intel/perf: Add and use a function to return platform OA format

The platform version check to return the OA format was duplicated
in a few places, so adding a function and dropping this duplication.

While at it, already making it future proof for Xe KMD support and
split i915 specific code to its own file.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29077>

src/intel/perf/i915/intel_perf.c

@@ -0,0 +1,20 @@
+/*
+ * Copyright 2024 Intel Corporation
+ * SPDX-License-Identifier: MIT
+ */
+
+#include "perf/i915/intel_perf.h"
+
+#include "perf/intel_perf.h"
+
+#include "drm-uapi/i915_drm.h"
+
+uint64_t i915_perf_get_oa_format(struct intel_perf_config *perf)
+{
+   if (perf->devinfo->verx10 <= 75)
+      return I915_OA_FORMAT_A45_B8_C8;
+   else if (perf->devinfo->verx10 <= 120)
+      return I915_OA_FORMAT_A32u40_A4u32_B8_C8;
+   else
+      return I915_OA_FORMAT_A24u40_A14u32_B8_C8;
+}

src/intel/perf/i915/intel_perf.h

@@ -0,0 +1,12 @@
+/*
+ * Copyright 2024 Intel Corporation
+ * SPDX-License-Identifier: MIT
+ */
+
+#pragma once
+
+#include <stdint.h>
+
+struct intel_perf_config;
+
+uint64_t i915_perf_get_oa_format(struct intel_perf_config *perf);

src/intel/perf/intel_perf.c

@@ -41,6 +41,7 @@
 #include "dev/intel_debug.h"
 #include "dev/intel_device_info.h"
 
+#include "perf/i915/intel_perf.h"
 #include "perf/intel_perf.h"
 #include "perf/intel_perf_regs.h"
 #include "perf/intel_perf_mdapi.h"
@@ -1567,3 +1568,15 @@ intel_perf_free(struct intel_perf_config *perf_cfg)
 {
    ralloc_free(perf_cfg);
 }
+
+uint64_t
+intel_perf_get_oa_format(struct intel_perf_config *perf_cfg)
+{
+   switch (perf_cfg->devinfo->kmd_type) {
+   case INTEL_KMD_TYPE_I915:
+      return i915_perf_get_oa_format(perf_cfg);
+   default:
+      unreachable("missing");
+      return 0;
+   }
+}

src/intel/perf/intel_perf.h

@@ -534,6 +534,8 @@ intel_perf_new(void *ctx)
 
 void intel_perf_free(struct intel_perf_config *perf_cfg);
 
+uint64_t intel_perf_get_oa_format(struct intel_perf_config *perf_cfg);
+
 /** Whether we have the ability to hold off preemption on a batch so we don't
  * have to look at the OA buffer to subtract unrelated workloads off the
  * values captured through MI_* commands.

src/intel/perf/intel_perf_mdapi.c

@@ -243,7 +243,6 @@ intel_perf_register_mdapi_oa_query(struct intel_perf_config *perf,
    switch (devinfo->ver) {
    case 7: {
       query = intel_perf_append_query_info(perf, 1 + 45 + 16 + 7);
-      query->oa_format = I915_OA_FORMAT_A45_B8_C8;
 
       struct gfx7_mdapi_metrics metric_data;
       query->data_size = sizeof(metric_data);
@@ -268,7 +267,6 @@ intel_perf_register_mdapi_oa_query(struct intel_perf_config *perf,
    }
    case 8: {
       query = intel_perf_append_query_info(perf, 2 + 36 + 16 + 16);
-      query->oa_format = I915_OA_FORMAT_A32u40_A4u32_B8_C8;
 
       struct gfx8_mdapi_metrics metric_data;
       query->data_size = sizeof(metric_data);
@@ -305,7 +303,6 @@ intel_perf_register_mdapi_oa_query(struct intel_perf_config *perf,
    case 11:
    case 12: {
       query = intel_perf_append_query_info(perf, 2 + 36 + 16 + 16 + 16 + 2);
-      query->oa_format = I915_OA_FORMAT_A32u40_A4u32_B8_C8;
 
       struct gfx9_mdapi_metrics metric_data;
       query->data_size = sizeof(metric_data);
@@ -349,6 +346,7 @@ intel_perf_register_mdapi_oa_query(struct intel_perf_config *perf,
       break;
    }
 
