tu, ir3: Move multi_pos_output to ir3_shader_variant

This means it has to be rederived rather than passed from the place
where we actually do the optimization, but it eliminates the clutter of
having to pass it around in turnip, which will only get worse with
graphics pipeline libraries.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18554>

src/freedreno/ir3/ir3_compiler_nir.c

@@ -4130,6 +4130,9 @@ setup_output(struct ir3_context *ctx, nir_intrinsic_instr *intr)
     */
    unsigned slot = io.location + (io.per_view ? 0 : offset);
 
+   if (io.per_view && offset > 0)
+      so->multi_pos_output = true;
+
    if (ctx->so->type == MESA_SHADER_FRAGMENT) {
       switch (slot) {
       case FRAG_RESULT_DEPTH:

src/freedreno/ir3/ir3_shader.h

@@ -568,6 +568,9 @@ struct ir3_shader_variant {
    /* Whether we should use the new per-wave layout rather than per-fiber. */
    bool pvtmem_per_wave;
 
+   /* Whether multi-position output is enabled. */
+   bool multi_pos_output;
+
    /* Size in bytes of required shared memory */
    unsigned shared_size;
 

src/freedreno/vulkan/tu_nir_lower_multiview.c

@@ -74,11 +74,8 @@ lower_multiview_mask(nir_shader *nir, uint32_t *mask)
 }
 
 bool
-tu_nir_lower_multiview(nir_shader *nir, uint32_t mask, bool *multi_pos_output,
-                       struct tu_device *dev)
+tu_nir_lower_multiview(nir_shader *nir, uint32_t mask, struct tu_device *dev)
 {
-   *multi_pos_output = false;
-
    bool progress = false;
 
    if (!dev->physical_device->info->a6xx.supports_multiview_mask)
@@ -105,8 +102,6 @@ tu_nir_lower_multiview(nir_shader *nir, uint32_t mask, bool *multi_pos_output,
    if (likely(!(dev->physical_device->instance->debug_flags & TU_DEBUG_NOMULTIPOS)) &&
        num_views <= max_views_for_multipos && num_outputs + (num_views - 1) <= 32 &&
        nir_can_lower_multiview(nir)) {
-      *multi_pos_output = true;
-
       /* It appears that the multiview mask is ignored when multi-position
        * output is enabled, so we have to write 0 to inactive views ourselves.
        */

src/freedreno/vulkan/tu_pipeline.c

@@ -1814,7 +1814,7 @@ tu6_emit_program(struct tu_cs *cs,
    gl_shader_stage stage = MESA_SHADER_VERTEX;
    uint32_t cps_per_patch = builder->create_info->pTessellationState ?
       builder->create_info->pTessellationState->patchControlPoints : 0;
-   bool multi_pos_output = builder->shaders->multi_pos_output;
+   bool multi_pos_output = vs->multi_pos_output;
 
   /* Don't use the binning pass variant when GS is present because we don't
    * support compiling correct binning pass variants with GS.
@@ -2782,7 +2782,6 @@ tu_shaders_serialize(struct vk_pipeline_cache_object *object,
 
    blob_write_bytes(blob, shaders->push_consts, sizeof(shaders->push_consts));
    blob_write_uint8(blob, shaders->active_desc_sets);
-   blob_write_uint8(blob, shaders->multi_pos_output);
 
    for (unsigned i = 0; i < ARRAY_SIZE(shaders->variants); i++) {
       if (shaders->variants[i]) {
@@ -2810,7 +2809,6 @@ tu_shaders_deserialize(struct vk_device *_device,
 
    blob_copy_bytes(blob, shaders->push_consts, sizeof(shaders->push_consts));
    shaders->active_desc_sets = blob_read_uint8(blob);
-   shaders->multi_pos_output = blob_read_uint8(blob);
 
    for (unsigned i = 0; i < ARRAY_SIZE(shaders->variants); i++) {
       bool has_shader = blob_read_uint8(blob);
@@ -3026,8 +3024,6 @@ tu_pipeline_builder_compile_shaders(struct tu_pipeline_builder *builder,
    }
 
    compiled_shaders->active_desc_sets = desc_sets;
-   compiled_shaders->multi_pos_output =
-      shaders[MESA_SHADER_VERTEX]->multi_pos_output;
 
    for (gl_shader_stage stage = MESA_SHADER_VERTEX;
         stage < ARRAY_SIZE(shaders); stage++) {

src/freedreno/vulkan/tu_pipeline.h

@@ -56,7 +56,6 @@ struct tu_compiled_shaders
 
    struct tu_push_constant_range push_consts[MESA_SHADER_STAGES];
    uint8_t active_desc_sets;
-   bool multi_pos_output;
 
    struct ir3_shader_variant *variants[MESA_SHADER_STAGES];
 };

src/freedreno/vulkan/tu_shader.c

@@ -780,8 +780,7 @@ tu_shader_create(struct tu_device *dev,
    ir3_nir_lower_io_to_temporaries(nir);
 
    if (nir->info.stage == MESA_SHADER_VERTEX && key->multiview_mask) {
-      tu_nir_lower_multiview(nir, key->multiview_mask,
-                             &shader->multi_pos_output, dev);
+      tu_nir_lower_multiview(nir, key->multiview_mask, dev);
    }
 
    if (nir->info.stage == MESA_SHADER_FRAGMENT && key->force_sample_interp) {

src/freedreno/vulkan/tu_shader.h

@@ -24,7 +24,6 @@ struct tu_shader
 
    struct tu_push_constant_range push_consts;
    uint8_t active_desc_sets;
-   bool multi_pos_output;
 };
 
 struct tu_shader_key {
@@ -34,8 +33,7 @@ struct tu_shader_key {
 };
 
 bool
-tu_nir_lower_multiview(nir_shader *nir, uint32_t mask, bool *multi_pos_output,
-                       struct tu_device *dev);
+tu_nir_lower_multiview(nir_shader *nir, uint32_t mask, struct tu_device *dev);
 
 nir_shader *
 tu_spirv_to_nir(struct tu_device *dev,