From 16bff57543897f5a208411cc208bfe32cade29a2 Mon Sep 17 00:00:00 2001
From: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Date: Fri, 21 May 2021 18:35:46 +0300
Subject: [PATCH] intel/perf: update Gen9/11 programming for AsyncCompute

Adding a register, similar to what was done for RenderBasic.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10930>
---
 src/intel/perf/oa-cflgt2.xml | 5 +++--
 src/intel/perf/oa-cflgt3.xml | 5 +++--
 src/intel/perf/oa-icl.xml    | 5 +++--
 src/intel/perf/oa-kblgt2.xml | 5 +++--
 src/intel/perf/oa-kblgt3.xml | 5 +++--
 src/intel/perf/oa-sklgt2.xml | 5 +++--
 src/intel/perf/oa-sklgt3.xml | 5 +++--
 src/intel/perf/oa-sklgt4.xml | 5 +++--
 8 files changed, 24 insertions(+), 16 deletions(-)

diff --git a/src/intel/perf/oa-cflgt2.xml b/src/intel/perf/oa-cflgt2.xml
index 020eeb5f34e..c5b9541390d 100644
--- a/src/intel/perf/oa-cflgt2.xml
+++ b/src/intel/perf/oa-cflgt2.xml
@@ -1,5 +1,5 @@
 <?xml version="1.0"?>
-<metrics version="1609841995" merge_md5="">
+<metrics version="1609843830" merge_md5="">
   <set name="Render Metrics Basic Gen9"
        chipset="CFLGT2"
        symbol_name="RenderBasic"
@@ -10871,7 +10871,7 @@
        symbol_name="AsyncCompute"
        underscore_name="async_compute"
        mdapi_supported_apis="VK OGL OGL4 OCL MEDIA IO BB"
-       hw_config_guid="d642768d-65a2-458d-9233-d15ff6f717e9"
+       hw_config_guid="9802484e-5e21-4ad3-8060-12b7f3bb81fe"
        >
     <counter name="GPU Time Elapsed"
              symbol_name="GpuTime"
@@ -11173,6 +11173,7 @@
         <register type="FLEX" address="0x0000E758" value="0x00011010" />
         <register type="FLEX" address="0x0000E45C" value="0x00061060" />
         <register type="FLEX" address="0x0000E55C" value="0x00000008" />
+        <register type="FLEX" address="0x0000E65C" value="0x00222222" />
     </register_config>
   </set>
 
diff --git a/src/intel/perf/oa-cflgt3.xml b/src/intel/perf/oa-cflgt3.xml
index bcf5ad71972..71f00b57b72 100644
--- a/src/intel/perf/oa-cflgt3.xml
+++ b/src/intel/perf/oa-cflgt3.xml
@@ -1,5 +1,5 @@
 <?xml version="1.0"?>
-<metrics version="1609842035" merge_md5="">
+<metrics version="1609843831" merge_md5="">
   <set name="Render Metrics Basic Gen9"
        chipset="CFLGT3"
        symbol_name="RenderBasic"
@@ -10938,7 +10938,7 @@
        symbol_name="AsyncCompute"
        underscore_name="async_compute"
        mdapi_supported_apis="VK OGL OGL4 OCL MEDIA IO BB"
-       hw_config_guid="4a97c6cf-8e84-4854-a031-6c6e7b42e07c"
+       hw_config_guid="0c824ec1-af53-4545-84a3-ede5e3572fc1"
        >
     <counter name="GPU Time Elapsed"
              symbol_name="GpuTime"
@@ -11240,6 +11240,7 @@
         <register type="FLEX" address="0x0000E758" value="0x00011010" />
         <register type="FLEX" address="0x0000E45C" value="0x00061060" />
         <register type="FLEX" address="0x0000E55C" value="0x00000008" />
+        <register type="FLEX" address="0x0000E65C" value="0x00222222" />
     </register_config>
   </set>
 
