dxil: Don't generate load_ubo_dxil directly

Just use load_ubo and let it get lowered appropriately later on.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23173>

diff --git a/src/gallium/drivers/d3d12/d3d12_compiler.cpp b/src/gallium/drivers/d3d12/d3d12_compiler.cpp
index 34d650e..73b9a79 100644 (file)
--- a/src/gallium/drivers/d3d12/d3d12_compiler.cpp
+++ b/src/gallium/drivers/d3d12/d3d12_compiler.cpp
@@ -138,7 +138,6 @@ compile_nir(struct d3d12_context *ctx, struct d3d12_shader_selector *sel,
          NIR_PASS_V(nir, nir_lower_clip_halfz);
       NIR_PASS_V(nir, d3d12_lower_yflip);
    }
-   NIR_PASS_V(nir, nir_lower_packed_ubo_loads);
    NIR_PASS_V(nir, d3d12_lower_load_draw_params);
    NIR_PASS_V(nir, d3d12_lower_load_patch_vertices_in);
    NIR_PASS_V(nir, d3d12_lower_state_vars, shader);

diff --git a/src/gallium/drivers/d3d12/d3d12_nir_passes.c b/src/gallium/drivers/d3d12/d3d12_nir_passes.c
index 2f84440..f04c5ce 100644 (file)
--- a/src/gallium/drivers/d3d12/d3d12_nir_passes.c
+++ b/src/gallium/drivers/d3d12/d3d12_nir_passes.c
@@ -628,42 +628,6 @@ d3d12_add_missing_dual_src_target(struct nir_shader *s,
                                nir_metadata_dominance);
 }
 
-static bool
-lower_load_ubo_packed_filter(const nir_instr *instr,
-                             UNUSED const void *_options) {
-   if (instr->type != nir_instr_type_intrinsic)
-      return false;
-
-   nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
-
-   return intr->intrinsic == nir_intrinsic_load_ubo;
-}
-
-static nir_ssa_def *
-lower_load_ubo_packed_impl(nir_builder *b, nir_instr *instr,
-                              UNUSED void *_options) {
-   nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
-
-   nir_ssa_def *buffer = intr->src[0].ssa;
-   nir_ssa_def *offset = intr->src[1].ssa;
-
-   nir_ssa_def *result =
-      build_load_ubo_dxil(b, buffer,
-                          offset,
-                          nir_dest_num_components(intr->dest),
-                          nir_dest_bit_size(intr->dest),
-                          nir_intrinsic_align(intr));
-   return result;
-}
-
-bool
-nir_lower_packed_ubo_loads(nir_shader *nir) {
-   return nir_shader_lower_instructions(nir,
-                                        lower_load_ubo_packed_filter,
-                                        lower_load_ubo_packed_impl,
-                                        NULL);
-}
-
 void
 d3d12_lower_primitive_id(nir_shader *shader)
 {

diff --git a/src/gallium/drivers/d3d12/d3d12_nir_passes.h b/src/gallium/drivers/d3d12/d3d12_nir_passes.h
index 8a39717..f80d2e6 100644 (file)
--- a/src/gallium/drivers/d3d12/d3d12_nir_passes.h
+++ b/src/gallium/drivers/d3d12/d3d12_nir_passes.h
@@ -88,9 +88,6 @@ d3d12_fix_io_uint_type(struct nir_shader *s, uint64_t in_mask, uint64_t out_mask
 void
 d3d12_nir_invert_depth(nir_shader *s, unsigned viewport_mask, bool clip_halfz);
 
-bool
-nir_lower_packed_ubo_loads(struct nir_shader *nir);
-
 void
 d3d12_lower_primitive_id(nir_shader *shader);
 

diff --git a/src/microsoft/clc/clc_nir.c b/src/microsoft/clc/clc_nir.c
index e5cbbd6..a5beedf 100644 (file)
--- a/src/microsoft/clc/clc_nir.c
+++ b/src/microsoft/clc/clc_nir.c
@@ -31,20 +31,28 @@
 #include "clc_compiler.h"
 #include "../compiler/dxil_nir.h"
 
