SWDEV-555889 - Support mipmap on rocr (#2082)

* SWDEV-555889 - Support mipmap on rocr

Support mipmap in hip-rt on rocr backend.
Enable all mipmap tests in Windows.
Some other minor improvement.

Add some SRD logs that will be removed finally.

* Add sampler.mipFilter to fix sampler issues on mipmap in rocr.
Fix format issues of view of leveled image and  mipmap image in blit kernel in rocr.
Enabled disabled mipmap tests.

* Rewrite view logic

* Set word4.f.PITCH = 0 for mipmap SRD on navi31 to fix unstable test issues.
Reset last error in nagative tests.

* Remove SRD dump log from hip-rt
Let Rocr mipmap log be in condition.

* minor format chang

* Exclude mipmap tests for mi200+ which don't support mipmap.

projects/rocr-runtime/runtime/hsa-runtime/core/runtime/hsa_ext_interface.cpp

@@ -503,9 +503,10 @@ hsa_status_t hsa_ext_image_destroy_v2(hsa_agent_t agent, hsa_ext_image_t image)
 hsa_status_t hsa_ext_image_mipmap_array_get_level(hsa_agent_t agent,
                                                   const hsa_ext_image_t* mipmap_array,
                                                   uint32_t mip_level,
+                                                  const hsa_ext_image_descriptor_v2_t* image_descriptor,
                                                   hsa_ext_image_t* level_view) {
   return rocr::core::Runtime::runtime_singleton_->extensions_.image_api
-      .hsa_ext_image_mipmap_array_get_level_fn(agent, mipmap_array, mip_level,
+      .hsa_ext_image_mipmap_array_get_level_fn(agent, mipmap_array, mip_level, image_descriptor,
                                                level_view);
 }
 

projects/rocr-runtime/runtime/hsa-runtime/image/hsa_ext_image.cpp

@@ -266,7 +266,7 @@ hsa_status_t hsa_ext_sampler_create(hsa_agent_t agent,
   }
   hsa_ext_sampler_descriptor_v2_t sampler_descriptor_v2 = {
       sampler_descriptor->coordinate_mode,
-      sampler_descriptor->filter_mode,
+     sampler_descriptor->filter_mode, HSA_EXT_SAMPLER_FILTER_MODE_NONE,
       {sampler_descriptor->address_mode,
           sampler_descriptor->address_mode, sampler_descriptor->address_mode}
   };
@@ -566,13 +566,15 @@ hsa_status_t hsa_ext_image_destroy_v2(hsa_agent_t agent, hsa_ext_image_t image)
 
 // per-level view retrieval implementation
 hsa_status_t HSA_API hsa_ext_image_mipmap_array_get_level(hsa_agent_t agent,
-                                                          const hsa_ext_image_t* mipmapped_array,
-                                                          uint32_t mip_level,
-                                                          hsa_ext_image_t* level_image_out) {
+                                         const hsa_ext_image_t* mipmapped_array,
+                                         uint32_t mip_level,
+                                         const hsa_ext_image_descriptor_v2_t* image_descriptor,
+                                         hsa_ext_image_t* level_image_out) {
   TRY;
   if (!mipmapped_array || !level_image_out) { return HSA_STATUS_ERROR_INVALID_ARGUMENT; }
 
-  return ImageRuntime::instance()->GetMipmapArrayLevelHandle(agent, *mipmapped_array, mip_level, *level_image_out);
+  return ImageRuntime::instance()->GetMipmapArrayLevelHandle(agent, *mipmapped_array, mip_level,
+                                                             image_descriptor, *level_image_out);
 
   CATCH;
 }

projects/rocr-runtime/runtime/hsa-runtime/image/image_manager_ai.cpp

@@ -483,7 +483,17 @@ hsa_status_t ImageManagerAi::PopulateSamplerSrd(Sampler& sampler) const {
   }
   word2.bits.XY_MIN_FILTER = word2.bits.XY_MAG_FILTER;
   word2.bits.Z_FILTER = SQ_TEX_Z_FILTER_NONE;
-  word2.bits.MIP_FILTER = SQ_TEX_MIP_FILTER_NONE;
+
+  switch (sampler_descriptor.mipmap_filter_mode) {
+    case HSA_EXT_SAMPLER_FILTER_MODE_NEAREST:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_POINT);
+      break;
+    case HSA_EXT_SAMPLER_FILTER_MODE_LINEAR:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_LINEAR);
+      break;
+    default:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_NONE);
+  }
 
   word3.u32All = 0;
 