diff --git a/src/intel/perf/oa-icl.xml b/src/intel/perf/oa-icl.xml
index 7fd0713d0ad..053507cc89e 100644
--- a/src/intel/perf/oa-icl.xml
+++ b/src/intel/perf/oa-icl.xml
@@ -1,5 +1,5 @@
 <?xml version="1.0"?>
-<metrics version="1609843424" merge_md5="">
+<metrics version="1609843831" merge_md5="">
   <set name="Render Metrics Basic Gen11"
        chipset="ICL"
        symbol_name="RenderBasic"
@@ -11874,7 +11874,7 @@
        symbol_name="AsyncCompute"
        underscore_name="async_compute"
        mdapi_supported_apis="VK OGL OGL4 OCL MEDIA IO BB"
-       hw_config_guid="7d863af6-be13-4f00-822d-33f58c8df9e7"
+       hw_config_guid="eca4858e-27a4-4bcf-a57f-2616e2f08950"
        >
     <counter name="GPU Time Elapsed"
              symbol_name="GpuTime"
@@ -12176,6 +12176,7 @@
         <register type="FLEX" address="0x0000E758" value="0x00011010" />
         <register type="FLEX" address="0x0000E45C" value="0x00061060" />
         <register type="FLEX" address="0x0000E55C" value="0x00000008" />
+        <register type="FLEX" address="0x0000E65C" value="0xFFFFFFFF" />
     </register_config>
   </set>
 
diff --git a/src/intel/perf/oa-kblgt2.xml b/src/intel/perf/oa-kblgt2.xml
index 619cda923bd..93f1e2c372c 100644
--- a/src/intel/perf/oa-kblgt2.xml
+++ b/src/intel/perf/oa-kblgt2.xml
@@ -1,5 +1,5 @@
 <?xml version="1.0"?>
-<metrics version="1609842130" merge_md5="">
+<metrics version="1609843830" merge_md5="">
   <set name="Render Metrics Basic Gen9"
        chipset="KBLGT2"
        symbol_name="RenderBasic"
@@ -10871,7 +10871,7 @@
        symbol_name="AsyncCompute"
        underscore_name="async_compute"
        mdapi_supported_apis="VK OGL OGL4 OCL MEDIA IO BB"
-       hw_config_guid="2f0740b8-da7c-47ff-b7ec-cc7bbf2057a0"
+       hw_config_guid="4032137b-8dcc-4f3f-912d-c0bb14a123bc"
        >
     <counter name="GPU Time Elapsed"
              symbol_name="GpuTime"
@@ -11173,6 +11173,7 @@
         <register type="FLEX" address="0x0000E758" value="0x00011010" />
         <register type="FLEX" address="0x0000E45C" value="0x00061060" />
         <register type="FLEX" address="0x0000E55C" value="0x00000008" />
+        <register type="FLEX" address="0x0000E65C" value="0x00222222" />
     </register_config>
   </set>
 
diff --git a/src/intel/perf/oa-kblgt3.xml b/src/intel/perf/oa-kblgt3.xml
index 64a54e96872..c1017a4701d 100644
--- a/src/intel/perf/oa-kblgt3.xml
+++ b/src/intel/perf/oa-kblgt3.xml
@@ -1,5 +1,5 @@
 <?xml version="1.0"?>
-<metrics version="1609842140" merge_md5="">
+<metrics version="1609843830" merge_md5="">
   <set name="Render Metrics Basic Gen9"
        chipset="KBLGT3"
        symbol_name="RenderBasic"
@@ -10938,7 +10938,7 @@
        symbol_name="AsyncCompute"
        underscore_name="async_compute"
        mdapi_supported_apis="VK OGL OGL4 OCL MEDIA IO BB"
-       hw_config_guid="cad0e478-5fec-44cc-ad01-76d4c61f52ad"
+       hw_config_guid="8d6d28f5-f699-4eec-89f7-216956bf2ed5"
        >
     <counter name="GPU Time Elapsed"
              symbol_name="GpuTime"
@@ -11240,6 +11240,7 @@
         <register type="FLEX" address="0x0000E758" value="0x00011010" />
         <register type="FLEX" address="0x0000E45C" value="0x00061060" />
         <register type="FLEX" address="0x0000E55C" value="0x00000008" />
+        <register type="FLEX" address="0x0000E65C" value="0x00222222" />
     </register_config>
   </set>
 
