[Media Common] [VP] Enable pat index setting when creating bo

Enable pat index setting for perf close #1680
intel · Jun 28, 2023 · cf94234 · cf94234
1 parent 3b5369d
commit cf94234
Show file tree

Hide file tree

Showing 14 changed files with 393 additions and 117 deletions.
diff --git a/media_driver/linux/ult/libdrm_mock/mos_bufmgr_api_mock.c b/media_driver/linux/ult/libdrm_mock/mos_bufmgr_api_mock.c
@@ -48,16 +48,16 @@
 
 struct mos_linux_bo *
 mos_bo_alloc(struct mos_bufmgr *bufmgr, const char *name,
-           unsigned long size, unsigned int alignment, int mem_type)
+           unsigned long size, unsigned int alignment, int mem_type, unsigned int pat_index, bool cpu_cacheable)
 {
-    return bufmgr->bo_alloc(bufmgr, name, size, alignment, mem_type);
+    return bufmgr->bo_alloc(bufmgr, name, size, alignment, mem_type, pat_index, cpu_cacheable);
 }
 
 struct mos_linux_bo *
 mos_bo_alloc_for_render(struct mos_bufmgr *bufmgr, const char *name,
-                  unsigned long size, unsigned int alignment, int mem_type)
+                  unsigned long size, unsigned int alignment, int mem_type, unsigned int pat_index, bool cpu_cacheable)
 {
-    return bufmgr->bo_alloc_for_render(bufmgr, name, size, alignment, mem_type);
+    return bufmgr->bo_alloc_for_render(bufmgr, name, size, alignment, mem_type, pat_index, cpu_cacheable);
 }
 
 struct mos_linux_bo *
@@ -77,10 +77,10 @@ mos_bo_alloc_userptr(struct mos_bufmgr *bufmgr,
 struct mos_linux_bo *
 mos_bo_alloc_tiled(struct mos_bufmgr *bufmgr, const char *name,
                         int x, int y, int cpp, uint32_t *tiling_mode,
-                        unsigned long *pitch, unsigned long flags, int mem_type)
+                        unsigned long *pitch, unsigned long flags, int mem_type, unsigned int pat_index, bool cpu_cacheable)
 {
     return bufmgr->bo_alloc_tiled(bufmgr, name, x, y, cpp,
-                      tiling_mode, pitch, flags, mem_type);
+                      tiling_mode, pitch, flags, mem_type, pat_index, cpu_cacheable);
 }
 
 void

diff --git a/media_driver/linux/ult/libdrm_mock/mos_bufmgr_mock.c b/media_driver/linux/ult/libdrm_mock/mos_bufmgr_mock.c
@@ -879,8 +879,12 @@ mos_gem_bo_alloc_internal(struct mos_bufmgr *bufmgr,
                 uint32_t tiling_mode,
                 unsigned long stride,
                 unsigned int alignment,
-                int mem_type)
+                int mem_type,
+                unsigned int pat_index,
+                bool cpu_cacheable)
 {
+    MOS_UNUSED(pat_index);
+    MOS_UNUSED(cpu_cacheable);
     struct mos_bufmgr_gem *bufmgr_gem = (struct mos_bufmgr_gem *) bufmgr;
     struct mos_bo_gem *bo_gem;
     unsigned int page_size = getpagesize();
@@ -1049,31 +1053,37 @@ mos_gem_bo_alloc_for_render(struct mos_bufmgr *bufmgr,
                   const char *name,
                   unsigned long size,
                   unsigned int alignment,
-                  int mem_type)
+                  int mem_type,
+                  unsigned int pat_index,
+                  bool cpu_cacheable)
 {
     return mos_gem_bo_alloc_internal(bufmgr, name, size,
                            I915_TILING_NONE, 0,
                            BO_ALLOC_FOR_RENDER,
                            alignment,
-                           mem_type);
+                           mem_type,
+                           pat_index,
+                           cpu_cacheable);
 }
 
 static struct mos_linux_bo *
 mos_gem_bo_alloc(struct mos_bufmgr *bufmgr,
                const char *name,
                unsigned long size,
                unsigned int alignment,
-               int mem_type)
+               int mem_type,
+               unsigned int pat_index,
+               bool cpu_cacheable)
 {
     return mos_gem_bo_alloc_internal(bufmgr, name, size, 0,
-                           I915_TILING_NONE, 0, 0, mem_type);
+                           I915_TILING_NONE, 0, 0, mem_type, pat_index, cpu_cacheable);
 }
 
