anv: Enable push constants on gen12-hp

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8342>

src/intel/vulkan/anv_cmd_buffer.c

@@ -1046,6 +1046,7 @@ anv_cmd_buffer_gfx_push_constants(struct anv_cmd_buffer *cmd_buffer)
 struct anv_state
 anv_cmd_buffer_cs_push_constants(struct anv_cmd_buffer *cmd_buffer)
 {
+   const struct gen_device_info *devinfo = &cmd_buffer->device->info;
    struct anv_push_constants *data =
       &cmd_buffer->state.compute.base.push_constants;
    struct anv_compute_pipeline *pipeline = cmd_buffer->state.compute.pipeline;
@@ -1062,10 +1063,16 @@ anv_cmd_buffer_cs_push_constants(struct anv_cmd_buffer *cmd_buffer)
       cmd_buffer->device->info.gen < 8 ? 32 : 64;
    const unsigned aligned_total_push_constants_size =
       ALIGN(total_push_constants_size, push_constant_alignment);
-   struct anv_state state =
+   struct anv_state state;
-      anv_cmd_buffer_alloc_dynamic_state(cmd_buffer,
+   if (devinfo->gen > 12 || gen_device_info_is_12hp(devinfo)) {
+      state = anv_state_stream_alloc(&cmd_buffer->general_state_stream,
                                      aligned_total_push_constants_size,
                                      push_constant_alignment);
+   } else {
+      state = anv_cmd_buffer_alloc_dynamic_state(cmd_buffer,
+                                                 aligned_total_push_constants_size,
+                                                 push_constant_alignment);
+   }
 
    void *dst = state.map;
    const void *src = (char *)data + (range->start * 32);

src/intel/vulkan/anv_nir_compute_push_layout.c

@@ -35,6 +35,7 @@ anv_nir_compute_push_layout(const struct anv_physical_device *pdevice,
                             void *mem_ctx)
 {
    const struct brw_compiler *compiler = pdevice->compiler;
+   const struct gen_device_info *devinfo = compiler->devinfo;
    memset(map->push_ranges, 0, sizeof(map->push_ranges));
 
    bool has_const_ubo = false;
@@ -91,7 +92,8 @@ anv_nir_compute_push_layout(const struct anv_physical_device *pdevice,
       push_end = MAX2(push_end, push_reg_mask_end);
    }
 
-   if (nir->info.stage == MESA_SHADER_COMPUTE) {
+   if (nir->info.stage == MESA_SHADER_COMPUTE &&
+       (devinfo->gen <= 12 && !gen_device_info_is_12hp(devinfo))) {
       /* For compute shaders, we always have to have the subgroup ID.  The
        * back-end compiler will "helpfully" add it for us in the last push
        * constant slot.  Yes, there is an off-by-one error here but that's

src/intel/vulkan/anv_private.h

@@ -2922,6 +2922,8 @@ struct anv_cmd_compute_state {
 
    bool pipeline_dirty;
 
+   struct anv_state push_data;
+
    struct anv_address num_workgroups;
 };
 

src/intel/vulkan/genX_cmd_buffer.c

@@ -4383,7 +4383,8 @@ void genX(CmdEndTransformFeedbackEXT)(
 void
 genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer)
 {
-   struct anv_compute_pipeline *pipeline = cmd_buffer->state.compute.pipeline;
+   struct anv_cmd_compute_state *comp_state = &cmd_buffer->state.compute;
+   struct anv_compute_pipeline *pipeline = comp_state->pipeline;
 
    assert(pipeline->cs);
 
@@ -4449,15 +4450,17 @@ genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer)
    }
 
    if (cmd_buffer->state.push_constants_dirty & VK_SHADER_STAGE_COMPUTE_BIT) {
-      struct anv_state push_state =
+      comp_state->push_data =
          anv_cmd_buffer_cs_push_constants(cmd_buffer);
 
-      if (push_state.alloc_size) {
+#if GEN_GEN <= 12 && !GEN_IS_GEN12HP
+      if (comp_state->push_data.alloc_size) {
          anv_batch_emit(&cmd_buffer->batch, GENX(MEDIA_CURBE_LOAD), curbe) {
-            curbe.CURBETotalDataLength    = push_state.alloc_size;
+            curbe.CURBETotalDataLength    = comp_state->push_data.alloc_size;
-            curbe.CURBEDataStartAddress   = push_state.offset;
+            curbe.CURBEDataStartAddress   = comp_state->push_data.offset;
          }
       }
+#endif
 
       cmd_buffer->state.push_constants_dirty &= ~VK_SHADER_STAGE_COMPUTE_BIT;
    }
@@ -4526,6 +4529,7 @@ emit_compute_walker(struct anv_cmd_buffer *cmd_buffer,
                     uint32_t groupCountX, uint32_t groupCountY,
                     uint32_t groupCountZ)
 {
+   struct anv_cmd_compute_state *comp_state = &cmd_buffer->state.compute;
    const struct anv_shader_bin *cs_bin = pipeline->cs;
    bool predicate = cmd_buffer->state.conditional_render_enabled;
    const struct anv_cs_parameters cs_params = anv_cs_parameters(pipeline);
@@ -4534,6 +4538,8 @@ emit_compute_walker(struct anv_cmd_buffer *cmd_buffer,
       cw.IndirectParameterEnable        = indirect;
       cw.PredicateEnable                = predicate;
       cw.SIMDSize                       = cs_params.simd_size / 16;
+      cw.IndirectDataStartAddress       = comp_state->push_data.offset;
+      cw.IndirectDataLength             = comp_state->push_data.alloc_size;
       cw.LocalXMaximum                  = prog_data->local_size[0] - 1;
       cw.LocalYMaximum                  = prog_data->local_size[1] - 1;
       cw.LocalZMaximum                  = prog_data->local_size[2] - 1;
@@ -4542,7 +4548,6 @@ emit_compute_walker(struct anv_cmd_buffer *cmd_buffer,
       cw.ThreadGroupIDZDimension        = groupCountZ;
       cw.ExecutionMask                  = pipeline->cs_right_mask;
 
-      assert(brw_cs_push_const_total_size(prog_data, cs_params.threads) == 0);
       cw.InterfaceDescriptor = (struct GENX(INTERFACE_DESCRIPTOR_DATA_HP)) {
          .KernelStartPointer = cs_bin->kernel.offset,
          .SamplerStatePointer =