MesaLib: update to 21.3.7

Of particular note is this change in Mesa 21.2:
  "32-bit x86 builds now default disable x87 math and use sse2."
  This might need to be revisited in pkgsrc.

Bugfixes omitted from ChangeLog.

New features in Mesa 21.3:
  VK_EXT_color_write_enable on lavapipe
  GL_ARB_texture_filter_anisotropic in llvmpipe
  Anisotropic texture filtering in lavapipe
  VK_EXT_shader_atomic_float2 on Intel and RADV.
  VK_EXT_vertex_input_dynamic_state on RADV.
  VK_KHR_timeline_semaphore on lavapipe
  VK_EXT_external_memory_host on lavapipe
  GL_AMD_pinned_memory on llvmpipe
  GL 4.5 compatibility on llvmpipe
  VK_EXT_primitive_topology_list_restart on RADV and lavapipe.
  ES 3.2 on zink
  VK_KHR_depth_stencil_resolve on lavapipe
  VK_KHR_shader_integer_dot_product on RADV.
  OpenGL FP16 support on llvmpipe
  VK_KHR_shader_float16_int8 on lavapipe
  VK_KHR_shader_subgroup_extended_types on lavapipe
  VK_KHR_spirv_1_4 on lavapipe
  Experimental raytracing support on RADV
  VK_KHR_synchronization2 on Intel
  NGG shader based culling is now enabled by default on GFX10.3 on RADV.
  VK_KHR_maintenance4 on RADV
  VK_KHR_format_feature_flags2 on RADV.
  EGL_EXT_present_opaque on wayland

New features in Mesa 21.2:
  zink supports GL_ARB_texture_filter_minmax, GL_ARB_shader_clock
  VK_EXT_provoking_vertex on RADV.
  VK_EXT_extended_dynamic_state2 on RADV.
  VK_EXT_global_priority_query on RADV.
  VK_EXT_physical_device_drm on RADV.
  VK_KHR_shader_subgroup_uniform_control_flow on Intel and RADV.
  VK_EXT_color_write_enable on RADV.
  32-bit x86 builds now default disable x87 math and use sse2.
  GL ES 3.1 on GT21x hardware.
  VK_EXT_acquire_drm_display on RADV and ANV.
  VK_EXT_vertex_input_dynamic_state on lavapipe
  wideLines on lavapipe
  VK_EXT_line_rasterization on lavapipe
  VK_EXT_multi_draw on ANV, lavapipe, and RADV
  VK_KHR_separate_depth_stencil_layouts on lavapipe
  VK_EXT_separate_stencil_usage on lavapipe
  VK_EXT_extended_dynamic_state2 on lavapipe
  NGG shader based primitive culling is now supported by RADV.
  Panfrost supports OpenGL ES 3.1
  New Asahi driver for the Apple M1
  GL_ARB_sample_locations on zink
  GL_ARB_sparse_buffer on zink
  GL_ARB_shader_group_vote on zink
  DRM format modifiers on zink
  freedreno+turnip: Initial support for a6xx gen4 (a660, a635)

New features in Mesa 21.1:
  VK_KHR_workgroup_memory_explicit_layout on Intel, RADV
  DRM format modifiers for AMD.
  VK_KHR_zero_initialize_workgroup_memory on Intel, RADV
  Zink exposes GL 4.6 and ES 3.1
  GL_EXT_depth_bounds_test on softpipe, zink
  GL_EXT_texture_filter_minmax on nvc0 (gm200+)
  GL_ARB_texture_filter_minmax on nvc0 (gm200+)
  GL_ARB_post_depth_coverage on zink
  VK_KHR_copy_commands2 on lavapipe
  lavapipe exposes Vulkan 1.1
  VRS attachment on RADV

New features in Mesa 21.0:
  GL_EXT_demote_to_helper_invocation on radeonsi
  GL_NV_compute_shader_derivatives on radeonsi
  EGL_MESA_platform_xcb
  Removed GL_NV_point_sprite for classic swrast.
  driconf: remove glx_disable_oml_sync_control, glx_disable_sgi_video_sync, and glx_disable_ext_buffer_age
  Removed support for loading DRI drivers older than Mesa 8.0, including all DRI1 support
  Add support for VK_VALVE_mutable_descriptor_type on RADV
  Removed classic OSMesa in favor of the newly improved gallium OSMesa
  VK_KHR_fragment_shading_rate on RADV (RDNA2 only)
  Freedreno a6xx exposes GL 3.3
  Classic swrast dri driver removed in favor of gallium swrast (llvmpipe or softpipe)
  Panfrost g31/g52/g72 exposes ES 3.0
  Panfrost t760+ exposes GL 3.1 (including on Bifrost)
  Sparse memory support on RADV
  Rapid packed math (16bit-vectorization) on RADV