@@ -999,6 +1009,8 @@ hsa_status_t ImageManagerAi::PopulateMipLevelSrd(
     MipmappedArray& level_view,
     const MipmappedArray& mipmap_array,
     uint32_t mip_level) const {
+  // Copy entire parent structure (srd is a fixed array, so it's deep-copied automatically)
+  level_view = mipmap_array;
 
   // SRD already copied from parent, just modify BASE_LEVEL/LAST_LEVEL fields
   uint32_t* srd_words = reinterpret_cast<uint32_t*>(level_view.srd);
@@ -1010,7 +1022,9 @@ hsa_status_t ImageManagerAi::PopulateMipLevelSrd(
   word3->f.base_level = mip_level;
   word3->f.last_level = mip_level;
 
-  debug_print("Set SRD mip selection: BASE_LEVEL=%u, LAST_LEVEL=%u", mip_level, mip_level);
+  if (core::Runtime::runtime_singleton_->flag().image_print_srd()) {
+    debug_print("Set SRD mip selection: BASE_LEVEL=%u, LAST_LEVEL=%u", mip_level, mip_level);
+  }
 
   return HSA_STATUS_SUCCESS;
 }

projects/rocr-runtime/runtime/hsa-runtime/image/image_manager_gfx11.cpp

@@ -601,7 +601,17 @@ hsa_status_t ImageManagerGfx11::PopulateSamplerSrd(Sampler& sampler) const {
   }
   word2.bits.XY_MIN_FILTER = word2.bits.XY_MAG_FILTER;
   word2.bits.Z_FILTER = SQ_TEX_Z_FILTER_NONE;
-  word2.bits.MIP_FILTER = SQ_TEX_MIP_FILTER_NONE;
+
+  switch (sampler_descriptor.mipmap_filter_mode) {
+    case HSA_EXT_SAMPLER_FILTER_MODE_NEAREST:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_POINT);
+      break;
+    case HSA_EXT_SAMPLER_FILTER_MODE_LINEAR:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_LINEAR);
+      break;
+    default:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_NONE);
+  }
 
   word3.u32All = 0;
 
@@ -924,9 +934,8 @@ hsa_status_t ImageManagerGfx11::PopulateMipmapSrd(MipmappedArray& mipmap) const
 
     // For 1d, 2d and 2d-msaa in gfx11 this is pitch-1
     if (!mipmap_array && !mipmap_3d) {
-      word4.f.PITCH = out.pitch - 1;
+      word4.f.PITCH = 0; // mipmap dosesn't support custom pitch, so set it as 0
     }
-
     word5.val = 0;
     word6.val = 0;
     word7.val = 0;
@@ -1199,6 +1208,8 @@ hsa_status_t ImageManagerGfx11::PopulateMipLevelSrd(
     MipmappedArray& level_view,
     const MipmappedArray& mipmap_array,
     uint32_t mip_level) const {
+  // Copy entire parent structure (srd is a fixed array, so it's deep-copied automatically)
+  level_view = mipmap_array;
 
   // SRD already copied from parent, just modify BASE_LEVEL/LAST_LEVEL fields
   uint32_t* srd_words = reinterpret_cast<uint32_t*>(level_view.srd);
@@ -1210,7 +1221,9 @@ hsa_status_t ImageManagerGfx11::PopulateMipLevelSrd(
   word3->f.BASE_LEVEL = mip_level;
   word3->f.LAST_LEVEL = mip_level;
 
-  debug_print("Set SRD mip selection: BASE_LEVEL=%u, LAST_LEVEL=%u", mip_level, mip_level);
+  if (core::Runtime::runtime_singleton_->flag().image_print_srd()) {
+    debug_print("Set SRD mip selection: BASE_LEVEL=%u, LAST_LEVEL=%u", mip_level, mip_level);
+  }
 
   return HSA_STATUS_SUCCESS;
 }

projects/rocr-runtime/runtime/hsa-runtime/image/image_manager_gfx12.cpp

@@ -637,7 +637,17 @@ hsa_status_t ImageManagerGfx12::PopulateSamplerSrd(Sampler& sampler) const {
   }
   word2.bits.XY_MIN_FILTER = word2.bits.XY_MAG_FILTER;
   word2.bits.Z_FILTER = SQ_TEX_Z_FILTER_NONE;
-  word2.bits.MIP_FILTER = SQ_TEX_MIP_FILTER_NONE;
+
+  switch (sampler_descriptor.mipmap_filter_mode) {
+    case HSA_EXT_SAMPLER_FILTER_MODE_NEAREST:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_POINT);
+      break;
+    case HSA_EXT_SAMPLER_FILTER_MODE_LINEAR:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_LINEAR);
+      break;
+    default:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_NONE);
+  }
 
   word3.u32All = 0;
 