+static nir_ssa_def *
+load_ubo(nir_builder *b, nir_intrinsic_instr *intr, nir_variable *var, unsigned offset)
+{
+   return nir_build_load_ubo(b,
+                             nir_dest_num_components(intr->dest),
+                             nir_dest_bit_size(intr->dest),
+                             nir_imm_int(b, var->data.binding),
+                             nir_imm_int(b, offset),
+                             .align_mul = 256,
+                             .align_offset = offset,
+                             .range_base = offset,
+                             .range = nir_dest_bit_size(intr->dest) * nir_dest_num_components(intr->dest) / 8);
+}
+
 static bool
 lower_load_base_global_invocation_id(nir_builder *b, nir_intrinsic_instr *intr,
                                     nir_variable *var)
 {
    b->cursor = nir_after_instr(&intr->instr);
 
-   nir_ssa_def *offset =
-      build_load_ubo_dxil(b, nir_imm_int(b, var->data.binding),
-                          nir_imm_int(b,
-                                      offsetof(struct clc_work_properties_data,
-                                               global_offset_x)),
-                          nir_dest_num_components(intr->dest),
-                          nir_dest_bit_size(intr->dest),
-                          sizeof(uint32_t) * 4);
+   nir_ssa_def *offset = load_ubo(b, intr, var, offsetof(struct clc_work_properties_data,
+                                                         global_offset_x));
    nir_ssa_def_rewrite_uses(&intr->dest.ssa, offset);
    nir_instr_remove(&intr->instr);
    return true;
@@ -56,14 +64,8 @@ lower_load_work_dim(nir_builder *b, nir_intrinsic_instr *intr,
 {
    b->cursor = nir_after_instr(&intr->instr);
 
-   nir_ssa_def *dim =
-      build_load_ubo_dxil(b, nir_imm_int(b, var->data.binding),
-                          nir_imm_int(b,
-                                      offsetof(struct clc_work_properties_data,
-                                               work_dim)),
-                          nir_dest_num_components(intr->dest),
-                          nir_dest_bit_size(intr->dest),
-                          sizeof(uint32_t));
+   nir_ssa_def *dim = load_ubo(b, intr, var, offsetof(struct clc_work_properties_data,
+                                                      work_dim));
    nir_ssa_def_rewrite_uses(&intr->dest.ssa, dim);
    nir_instr_remove(&intr->instr);
    return true;
@@ -76,13 +78,8 @@ lower_load_num_workgroups(nir_builder *b, nir_intrinsic_instr *intr,
    b->cursor = nir_after_instr(&intr->instr);
 
    nir_ssa_def *count =
-      build_load_ubo_dxil(b, nir_imm_int(b, var->data.binding),
-                         nir_imm_int(b,
-                                     offsetof(struct clc_work_properties_data,
-                                              group_count_total_x)),
-                         nir_dest_num_components(intr->dest),
-                         nir_dest_bit_size(intr->dest),
-                         sizeof(uint32_t) * 4);
+      load_ubo(b, intr, var, offsetof(struct clc_work_properties_data,
+                                      group_count_total_x));
    nir_ssa_def_rewrite_uses(&intr->dest.ssa, count);
    nir_instr_remove(&intr->instr);
    return true;
@@ -95,13 +92,8 @@ lower_load_base_workgroup_id(nir_builder *b, nir_intrinsic_instr *intr,
    b->cursor = nir_after_instr(&intr->instr);
 
    nir_ssa_def *offset =
-      build_load_ubo_dxil(b, nir_imm_int(b, var->data.binding),
-                         nir_imm_int(b,
-                                     offsetof(struct clc_work_properties_data,
-                                              group_id_offset_x)),
-                         nir_dest_num_components(intr->dest),
-                         nir_dest_bit_size(intr->dest),
-                         sizeof(uint32_t) * 4);
+      load_ubo(b, intr, var, offsetof(struct clc_work_properties_data,
+                                      group_id_offset_x));
    nir_ssa_def_rewrite_uses(&intr->dest.ssa, offset);
    nir_instr_remove(&intr->instr);
    return true;

diff --git a/src/microsoft/spirv_to_dxil/dxil_spirv_nir.c b/src/microsoft/spirv_to_dxil/dxil_spirv_nir.c
index e591bec..469ed61 100644 (file)
--- a/src/microsoft/spirv_to_dxil/dxil_spirv_nir.c
+++ b/src/microsoft/spirv_to_dxil/dxil_spirv_nir.c
@@ -252,15 +252,16 @@ lower_shader_system_values(struct nir_builder *builder, nir_instr *instr,
       nir_address_format_bit_size(ubo_format),
       index, .desc_type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER);
 
