radv: Implement VK_KHR_zero_initialize_workgroup_memory.

Reuses the pass that was implemented for ANV.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8751>

diff --git a/docs/relnotes/new_features.txt b/docs/relnotes/new_features.txt
index b17db63c782f5b3222cfba81637c2e8142f43655..8d594a1dbec1ffd319e6d142f2663f0937d5a11b 100644 (file)
--- a/docs/relnotes/new_features.txt
+++ b/docs/relnotes/new_features.txt
@@ -15,4 +15,4 @@ Sparse memory support on RADV
 Rapid packed math (16bit-vectorization) on RADV
 VK_KHR_workgroup_memory_explicit_layout on Intel, RADV
 DRM format modifiers for AMD.
-VK_KHR_zero_initialize_workgroup_memory on Intel
+VK_KHR_zero_initialize_workgroup_memory on Intel, RADV

diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index fad1f46695d7e1ed3ff7c440520b12dd0d2d8264..73e1797ea299532e1b117653d91f8a4c1ec2b55d 100644 (file)
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -1444,6 +1444,12 @@ void radv_GetPhysicalDeviceFeatures2(
                        features->workgroupMemoryExplicitLayout16BitAccess = true;
                        break;
                }
+               case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ZERO_INITIALIZE_WORKGROUP_MEMORY_FEATURES_KHR: {
+                       VkPhysicalDeviceZeroInitializeWorkgroupMemoryFeaturesKHR *features =
+                               (VkPhysicalDeviceZeroInitializeWorkgroupMemoryFeaturesKHR*)ext;
+                       features->shaderZeroInitializeWorkgroupMemory = true;
+                       break;
+               }
                default:
                        break;
                }

diff --git a/src/amd/vulkan/radv_extensions.py b/src/amd/vulkan/radv_extensions.py
index f618dd9e2d76c92bfdab18ad185e0d1e23b5e56c..3f6373469095b36ff36b281a19113a074be6fca9 100644 (file)
--- a/src/amd/vulkan/radv_extensions.py
+++ b/src/amd/vulkan/radv_extensions.py
@@ -119,6 +119,7 @@ EXTENSIONS = [
     Extension('VK_KHR_workgroup_memory_explicit_layout',  1, True),
     Extension('VK_KHR_xcb_surface',                       6, 'VK_USE_PLATFORM_XCB_KHR'),
     Extension('VK_KHR_xlib_surface',                      6, 'VK_USE_PLATFORM_XLIB_KHR'),
+    Extension('VK_KHR_zero_initialize_workgroup_memory',  1, True),
     Extension('VK_EXT_4444_formats',                      1, True),
     Extension('VK_EXT_acquire_xlib_display',              1, 'VK_USE_PLATFORM_XLIB_XRANDR_EXT'),
     Extension('VK_EXT_buffer_device_address',             2, True),

diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c
index e0660f88631d19fa0875e6b965080aeb2640953a..c25af07df930629d85755c9386e8aed398d148a0 100644 (file)
--- a/src/amd/vulkan/radv_shader.c
+++ b/src/amd/vulkan/radv_shader.c
@@ -681,6 +681,14 @@ radv_shader_compile_to_nir(struct radv_device *device,
                }
                NIR_PASS_V(nir, nir_lower_explicit_io,
                           nir_var_mem_shared, nir_address_format_32bit_offset);
+
+               if (nir->info.cs.zero_initialize_shared_memory &&
+                   nir->info.cs.shared_size > 0) {
+                       const unsigned chunk_size = 16; /* max single store size */
+                       const unsigned shared_size = ALIGN(nir->info.cs.shared_size, chunk_size);
+                       NIR_PASS_V(nir, nir_zero_initialize_shared_memory,
+                                  shared_size, chunk_size);
+               }
        }
 
        nir_lower_explicit_io(nir, nir_var_mem_global,