tu: increase maxPushConstantsSize to 256.

Now there are two paths for push constants.

When it's range is under 128b, we can use shared consts.
When it's over 128b, we can instead do loading data through
regular path, which is same as the previous way.

Now we can satisfy emulations like vkd3d that requires 256b for
its root signatures and we think it fairly maps to push constants
rather than inline uniform blocks that requires one indirection.

Signed-off-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15503>

src/freedreno/vulkan/tu_pipeline.c

@@ -1692,7 +1692,8 @@ tu6_emit_program_config(struct tu_cs *cs,
 
    STATIC_ASSERT(MESA_SHADER_VERTEX == 0);
 
-   bool shared_consts_enable = builder->layout->push_constant_size > 0;
+   bool shared_consts_enable = tu6_shared_constants_enable(builder->layout,
+         builder->device->compiler);
    tu6_emit_shared_consts_enable(cs, shared_consts_enable);
 
    tu_cs_emit_regs(cs, A6XX_HLSQ_INVALIDATE_CMD(
@@ -2811,7 +2812,7 @@ tu_pipeline_builder_compile_shaders(struct tu_pipeline_builder *builder,
       stage_infos[stage] = &builder->create_info->pStages[i];
    }
 
-   if (builder->layout->push_constant_size > 0) {
+   if (tu6_shared_constants_enable(builder->layout, builder->device->compiler)) {
       pipeline->shared_consts = (struct tu_push_constant_range) {
          .lo = 0,
          .dwords = builder->layout->push_constant_size / 4,
@@ -4064,7 +4065,7 @@ tu_compute_pipeline_create(VkDevice device,
          VK_PIPELINE_CREATION_FEEDBACK_APPLICATION_PIPELINE_CACHE_HIT_BIT;
    }
 
-   if (layout->push_constant_size > 0) {
+   if (tu6_shared_constants_enable(layout, dev->compiler)) {
       pipeline->shared_consts = (struct tu_push_constant_range) {
          .lo = 0,
          .dwords = layout->push_constant_size / 4,

src/freedreno/vulkan/tu_private.h

@@ -114,7 +114,7 @@ typedef uint32_t xcb_window_t;
 #define MAX_VIEWPORT_SIZE (1 << 14)
 #define MAX_SCISSORS 16
 #define MAX_DISCARD_RECTANGLES 4
-#define MAX_PUSH_CONSTANTS_SIZE 128
+#define MAX_PUSH_CONSTANTS_SIZE 256
 #define MAX_PUSH_DESCRIPTORS 32
 #define MAX_DYNAMIC_UNIFORM_BUFFERS 16
 #define MAX_DYNAMIC_STORAGE_BUFFERS 8
@@ -1430,6 +1430,15 @@ tu_shader_destroy(struct tu_device *dev,
                   struct tu_shader *shader,
                   const VkAllocationCallbacks *alloc);
 
+static bool inline
+tu6_shared_constants_enable(const struct tu_pipeline_layout *layout,
+                            const struct ir3_compiler *compiler)
+{
+   return layout->push_constant_size > 0 &&
+          layout->push_constant_size <= (compiler->shared_consts_size * 16);
+}
+
+
 struct tu_program_descriptor_linkage
 {
    struct ir3_const_state const_state;

src/freedreno/vulkan/tu_shader.c

@@ -141,18 +141,22 @@ static void
 lower_load_push_constant(struct tu_device *dev,
                          nir_builder *b,
                          nir_intrinsic_instr *instr,
-                         struct tu_shader *shader)
+                         struct tu_shader *shader,
+                         const struct tu_pipeline_layout *layout)
 {
    uint32_t base = nir_intrinsic_base(instr);
    assert(base % 4 == 0);
    assert(base >= shader->push_consts.lo * 4);
    base -= shader->push_consts.lo * 4;
 
+   if (tu6_shared_constants_enable(layout, dev->compiler))
+      base += dev->compiler->shared_consts_base_offset * 4;
+
    nir_ssa_def *load =
       nir_load_uniform(b, instr->num_components,
             instr->dest.ssa.bit_size,
             nir_ushr(b, instr->src[0].ssa, nir_imm_int(b, 2)),
-            .base = base + dev->compiler->shared_consts_base_offset * 4);
+            .base = base);
 
    nir_ssa_def_rewrite_uses(&instr->dest.ssa, load);
 
@@ -401,7 +405,7 @@ lower_intrinsic(nir_builder *b, nir_intrinsic_instr *instr,
 {
    switch (instr->intrinsic) {
    case nir_intrinsic_load_push_constant:
-      lower_load_push_constant(dev, b, instr, shader);
+      lower_load_push_constant(dev, b, instr, shader, layout);
       return true;
 
    case nir_intrinsic_load_vulkan_descriptor:
@@ -827,10 +831,13 @@ tu_shader_create(struct tu_device *dev,
 
    ir3_finalize_nir(dev->compiler, nir);
 
+   uint32_t reserved_consts_vec4 = align(shader->push_consts.dwords, 16) / 4;
+   bool shared_consts_enable = tu6_shared_constants_enable(layout, dev->compiler);
+
    shader->ir3_shader =
       ir3_shader_from_nir(dev->compiler, nir, &(struct ir3_shader_options) {
-                           .reserved_user_consts = 0,
-                           .shared_consts_enable = layout->push_constant_size > 0,
+                           .reserved_user_consts = reserved_consts_vec4,
+                           .shared_consts_enable = shared_consts_enable,
                            .api_wavesize = key->api_wavesize,
                            .real_wavesize = key->real_wavesize,
                           }, &so_info);