-   unsigned num_components = nir_dest_num_components(intrin->dest);
-   unsigned alignment = (num_components == 3 ? 4 : num_components) *
-      nir_dest_bit_size(intrin->dest) / 8;
-   assert(offset % alignment == 0);
-   nir_ssa_def *load_data = build_load_ubo_dxil(
-      builder, nir_channel(builder, load_desc, 0),
+   nir_ssa_def *load_data = nir_build_load_ubo(
+      builder, 
+      nir_dest_num_components(intrin->dest),
+      nir_dest_bit_size(intrin->dest),
+      nir_channel(builder, load_desc, 0),
       nir_imm_int(builder, offset),
-      num_components, nir_dest_bit_size(intrin->dest),
-      alignment);
+      .align_mul = 256,
+      .align_offset = offset,
+      .range_base = offset,
+      .range = nir_dest_bit_size(intrin->dest) * nir_dest_num_components(intrin->dest) / 8);
 
    nir_ssa_def_rewrite_uses(&intrin->dest.ssa, load_data);
    nir_instr_remove(instr);
@@ -340,11 +341,16 @@ lower_load_push_constant(struct nir_builder *builder, nir_instr *instr,
       index, .desc_type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER);
 
    nir_ssa_def *offset = nir_ssa_for_src(builder, intrin->src[0], 1);
-   nir_ssa_def *load_data = build_load_ubo_dxil(
-      builder, nir_channel(builder, load_desc, 0),
+   nir_ssa_def *load_data = nir_build_load_ubo(
+      builder, 
+      nir_dest_num_components(intrin->dest),
+      nir_dest_bit_size(intrin->dest), 
+      nir_channel(builder, load_desc, 0),
       nir_iadd_imm(builder, offset, base),
-      nir_dest_num_components(intrin->dest), nir_dest_bit_size(intrin->dest),
-      nir_intrinsic_align(intrin));
+      .align_mul = nir_intrinsic_align_mul(intrin),
+      .align_offset = nir_intrinsic_align_offset(intrin),
+      .range_base = base,
+      .range = range);
 
    nir_ssa_def_rewrite_uses(&intrin->dest.ssa, load_data);
    nir_instr_remove(instr);
@@ -431,9 +437,13 @@ lower_yz_flip(struct nir_builder *builder, nir_instr *instr,
          index, .desc_type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER);
 
       dyn_yz_flip_mask =
-         build_load_ubo_dxil(builder,
-                             nir_channel(builder, load_desc, 0),
-                             nir_imm_int(builder, offset), 1, 32, 4);
+         nir_build_load_ubo(builder, 1, 32,
+                            nir_channel(builder, load_desc, 0),
+                            nir_imm_int(builder, offset),
+                            .align_mul = 256,
+                            .align_offset = offset,
+                            .range_base = offset,
+                            .range = 4);
       *data->reads_sysval_ubo = true;
    }
 
@@ -728,9 +738,12 @@ write_pntc_with_pos(nir_builder *b, nir_instr *instr, void *_data)
       index, .desc_type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER);
 
    nir_ssa_def *transform = nir_channels(b,
-                                         build_load_ubo_dxil(b,
-                                                             nir_channel(b, load_desc, 0),
-                                                             nir_imm_int(b, offset), 4, 32, 16),
+                                         nir_build_load_ubo(b, 4, 32,
+                                                            nir_channel(b, load_desc, 0),
+                                                            nir_imm_int(b, offset),
+                                                            .align_mul = 16,
+                                                            .range_base = offset,
+                                                            .range = 16),
                                          0x6);
    nir_ssa_def *point_center_in_clip = nir_fmul(b, nir_trim_vector(b, pos, 2),
                                                 nir_frcp(b, nir_channel(b, pos, 3)));

diff --git a/src/microsoft/vulkan/dzn_nir.c b/src/microsoft/vulkan/dzn_nir.c
index 20f50bd..9fef765 100644 (file)
--- a/src/microsoft/vulkan/dzn_nir.c
+++ b/src/microsoft/vulkan/dzn_nir.c
@@ -829,10 +829,12 @@ load_dynamic_depth_bias(nir_builder *b, struct dzn_nir_point_gs_info *info)
       nir_address_format_bit_size(ubo_format),
       index, .desc_type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER);
 
-   return build_load_ubo_dxil(
-      b, nir_channel(b, load_desc, 0),
+   return nir_build_load_ubo(
+      b, 1, 32,
+      nir_channel(b, load_desc, 0),
       nir_imm_int(b, offset),
-      1, 32, 4);
+      .align_mul = 256,
+      .align_offset = offset);
 }
 
 nir_shader *