@@ -1051,7 +1061,7 @@ hsa_status_t ImageManagerGfx12::PopulateMipmapSrd(MipmappedArray& mipmap) const
 
     // For 1d, 2d and 2d-msaa, fields DEPTH+PITCH_MSB encode pitch-1
     if (!mipmap_array && !mipmap_3d) {
-      uint32_t encPitch = out.pitch - 1;
+      uint32_t encPitch = 0;  // mipmap dosesn't support custom pitch, so set it as 0
       word4.f.DEPTH = encPitch & 0x3fff;           // first 14 bits
       word4.f.PITCH_MSB = (encPitch >> 14) & 0x3;  // last 2 bits
     } else {
@@ -1358,6 +1368,8 @@ hsa_status_t ImageManagerGfx12::PopulateMipLevelSrd(
     MipmappedArray& level_view,
     const MipmappedArray& mipmap_array,
     uint32_t mip_level) const {
+  // Copy entire parent structure (srd is a fixed array, so it's deep-copied automatically)
+  level_view = mipmap_array;
 
   // SRD already copied from parent, just modify BASE_LEVEL/LAST_LEVEL fields
   uint32_t* srd_words = reinterpret_cast<uint32_t*>(level_view.srd);
@@ -1370,7 +1382,9 @@ hsa_status_t ImageManagerGfx12::PopulateMipLevelSrd(
   word1->f.BASE_LEVEL = mip_level;
   word3->f.LAST_LEVEL = mip_level;
 
-  debug_print("Set SRD mip selection: BASE_LEVEL=%u, LAST_LEVEL=%u", mip_level, mip_level);
+  if (core::Runtime::runtime_singleton_->flag().image_print_srd()) {
+    debug_print("Set SRD mip selection: BASE_LEVEL=%u, LAST_LEVEL=%u", mip_level, mip_level);
+  }
 
   return HSA_STATUS_SUCCESS;
 }

projects/rocr-runtime/runtime/hsa-runtime/image/image_manager_nv.cpp

@@ -594,7 +594,17 @@ hsa_status_t ImageManagerNv::PopulateSamplerSrd(Sampler& sampler) const {
   }
   word2.bits.XY_MIN_FILTER = word2.bits.XY_MAG_FILTER;
   word2.bits.Z_FILTER = SQ_TEX_Z_FILTER_NONE;
-  word2.bits.MIP_FILTER = SQ_TEX_MIP_FILTER_NONE;
+
+  switch (sampler_descriptor.mipmap_filter_mode) {
+    case HSA_EXT_SAMPLER_FILTER_MODE_NEAREST:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_POINT);
+      break;
+    case HSA_EXT_SAMPLER_FILTER_MODE_LINEAR:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_LINEAR);
+      break;
+    default:
+      word2.bits.MIP_FILTER = static_cast<int>(SQ_TEX_MIP_FILTER_NONE);
+  }
 
   word3.u32All = 0;
 
@@ -923,7 +933,7 @@ hsa_status_t ImageManagerNv::PopulateMipmapSrd(MipmappedArray& mipmap) const {
     uint32_t minor_ver = MinorVerFromDevID(chip_id_);
     // For 1d, 2d and 2d-msaa in gfx1030 and beyond this is pitch-1
     if ((minor_ver >= 3) && !mipmap_array && !mipmap_3d)
-      word4.f.PITCH = out.pitch - 1;
+      word4.f.PITCH = 0;  // mipmap dosesn't support custom pitch, so set it as 0
 
     word5.val = 0;
     word5.f.MAX_MIP = mipmap.num_levels - 1;
@@ -1119,6 +1129,8 @@ hsa_status_t ImageManagerNv::PopulateMipLevelSrd(
     MipmappedArray& level_view,
     const MipmappedArray& mipmap_array,
     uint32_t mip_level) const {
+  // Copy entire parent structure (srd is a fixed array, so it's deep-copied automatically)
+  level_view = mipmap_array;
 