New features in Mesa 20.3:
  GL 4.5 on llvmpipe
  GL_INTEL_blackhole_render on radeonsi
  GL_NV_copy_depth_to_color for NIR
  GL_NV_half_float
  GL_NV_shader_atomic_int64 on radeonsi
  EGL_KHR_swap_buffers_with_damage on X11 (DRI3)
  VK_PRESENT_MODE_FIFO_RELAXED on X11
  GLX_EXT_swap_control for DRI2 and DRI3
  GLX_EXT_swap_control_tear for DRI3
  VK_KHR_copy_commands2 on RADV
  VK_KHR_shader_terminate_invocation on RADV
  NGG GS support in ACO
  VK_KHR_shader_terminate_invocation on ANV
  driconf: add glx_extension_override
  driconf: add indirect_gl_extension_override
  VK_AMD_mixed_attachment_samples on RADV (GFX6-GFX7).
  GL_MESA_pack_invert on r100 and vieux
  GL_ANGLE_pack_reverse_row_order
  VK_EXT_shader_image_atomic_int64 on RADV

New features in mesa 20.2:
  GL_ARB_compute_variable_group_size on Iris.
  GL_ARB_gpu_shader5 on llvmpipe
  GL_ARB_post_depth_coverage on llvmpipe
  GLES 3.2 on llvmpipe
  GL_EXT_shader_group_vote on GLES3.
  GL_EXT_texture_shadow_lod on llvmpipe
  VK_AMD_texture_gather_bias_lod on RADV.
  VK_AMD_gpu_shader_half_float on RADV/ACO.
  VK_AMD_gpu_shader_int16 on RADV/ACO.
  VK_EXT_extended_dynamic_state on ANV and RADV.
  VK_EXT_image_robustness on RADV.
  VK_EXT_private_data on ANV and RADV.
  VK_EXT_custom_border_color on ANV and RADV.
  VK_EXT_pipeline_creation_cache_control on ANV and RADV.
  VK_EXT_shader_demote_to_helper_invocation on RADV/LLVM.
  VK_EXT_subgroup_size_control on RADV/ACO.
  VK_GOOGLE_user_type on ANV and RADV.
  VK_KHR_shader_subgroup_extended_types on RADV/ACO.
  GL_ARB_gl_spirv on nvc0/nir.
  GL_ARB_spirv_extensions on nvc0/nir.
  RADV now uses ACO per default as backend
  RADV_DEBUG=llvm option to enable LLVM backend for RADV
  VK_EXT_image_robustness for ANV
  VK_EXT_shader_atomic_float on ANV
  VK_EXT_4444_formats on ANV and RADV.
  VK_KHR_memory_model on RADV.
  GL 4.5 on llvmpipe
  EGL_KHR_swap_buffers_with_damage on X11 (DRI3)

New features in mesa 20.1:
  GL_ARB_compute_variable_group_size on i965.
  GL_EXT_depth_bounds_test on Iris.
  GL_EXT_texture_shadow_lod on radeonsi, nvc0.
  GL_NV_alpha_to_coverage_dither_control on radeonsi
  GL_NV_copy_image on all gallium drivers.
  GL_NV_pixel_buffer_object on all gallium drivers, i915, i965, swrast.
  GL_NV_viewport_array2 on nvc0 (GM200+).
  GL_NV_viewport_swizzle on nvc0 (GM200+).
  VK_AMD_memory_overallocation_behavior on RADV.
  VK_KHR_shader_non_semantic_info on Intel, RADV.
  GL_EXT_draw_instanced on gles2
  VK_KHR_8bit_storage for ACO on GFX8+
  VK_KHR_16bit_storage for ACO on GFX8+ (storageInputOutput16 is still unsupported)
  shaderInt16 for ACO on GFX9+
  VK_KHR_shader_float16_int8 for ACO on GFX8+ (shaderFloat16 is still unsupported)
  VK_EXT_robustness2 on Intel, RADV.
  Add Rocket Lake (RKL) support on anvil and iris.

1 files changed, 41 insertions, 67 deletions

diff --git a/graphics/MesaLib/patches/patch-src_util_u__thread.h b/graphics/MesaLib/patches/patch-src_util_u__thread.h
index 64db83fb096..d4e6882b489 100644
--- a/graphics/MesaLib/patches/patch-src_util_u__thread.h
+++ b/graphics/MesaLib/patches/patch-src_util_u__thread.h
@@ -1,95 +1,69 @@
-$NetBSD: patch-src_util_u__thread.h,v 1.4 2020/03/08 10:35:03 tnn Exp $
+$NetBSD: patch-src_util_u__thread.h,v 1.5 2022/03/13 15:50:05 tnn Exp $
 
 Oracle Solaris has pthread_setname_np. illumos does not.
+(actually, newer Illumos does have it, but never mind.)
 