 static struct mos_linux_bo *
 mos_gem_bo_alloc_tiled(struct mos_bufmgr *bufmgr, const char *name,
                  int x, int y, int cpp, uint32_t *tiling_mode,
                  unsigned long *pitch, unsigned long flags,
-                 int mem_type)
+                 int mem_type, unsigned int pat_index, bool cpu_cacheable)
 {
     struct mos_bufmgr_gem *bufmgr_gem = (struct mos_bufmgr_gem *)bufmgr;
     unsigned long size, stride;
@@ -1116,7 +1126,7 @@ mos_gem_bo_alloc_tiled(struct mos_bufmgr *bufmgr, const char *name,
     if (tiling == I915_TILING_NONE)
         stride = 0;
     return mos_gem_bo_alloc_internal(bufmgr, name, size, flags,
-                                               tiling, stride, 0, mem_type);
+                                               tiling, stride, 0, mem_type, pat_index, cpu_cacheable);
 }
 
 static struct mos_linux_bo *

diff --git a/media_softlet/agnostic/common/os/mos_interface.h b/media_softlet/agnostic/common/os/mos_interface.h
@@ -308,6 +308,21 @@ class MosInterface
     static GMM_CLIENT_CONTEXT *GetGmmClientContext(
         MOS_STREAM_HANDLE streamState);
 
+    //!
+    //! \brief  Get PAT index from gmm
+    //!
+    //! \param  [in] gmmClient
+    //!         GMM client context
+    //! \param  [in] gmmResourceInfo
+    //!         gmm resource info
+    //!
+    //! \return unsigned int
+    //!         Pat index
+    //!
+    static unsigned int GetPATIndexFromGmm(
+        GMM_CLIENT_CONTEXT *gmmClient,
+        GMM_RESOURCE_INFO *gmmResourceInfo);
+
     //!
     //! \brief    Get current Gpu context priority
     //! \details  Get current Gpu context priority

diff --git a/media_softlet/linux/common/ddi/media_libva_util_next.cpp b/media_softlet/linux/common/ddi/media_libva_util_next.cpp
@@ -734,16 +734,19 @@ VAStatus MediaLibvaUtilNext::CreateInternalSurface(
 
     params.memType = MemoryPolicyManager::UpdateMemoryPolicy(&memPolicyPar);
 
+    unsigned int patIndex = MosInterface::GetPATIndexFromGmm(mediaDrvCtx->pGmmClientContext, gmmResourceInfo);
+    bool isCpuCacheable   = gmmResourceInfo->GetResFlags().Info.Cacheable;
+
     if ( params.tileFormat == I915_TILING_NONE )
     {
-        bo = mos_bo_alloc(mediaDrvCtx->pDrmBufMgr, "MEDIA", gmmSize, 4096, params.memType);
+        bo = mos_bo_alloc(mediaDrvCtx->pDrmBufMgr, "MEDIA", gmmSize, 4096, params.memType, patIndex, isCpuCacheable);
         params.pitch = gmmPitch;
     }
     else
     {
         unsigned long  ulPitch = 0;
         bo = mos_bo_alloc_tiled(mediaDrvCtx->pDrmBufMgr, "MEDIA", gmmPitch, (gmmSize + gmmPitch -1)/gmmPitch, 1, &params.tileFormat, 
-            (unsigned long *)&ulPitch, 0, params.memType);
+            (unsigned long *)&ulPitch, 0, params.memType, patIndex, isCpuCacheable);
         params.pitch = ulPitch;
     }
 
@@ -1658,8 +1661,11 @@ VAStatus MediaLibvaUtilNext::Allocate2DBuffer(
 
     mem_type = MemoryPolicyManager::UpdateMemoryPolicy(&memPolicyPar);
 
+    unsigned int patIndex = MosInterface::GetPATIndexFromGmm(mediaBuffer->pMediaCtx->pGmmClientContext, gmmResourceInfo);
+    bool isCpuCacheable   = gmmResourceInfo->GetResFlags().Info.Cacheable;
+
     MOS_LINUX_BO  *bo;
-    bo = mos_bo_alloc(bufmgr, "Media 2D Buffer", gmmSize, 4096, mem_type);
+    bo = mos_bo_alloc(bufmgr, "Media 2D Buffer", gmmSize, 4096, mem_type, patIndex, isCpuCacheable);
 
