invocation_id(nir_builder *b) * accidentally end up with a gl_LocalInvocationIndex that is too * large so it can safely be omitted. */ - nir_ssa_def *local_index = nir_load_local_invocation_index(b); - nir_ssa_def *local_size = build_local_group_size(b); + nir_ssa_def *local_index = nir_load_local_invocation_index(b, bit_size); + nir_ssa_def *local_size = build_local_group_size(b, bit_size); </blockquote><div> </div><div>In my iris clover branch, I instead have a line at that just does "if (bit_size == 64) id = nir_u2u64(b, id)" where "id" is the final computed 3D ID just before returning it. This lets us satisfy the CL requirements while still only using 32-bit system values and doing the calculation in 32 bits. </div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> nir_ssa_def *id_x, *id_y, *id_z; id_x = nir_umod(b, local_index, @@ -86,7 +92,7 @@ build_local_invocation_id(nir_builder *b) nir_channel(b, local_size, 1))); </blockquote><div> </div><div>In other words, righ here we have</div><div> </div><div>nir_ssa_def *id = nir_vec3(b, id_x, id_y, id_z);</div><div> </div><div>if (bit_size == 64)</div><div> id = nir_u2u64(b, id);</div><div> </div><div>return id;</div><div> </div><div>It's kind of nice to avoid unneeded 64-bit math especially when some of that math is division/modulus. :-) </div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> return nir_vec3(b, id_x, id_y, id_z); } else { - return nir_load_local_invocation_id(b); + return nir_load_local_invocation_id(b, bit_size); } } @@ -120,6 +126,7 @@ convert_block(nir_block *block, nir_builder *b) b->cursor = nir_after_instr(&load_deref->instr); + unsigned bit_size = nir_dest_bit_size(load_deref->dest); nir_ssa_def *sysval = NULL; switch (var->data.location) { case SYSTEM_VALUE_GLOBAL_INVOCATION_ID: { @@ -128,9 +135,9 @@ convert_block(nir_block *block, nir_builder *b) * "The value of gl_GlobalInvocationID is equal to * gl_WorkGroupID * gl_WorkGroupSize + gl_LocalInvocationID" */ - nir_ssa_def *group_size = build_local_group_size(b); - nir_ssa_def *group_id = nir_load_work_group_id(b); - nir_ssa_def *local_id = build_local_invocation_id(b); + nir_ssa_def *group_size = build_local_group_size(b, bit_size); + nir_ssa_def *group_id = nir_load_work_group_id(b, bit_size); + nir_ssa_def *local_id = build_local_invocation_id(b, bit_size); sysval = nir_iadd(b, nir_imul(b, group_id, group_size), local_id); break; @@ -150,7 +157,7 @@ convert_block(nir_block *block, nir_builder *b) * gl_WorkGroupSize.y + gl_LocalInvocationID.y * * gl_WorkGroupSize.x + gl_LocalInvocationID.x" */ - nir_ssa_def *local_id = nir_load_local_invocation_id(b); + nir_ssa_def *local_id = nir_load_local_invocation_id(b, bit_size); nir_ssa_def *size_x = nir_imm_int(b, b->shader->info.cs.local_size[0]); @@ -170,11 +177,11 @@ convert_block(nir_block *block, nir_builder *b) * index from the local id. */ if (b->shader->options->lower_cs_local_id_from_index) - sysval = build_local_invocation_id(b); + sysval = build_local_invocation_id(b, bit_size); break; case SYSTEM_VALUE_LOCAL_GROUP_SIZE: { - sysval = build_local_group_size(b); + sysval = build_local_group_size(b, bit_size); break; } @@ -248,8 +255,8 @@ convert_block(nir_block *block, nir_builder *b) break; case SYSTEM_VALUE_GLOBAL_GROUP_SIZE: { - nir_ssa_def *group_size = build_local_group_size(b); - nir_ssa_def *num_work_groups = nir_load_num_work_groups(b); + nir_ssa_def *group_size = build_local_group_size(b, bit_size); + nir_ssa_def *num_work_groups = nir_load_num_work_groups(b, bit_size); sysval = nir_imul(b, group_size, num_work_groups); break; } -- 2.19.2 </blockquote></div></div>