freedreno, tu, ir3: Add last_helper statistic

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24433>

src/freedreno/ir3/ir3.c

@@ -293,6 +293,7 @@ ir3_collect_info(struct ir3_shader_variant *v)
    info->sizedwords = info->size / 4;
 
    bool in_preamble = false;
+   bool has_eq = false;
 
    foreach_block (block, &shader->block_list) {
       int sfu_delay = 0, mem_delay = 0;
@@ -325,6 +326,15 @@ ir3_collect_info(struct ir3_shader_variant *v)
              (instr->dsts[0]->flags & IR3_REG_EI))
             info->last_baryf = info->instrs_count;
 
+         if ((instr->opc == OPC_NOP) && (instr->flags & IR3_INSTR_EQ)) {
+            info->last_helper = info->instrs_count;
+            has_eq = true;
+         }
+
+         if (v->type == MESA_SHADER_FRAGMENT && v->need_pixlod &&
+             instr->opc == OPC_END && !v->prefetch_end_of_quad && !has_eq)
+            info->last_helper = info->instrs_count;
+
          if (instr->opc == OPC_SHPS)
             in_preamble = true;
 

src/freedreno/ir3/ir3.h

@@ -86,6 +86,8 @@ struct ir3_info {
 
    uint16_t last_baryf; /* instruction # of last varying fetch */
 
+   uint16_t last_helper; /* last instruction to use helper invocations */
+
    /* Number of instructions of a given category: */
    uint16_t instrs_per_cat[8];
 };

src/freedreno/ir3/ir3_shader.c

@@ -872,9 +872,10 @@ ir3_shader_disasm(struct ir3_shader_variant *so, uint32_t *bin, FILE *out)
       so->info.cov_count, so->info.sizedwords);
 
    fprintf(out,
-           "; %s prog %d/%d: %u last-baryf, %d half, %d full, %u constlen\n",
+           "; %s prog %d/%d: %u last-baryf, %u last-helper, %d half, %d full, %u constlen\n",
            type, so->shader_id, so->id, so->info.last_baryf,
-           so->info.max_half_reg + 1, so->info.max_reg + 1, so->constlen);
+           so->info.last_helper, so->info.max_half_reg + 1,
+           so->info.max_reg + 1, so->constlen);
 
    fprintf(
       out,

src/freedreno/vulkan/tu_pipeline.cc

@@ -5722,6 +5722,14 @@ tu_GetPipelineExecutableStatisticsKHR(
       stat->value.u64 = exe->stats.last_baryf;
    }
 
+   vk_outarray_append_typed(VkPipelineExecutableStatisticKHR, &out, stat) {
+      WRITE_STR(stat->name, "Last helper instruction");
+      WRITE_STR(stat->description,
+                "The instruction where helper invocations are killed");
+      stat->format = VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_UINT64_KHR;
+      stat->value.u64 = exe->stats.last_helper;
+   }
+
    vk_outarray_append_typed(VkPipelineExecutableStatisticKHR, &out, stat) {
       WRITE_STR(stat->name, "Instructions with SS sync bit");
       WRITE_STR(stat->description,

src/gallium/drivers/freedreno/ir3/ir3_gallium.c

@@ -83,14 +83,15 @@ dump_shader_info(struct ir3_shader_variant *v,
    util_debug_message(
       debug, SHADER_INFO,
       "%s shader: %u inst, %u nops, %u non-nops, %u mov, %u cov, "
-      "%u dwords, %u last-baryf, %u half, %u full, %u constlen, "
+      "%u dwords, %u last-baryf, %u last-helper, %u half, %u full, %u constlen, "
       "%u cat0, %u cat1, %u cat2, %u cat3, %u cat4, %u cat5, %u cat6, %u cat7, "
       "%u stp, %u ldp, %u sstall, %u (ss), %u systall, %u (sy), %d waves, "
       "%d loops\n",
       ir3_shader_stage(v), v->info.instrs_count, v->info.nops_count,
       v->info.instrs_count - v->info.nops_count, v->info.mov_count,
       v->info.cov_count, v->info.sizedwords, v->info.last_baryf,
-      v->info.max_half_reg + 1, v->info.max_reg + 1, v->constlen,
+      v->info.last_helper, v->info.max_half_reg + 1, v->info.max_reg + 1,
+      v->constlen,
       v->info.instrs_per_cat[0], v->info.instrs_per_cat[1],
       v->info.instrs_per_cat[2], v->info.instrs_per_cat[3],
       v->info.instrs_per_cat[4], v->info.instrs_per_cat[5],