     mediaBuffer->bMapped = false;
     if (bo)
@@ -1731,7 +1737,11 @@ VAStatus MediaLibvaUtilNext::AllocateBuffer(
     memPolicyPar.preferredMemType = mediaBuffer->bUseSysGfxMem ? MOS_MEMPOOL_SYSTEMMEMORY : 0;
 
     mem_type = MemoryPolicyManager::UpdateMemoryPolicy(&memPolicyPar);
-    MOS_LINUX_BO *bo  = mos_bo_alloc(bufmgr, "Media Buffer", size, 4096, mem_type);
+
+    unsigned int patIndex = MosInterface::GetPATIndexFromGmm(mediaBuffer->pMediaCtx->pGmmClientContext, mediaBuffer->pGmmResourceInfo);
+    bool isCpuCacheable   = mediaBuffer->pGmmResourceInfo->GetResFlags().Info.Cacheable;
+
+    MOS_LINUX_BO *bo  = mos_bo_alloc(bufmgr, "Media Buffer", size, 4096, mem_type, patIndex, isCpuCacheable);
     mediaBuffer->bMapped = false;
     if (bo)
     {

diff --git a/media_softlet/linux/common/os/i915/include/mos_bufmgr.h b/media_softlet/linux/common/os/i915/include/mos_bufmgr.h
@@ -170,13 +170,17 @@ struct mos_aub_annotation {
 
 #define BO_ALLOC_FOR_RENDER (1<<0)
 
+#define PAT_INDEX_INVALID ((uint32_t)-1)
+
 struct mos_linux_bo *mos_bo_alloc(struct mos_bufmgr *bufmgr, const char *name,
-                 unsigned long size, unsigned int alignment, int mem_type);
+                 unsigned long size, unsigned int alignment, int mem_type, unsigned int pat_index = PAT_INDEX_INVALID, bool cpu_cacheable = true);
 struct mos_linux_bo *mos_bo_alloc_for_render(struct mos_bufmgr *bufmgr,
                         const char *name,
                         unsigned long size,
                         unsigned int alignment,
-                        int mem_type);
+                        int mem_type,
+                        unsigned int pat_index = PAT_INDEX_INVALID,
+                        bool cpu_cacheable = true);
 struct mos_linux_bo *mos_bo_alloc_userptr(struct mos_bufmgr *bufmgr,
                     const char *name,
                     void *addr, uint32_t tiling_mode,
@@ -188,7 +192,9 @@ struct mos_linux_bo *mos_bo_alloc_tiled(struct mos_bufmgr *bufmgr,
                        uint32_t *tiling_mode,
                        unsigned long *pitch,
                        unsigned long flags,
-                       int mem_type);
+                       int mem_type,
+                       unsigned int pat_index = PAT_INDEX_INVALID,
+                       bool cpu_cacheable = true);
 void mos_bo_reference(struct mos_linux_bo *bo);
 void mos_bo_unreference(struct mos_linux_bo *bo);
 int mos_bo_map(struct mos_linux_bo *bo, int write_enable);

diff --git a/media_softlet/linux/common/os/i915/include/mos_bufmgr_priv.h b/media_softlet/linux/common/os/i915/include/mos_bufmgr_priv.h
@@ -48,7 +48,7 @@ struct mos_bufmgr {
      * using bo_map() or drm_intel_gem_bo_map_gtt() to be used by the CPU.
      */
     struct mos_linux_bo *(*bo_alloc) (struct mos_bufmgr *bufmgr, const char *name,
-                   unsigned long size, unsigned int alignment, int mem_type);
+                   unsigned long size, unsigned int alignment, int mem_type, unsigned int pat_index, bool cpu_cacheable);
 
     /**
      * Allocate a buffer object, hinting that it will be used as a
@@ -60,7 +60,9 @@ struct mos_bufmgr {
                           const char *name,
                           unsigned long size,
                           unsigned int alignment,
-                          int mem_type);
+                          int mem_type,
+                          unsigned int pat_index,
+                          bool cpu_cacheable);
 
     /**
      * Allocate a buffer object from an existing user accessible
@@ -95,7 +97,9 @@ struct mos_bufmgr {
                      uint32_t *tiling_mode,
                      unsigned long *pitch,
                      unsigned long flags,
-                     int mem_type);
+                     int mem_type,
+                     unsigned int pat_index,
+                     bool cpu_cacheable);
 
     /** Takes a reference on a buffer object */
     void (*bo_reference) (struct mos_linux_bo *bo);

diff --git a/media_softlet/linux/common/os/i915/include/uapi/README b/media_softlet/linux/common/os/i915/include/uapi/README
@@ -13,45 +13,55 @@ $ make headers_install INSTALL_HDR_PATH=/path/to/install
 