+   query->oa_format = intel_perf_get_oa_format(perf);
    query->kind = INTEL_PERF_QUERY_TYPE_RAW;
    query->name = "Intel_Raw_Hardware_Counters_Set_0_Query";
    query->guid = INTEL_PERF_QUERY_GUID_MDAPI;

src/intel/perf/intel_perf_setup.h

@@ -38,10 +38,10 @@ intel_query_alloc(struct intel_perf_config *perf, int ncounters)
    query->n_counters = 0;
    query->oa_metrics_set_id = 0; /* determined at runtime, via sysfs */
    query->counters = rzalloc_array(query, struct intel_perf_query_counter, ncounters);
+   query->oa_format = intel_perf_get_oa_format(perf);
 
    /* Accumulation buffer offsets... */
    if (perf->devinfo->verx10 <= 75) {
-      query->oa_format = I915_OA_FORMAT_A45_B8_C8;
       query->gpu_time_offset = 0;
       query->a_offset = query->gpu_time_offset + 1;
       query->b_offset = query->a_offset + 45;
@@ -49,7 +49,6 @@ intel_query_alloc(struct intel_perf_config *perf, int ncounters)
       query->perfcnt_offset = query->c_offset + 8;
       query->rpstat_offset = query->perfcnt_offset + 2;
    } else if (perf->devinfo->verx10 <= 120) {
-      query->oa_format = I915_OA_FORMAT_A32u40_A4u32_B8_C8;
       query->gpu_time_offset = 0;
       query->gpu_clock_offset = query->gpu_time_offset + 1;
       query->a_offset = query->gpu_clock_offset + 1;
@@ -58,7 +57,6 @@ intel_query_alloc(struct intel_perf_config *perf, int ncounters)
       query->perfcnt_offset = query->c_offset + 8;
       query->rpstat_offset = query->perfcnt_offset + 2;
    } else {
-      query->oa_format = I915_OA_FORMAT_A24u40_A14u32_B8_C8;
       query->gpu_time_offset = 0;
       query->gpu_clock_offset = query->gpu_time_offset + 1;
       query->a_offset = query->gpu_clock_offset + 1;

src/intel/perf/meson.build

@@ -17,6 +17,8 @@ foreach hw : intel_hw_metrics
 endforeach
 
 intel_perf_sources = [
+  'i915/intel_perf.c',
+  'i915/intel_perf.h',
   'intel_perf.c',
   'intel_perf_query.c',
   'intel_perf_mdapi.c',

src/intel/vulkan/anv_perf.c

@@ -107,10 +107,7 @@ anv_device_perf_open(struct anv_device *device, uint64_t metric_id)
    properties[p++] = metric_id;
 
    properties[p++] = DRM_I915_PERF_PROP_OA_FORMAT;
-   properties[p++] =
+   properties[p++] = intel_perf_get_oa_format(device->physical->perf);
-      device->info->verx10 >= 125 ?
-      I915_OA_FORMAT_A24u40_A14u32_B8_C8 :
-      I915_OA_FORMAT_A32u40_A4u32_B8_C8;
 
    properties[p++] = DRM_I915_PERF_PROP_OA_EXPONENT;
    properties[p++] = 31; /* slowest sampling period */

src/intel/vulkan_hasvk/anv_perf.c

@@ -115,9 +115,7 @@ anv_device_perf_open(struct anv_device *device, uint64_t metric_id)
    properties[p++] = metric_id;
 
    properties[p++] = DRM_I915_PERF_PROP_OA_FORMAT;
-   properties[p++] = device->info->ver >= 8 ?
+   properties[p++] = intel_perf_get_oa_format(device->physical->perf);
-      I915_OA_FORMAT_A32u40_A4u32_B8_C8 :
-      I915_OA_FORMAT_A45_B8_C8;
 
    properties[p++] = DRM_I915_PERF_PROP_OA_EXPONENT;
    properties[p++] = 31; /* slowest sampling period */