diff --git a/src/intel/perf/oa-sklgt2.xml b/src/intel/perf/oa-sklgt2.xml
index 12e01180d59..fbd49ba6554 100644
--- a/src/intel/perf/oa-sklgt2.xml
+++ b/src/intel/perf/oa-sklgt2.xml
@@ -1,5 +1,5 @@
 <?xml version="1.0"?>
-<metrics version="1609842167" merge_md5="">
+<metrics version="1609843830" merge_md5="">
   <set name="Render Metrics Basic Gen9"
        chipset="SKLGT2"
        symbol_name="RenderBasic"
@@ -11900,7 +11900,7 @@
        symbol_name="AsyncCompute"
        underscore_name="async_compute"
        mdapi_supported_apis="VK OGL OGL4 OCL MEDIA IO BB"
-       hw_config_guid="ea700484-6e4d-4765-94a3-7867fba6805a"
+       hw_config_guid="04769cac-6809-4f2a-af5d-40022435a8b2"
        >
     <counter name="GPU Time Elapsed"
              symbol_name="GpuTime"
@@ -12202,6 +12202,7 @@
         <register type="FLEX" address="0x0000E758" value="0x00011010" />
         <register type="FLEX" address="0x0000E45C" value="0x00061060" />
         <register type="FLEX" address="0x0000E55C" value="0x00000008" />
+        <register type="FLEX" address="0x0000E65C" value="0x00222222" />
     </register_config>
   </set>
 
diff --git a/src/intel/perf/oa-sklgt3.xml b/src/intel/perf/oa-sklgt3.xml
index 2e32bc7d828..98c3374bd41 100644
--- a/src/intel/perf/oa-sklgt3.xml
+++ b/src/intel/perf/oa-sklgt3.xml
@@ -1,5 +1,5 @@
 <?xml version="1.0"?>
-<metrics version="1609842186" merge_md5="">
+<metrics version="1609843830" merge_md5="">
   <set name="Render Metrics Basic Gen9"
        chipset="SKLGT3"
        symbol_name="RenderBasic"
@@ -10938,7 +10938,7 @@
        symbol_name="AsyncCompute"
        underscore_name="async_compute"
        mdapi_supported_apis="VK OGL OGL4 OCL MEDIA IO BB"
-       hw_config_guid="2acbf308-143d-4846-a976-6ad7cc83148a"
+       hw_config_guid="9d8e6b6c-c6ec-4591-99b5-7d9983e10d64"
        >
     <counter name="GPU Time Elapsed"
              symbol_name="GpuTime"
@@ -11240,6 +11240,7 @@
         <register type="FLEX" address="0x0000E758" value="0x00011010" />
         <register type="FLEX" address="0x0000E45C" value="0x00061060" />
         <register type="FLEX" address="0x0000E55C" value="0x00000008" />
+        <register type="FLEX" address="0x0000E65C" value="0x00222222" />
     </register_config>
   </set>
 
diff --git a/src/intel/perf/oa-sklgt4.xml b/src/intel/perf/oa-sklgt4.xml
index 570919df72c..77415661e65 100644
--- a/src/intel/perf/oa-sklgt4.xml
+++ b/src/intel/perf/oa-sklgt4.xml
@@ -1,5 +1,5 @@
 <?xml version="1.0"?>
-<metrics version="1609842208" merge_md5="">
+<metrics version="1609843830" merge_md5="">
   <set name="Render Metrics Basic Gen9"
        chipset="SKLGT4"
        symbol_name="RenderBasic"
@@ -10961,7 +10961,7 @@
        symbol_name="AsyncCompute"
        underscore_name="async_compute"
        mdapi_supported_apis="VK OGL OGL4 OCL MEDIA IO BB"
-       hw_config_guid="93685e9d-f4ba-4f30-a087-a4f56246fc70"
+       hw_config_guid="796044f3-8f9d-4439-ab2c-aca5b046a1ef"
        >
     <counter name="GPU Time Elapsed"
              symbol_name="GpuTime"
@@ -11263,6 +11263,7 @@
         <register type="FLEX" address="0x0000E758" value="0x00011010" />
         <register type="FLEX" address="0x0000E45C" value="0x00061060" />
         <register type="FLEX" address="0x0000E55C" value="0x00000008" />
+        <register type="FLEX" address="0x0000E65C" value="0x00222222" />
     </register_config>
   </set>