 The last update was done at the following kernel commit:
 
-commit 1cc064dce4ed0ff111b6d6cb06b3cccf1cba29f5
-Author: Umesh Nerlige Ramappa <[email protected]>
-Date:   Thu Mar 23 15:58:59 2023 -0700
-
-    drm/i915/perf: Add support for OA media units
-
-    MTL introduces additional OA units dedicated to media use cases. Add
-    support for programming these OA units by passing the media engine class
-    and instance parameters.
-
-    UMD specific changes for GPUvis support:
-    https://patchwork.freedesktop.org/patch/522827/?series=114023
-    https://patchwork.freedesktop.org/patch/522822/?series=114023
-    https://patchwork.freedesktop.org/patch/522826/?series=114023
-    https://patchwork.freedesktop.org/patch/522828/?series=114023
-    https://patchwork.freedesktop.org/patch/522816/?series=114023
-    https://patchwork.freedesktop.org/patch/522825/?series=114023
-
-    v2: (Ashutosh)
-    - check for IP_VER(12, 70) instead of MTL
-    - remove PERF_GROUP_OAG comment in mtl_oa_base
-    - remove oa_buffer.group
-    - use engine->oa_group->type in engine_supports_oa_format
-    - remove fw_domains and use FORCEWAKE_ALL
-    - remove MPES/MPEC comment
-    - s/xehp/mtl/ in b counter validation function name
-    - remove engine_supports_oa in __oa_engine_group
-    - remove warn_ON from __oam_engine_group
-    - refactor oa_init_groups and oa_init_regs
-    - assign g->type correctly
-    - use enum oa_type definition
-
-    v3: (Ashutosh)
-    - Drop oa_unit_functional as engine_supports_oa is enough
-
-    v4:
-    - s/DRM_DEBUG/drm_dbg/
-
-    Signed-off-by: Umesh Nerlige Ramappa <[email protected]>
-    Reviewed-by: Ashutosh Dixit <[email protected]>
-    Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
+commit 81b1b599dfd71c958418dad586fa72c8d30d1065
+Author: Fei Yang <[email protected]>
+Date:   Tue Jun 6 12:00:42 2023 +0200
+
+    drm/i915: Allow user to set cache at BO creation
+
+    To comply with the design that buffer objects shall have immutable
+    cache setting through out their life cycle, {set, get}_caching ioctl's
+    are no longer supported from MTL onward. With that change caching
+    policy can only be set at object creation time. The current code
+    applies a default (platform dependent) cache setting for all objects.
+    However this is not optimal for performance tuning. The patch extends
+    the existing gem_create uAPI to let user set PAT index for the object
+    at creation time.
+    The new extension is platform independent, so UMD's can switch to using
+    this extension for older platforms as well, while {set, get}_caching are
+    still supported on these legacy paltforms for compatibility reason.
+    However, since PAT index was not clearly defined for platforms prior to
+    GEN12 (TGL), so we are limiting this externsion to GEN12+ platforms
+    only. See ext_set_pat() in for the implementation details.
+
+    The documentation related to the PAT/MOCS tables is currently available
+    for Tiger Lake here:
+    https://www.intel.com/content/www/us/en/docs/graphics-for-linux/developer-reference/1-0/tiger-lake.html
+
+    The documentation for other platforms is currently being updated.
+
+    BSpec: 45101
+
+    Mesa support has been submitted in this merge request:
+    https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878
+
+    The media driver supprt has bin submitted in this merge request:
+    https://github.com/intel/media-driver/pull/1680
+
+    The IGT test related to this change is
+    igt@gem_create@create-ext-set-pat
+
+    Signed-off-by: Fei Yang <[email protected]>
+    Cc: Chris Wilson <[email protected]>
+    Cc: Matt Roper <[email protected]>
+    Cc: Andi Shyti <[email protected]>
+    Reviewed-by: Andi Shyti <[email protected]>
+    Acked-by: Jordan Justen <[email protected]>
+    Tested-by: Jordan Justen <[email protected]>
+    Acked-by: Carl Zhang <[email protected]>
+    Tested-by: Lihao Gu <[email protected]>
+    Signed-off-by: Andi Shyti <[email protected]>
+    Acked-by: Tvrtko Ursulin <[email protected]>
+    Acked-by: Slawomir Milczarek <[email protected]>
+    Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]