   // SRD already copied from parent, just modify BASE_LEVEL/LAST_LEVEL fields
   uint32_t* srd_words = reinterpret_cast<uint32_t*>(level_view.srd);
@@ -1130,7 +1142,9 @@ hsa_status_t ImageManagerNv::PopulateMipLevelSrd(
   word3->f.BASE_LEVEL = mip_level;
   word3->f.LAST_LEVEL = mip_level;
 
-  debug_print("Set SRD mip selection: BASE_LEVEL=%u, LAST_LEVEL=%u", mip_level, mip_level);
+  if (core::Runtime::runtime_singleton_->flag().image_print_srd()) {
+    debug_print("Set SRD mip selection: BASE_LEVEL=%u, LAST_LEVEL=%u", mip_level, mip_level);
+  }
 
   return HSA_STATUS_SUCCESS;
 }

projects/rocr-runtime/runtime/hsa-runtime/image/image_runtime.cpp

@@ -771,8 +771,6 @@ hsa_status_t ImageRuntime::CreateMipmapArrayHandle(
     mipmap_array->tile_mode = Image::TileMode::TILED;
   }
 
-  debug_print("Tile mode = %u (0: LINEAR, 1: TILED)", mipmap_array->tile_mode);
-
   // Initialize the mipmapped array object
   mipmap_array->component = component;
   mipmap_array->data = const_cast<void*>(image_data);
@@ -782,15 +780,18 @@ hsa_status_t ImageRuntime::CreateMipmapArrayHandle(
   mipmap_array->flags = 0;
 
   manager->PopulateMipmapSrd(*mipmap_array);
-  debug_print("Populating mipmapped array SRD...");
-  if (core::Runtime::runtime_singleton_->flag().image_print_srd())
-    mipmap_array->printSRD();
-
-  manager->printSRDDetailed(mipmap_array->srd);
 
   // assert(mipmap_array->size == required_size);
   image_handle.handle = mipmap_array->Convert();
-  debug_print("output handle = %lu", image_handle.handle);
+
+  if (core::Runtime::runtime_singleton_->flag().image_print_srd()) {
+    debug_print("Tile mode = %u (0: LINEAR, 1: TILED)", mipmap_array->tile_mode);
+    debug_print("Populating mipmapped array SRD...");
+    mipmap_array->printSRD();
+    manager->printSRDDetailed(mipmap_array->srd);
+    debug_print("output handle = %lu", image_handle.handle);
+  }
+
   return HSA_STATUS_SUCCESS;
 }
 
@@ -809,7 +810,12 @@ hsa_status_t ImageRuntime::DestroyMipmapArrayHandle(
 
 hsa_status_t ImageRuntime::GetMipmapArrayLevelHandle(
     hsa_agent_t component, const hsa_ext_image_t& mipmapped_array,
-    uint32_t mip_level, hsa_ext_image_t& level_image_out) {
+    uint32_t mip_level, const hsa_ext_image_descriptor_v2_t* image_descriptor,
+    hsa_ext_image_t& level_image_out) {
+  ImageManager * manager = image_manager(component);
+  if (!manager) {
+    return HSA_STATUS_ERROR_OUT_OF_RESOURCES;
+  }
 
   level_image_out.handle = 0;
 
@@ -837,35 +843,42 @@ hsa_status_t ImageRuntime::GetMipmapArrayLevelHandle(
       return HSA_STATUS_ERROR_INVALID_ARGUMENT;
   }
 
-  debug_print("Creating mip level %u view for %u level mipmap\n",
+  if (core::Runtime::runtime_singleton_->flag().image_print_srd()) {
+    debug_print("Creating mip level %u view for %u level mipmap\n",
               mip_level, array->num_levels);
+  }
 
   // Create a view that references the parent mipmap array
   MipmappedArray* level_view = MipmappedArray::Create(component);
   if (!level_view) return HSA_STATUS_ERROR_OUT_OF_RESOURCES;
 
-  // Copy entire parent structure (srd is a fixed array, so it's deep-copied automatically)
-  *level_view = *array;
-
-  // Modify SRD to select only the specific mip level
-  ImageManager* manager = image_manager(component);
-  if (!manager) {
-    MipmappedArray::Destroy(level_view);
-    return HSA_STATUS_ERROR_OUT_OF_RESOURCES;
+  auto format = image_descriptor ? &image_descriptor->format : nullptr;
+   if (format &&
+       (array->desc.format.channel_type != format->channel_type ||
+        array->desc.format.channel_order != format->channel_order)) {
+    MipmappedArray tempArray = *array;
+    tempArray.desc.format.channel_type = format->channel_type;
+    tempArray.desc.format.channel_order = format->channel_order;
+    status = manager->PopulateMipmapSrd(tempArray);
+    if (status == HSA_STATUS_SUCCESS) {
+      status = manager->PopulateMipLevelSrd(*level_view, tempArray, mip_level);
+    }
+    else {
+      debug_print("PopulateMipmapSrd() failed with status %d", status);
+    }
+  }
+  else {
+    status = manager->PopulateMipLevelSrd(*level_view, *array, mip_level);
   }
-
-  status = manager->PopulateMipLevelSrd(*level_view, *array, mip_level);
   if (status != HSA_STATUS_SUCCESS) {
     MipmappedArray::Destroy(level_view);
     return status;
   }
 
-  debug_print("Created mip level view using SRD fields");
-  if (core::Runtime::runtime_singleton_->flag().image_print_srd())
+  if (core::Runtime::runtime_singleton_->flag().image_print_srd()) {
     level_view->printSRD();
-
-  manager->printSRDDetailed(level_view->srd);
-
+    manager->printSRDDetailed(level_view->srd);
+  }
   // Return handle
   level_image_out.handle = level_view->Convert();
   return HSA_STATUS_SUCCESS;

projects/rocr-runtime/runtime/hsa-runtime/image/image_runtime.h

@@ -171,7 +171,8 @@ class ImageRuntime {
   /// @brief Get the handle for a specific mipmap level in a mipmapped array.
   hsa_status_t GetMipmapArrayLevelHandle(
     hsa_agent_t agent, const hsa_ext_image_t& mipmapped_array,
-    uint32_t mip_level, hsa_ext_image_t& level_image_out);
+    uint32_t mip_level, const hsa_ext_image_descriptor_v2_t* image_descriptor,
+    hsa_ext_image_t& level_image_out);
 
   ImageManager* image_manager(hsa_agent_t agent) {
     std::map<uint64_t, ImageManager*>::iterator it = image_managers_.find(agent.handle);

projects/rocr-runtime/runtime/hsa-runtime/image/resource.h

@@ -223,9 +223,9 @@ private:
     tile_mode = LINEAR;
   }
 
-  ~MipmappedArray() {}
-
 public:
+ ~MipmappedArray() {}
+
  /// @brief Create a MipmappedArray.
  /// Only internal metadata is allocated; image data must be provided by the user.
  static MipmappedArray* Create(hsa_agent_t agent);

projects/rocr-runtime/runtime/hsa-runtime/inc/hsa_ext_image.h

@@ -1053,6 +1053,7 @@ hsa_status_t HSA_API hsa_ext_image_destroy_v2(hsa_agent_t agent, hsa_ext_image_t
 hsa_status_t HSA_API hsa_ext_image_mipmap_array_get_level(hsa_agent_t agent,
                                       const hsa_ext_image_t* mipmapped_array,
                                       uint32_t mip_level,
+                                      const hsa_ext_image_descriptor_v2_t* image_descriptor,
                                       hsa_ext_image_t* level_image_out);
 
 /**
@@ -1381,7 +1382,12 @@ typedef enum {
    * square block or 2x2x2 cube block around the specified coordinate. The
    * elements are combined using linear interpolation.
    */
-  HSA_EXT_SAMPLER_FILTER_MODE_LINEAR = 1
+  HSA_EXT_SAMPLER_FILTER_MODE_LINEAR = 1,
+
+  /**
+   * None filter. Used for mipmap filter mode of non-mipmap images.
+   */
+  HSA_EXT_SAMPLER_FILTER_MODE_NONE = 2
 
 } hsa_ext_sampler_filter_mode_t;
 
@@ -1422,10 +1428,15 @@ typedef struct hsa_ext_sampler_descriptor_v2_s {
   hsa_ext_sampler_coordinate_mode32_t coordinate_mode;
 
   /**
-   * Sampler filter type describes the type of sampling performed.
+   * Sampler filter type describes the type of sampling performed on regular image.
    */
   hsa_ext_sampler_filter_mode32_t filter_mode;
 
+  /**
+   * Sampler filter type describes the type of sampling performed on mipmap image.
+  */
+  hsa_ext_sampler_filter_mode32_t mipmap_filter_mode;
+
   /**
    * Sampler address mode describes the processing of out-of-range image
    * coordinates.