 Don't hard error when there's no pthread_setname_np.
 
 handle NetBSD-style pthread_setaffinity_np(3)
 
---- src/util/u_thread.h.orig	2020-03-05 21:34:32.000000000 +0000
+--- src/util/u_thread.h.orig	2021-08-04 18:49:29.374474500 +0000
 +++ src/util/u_thread.h
-@@ -73,7 +73,7 @@ static inline thrd_t u_thread_create(int
+@@ -129,7 +129,7 @@ static inline thrd_t u_thread_create(int
  static inline void u_thread_setname( const char *name )
  {
  #if defined(HAVE_PTHREAD)
 -#if DETECT_OS_LINUX || DETECT_OS_CYGWIN || DETECT_OS_SOLARIS
 +#if DETECT_OS_LINUX || DETECT_OS_CYGWIN
-    pthread_setname_np(pthread_self(), name);
- #elif DETECT_OS_FREEBSD || DETECT_OS_OPENBSD
-    pthread_set_name_np(pthread_self(), name);
-@@ -104,6 +104,17 @@ static inline void
- util_pin_thread_to_L3(thrd_t thread, unsigned L3_index, unsigned cores_per_L3)
+    int ret = pthread_setname_np(pthread_self(), name);
+    if (ret == ERANGE) {
+       char buf[16];
+@@ -169,8 +169,32 @@ util_set_thread_affinity(thrd_t thread,
+                          unsigned num_mask_bits)
  {
  #if defined(HAVE_PTHREAD_SETAFFINITY)
-+#if defined(__NetBSD__)
-+   cpuset_t *cpuset;
+-   cpu_set_t cpuset;
++# if defined(__NetBSD__)
++   cpu_set_t *cpuset;
 +   cpuset = cpuset_create();
 +   if (cpuset == NULL)
 +      return;
-+   cpuset_zero(cpuset);
-+   for (unsigned i = 0; i < cores_per_L3; i++)
-+      cpuset_set(L3_index * cores_per_L3 + i, cpuset);
-+   pthread_setaffinity_np(thread, cpuset_size(cpuset), cpuset);
-+   cpuset_destroy(cpuset);
-+#else
-    cpu_set_t cpuset;
++   if (old_mask) {
++      if (pthread_getaffinity_np(thread, cpuset_size(cpuset), cpuset) != 0)
++         return false;
  
-    CPU_ZERO(&cpuset);
-@@ -111,6 +122,7 @@ util_pin_thread_to_L3(thrd_t thread, uns
-       CPU_SET(L3_index * cores_per_L3 + i, &cpuset);
-    pthread_setaffinity_np(thread, sizeof(cpuset), &cpuset);
- #endif
-+#endif
- }
- 
- /**
-@@ -124,6 +136,35 @@ static inline int
- util_get_L3_for_pinned_thread(thrd_t thread, unsigned cores_per_L3)
- {
- #if defined(HAVE_PTHREAD_SETAFFINITY)
-+#if defined(__NetBSD__)
-+   cpuset_t *cpuset;
-+
-+   cpuset = cpuset_create();
-+   if (cpuset == NULL)
-+      return -1;
-+
-+   if (pthread_getaffinity_np(thread, cpuset_size(cpuset), cpuset) == 0) {
-+      int L3_index = -1;
-+
-+      for (unsigned i = 0; i < cpuset_size(cpuset); i++) {
-+         if (cpuset_isset(i, cpuset)) {
-+            int x = i / cores_per_L3;
-+
-+            if (L3_index != x) {
-+               if (L3_index == -1)
-+                  L3_index = x;
-+               else {
-+                  cpuset_destroy(cpuset);
-+                  return -1; /* multiple L3s are set */
-+               }
-+            }
-+         }
++      memset(old_mask, 0, num_mask_bits / 8);
++      for (unsigned i = 0; i < num_mask_bits && i < CPU_SETSIZE; i++) {
++         if (cpuset_isset(i, cpuset))
++            old_mask[i / 32] |= 1u << (i % 32);
 +      }
-+      cpuset_destroy(cpuset);
-+      return L3_index;
 +   }
++
++   cpuset_zero(cpuset);
++   for (unsigned i = 0; i < num_mask_bits && i < CPU_SETSIZE; i++) {
++      if (mask[i / 32] & (1u << (i % 32)))
++         cpuset_set(i, cpuset);
++   }
++   int err = pthread_setaffinity_np(thread, cpuset_size(cpuset), cpuset);
 +   cpuset_destroy(cpuset);
-+#else
-    cpu_set_t cpuset;
- 
-    if (pthread_getaffinity_np(thread, sizeof(cpuset), &cpuset) == 0) {
-@@ -144,6 +185,7 @@ util_get_L3_for_pinned_thread(thrd_t thr
-       return L3_index;
++   return err == 0;
++# else
++   cpu_set_t cpuset;
+    if (old_mask) {
+       if (pthread_getaffinity_np(thread, sizeof(cpuset), &cpuset) != 0)
+          return false;
+@@ -188,7 +212,7 @@ util_set_thread_affinity(thrd_t thread,
+          CPU_SET(i, &cpuset);
     }
- #endif
-+#endif
-    return -1;
- }
+    return pthread_setaffinity_np(thread, sizeof(cpuset), &cpuset) == 0;
+-
++# endif
+ #elif defined(_WIN32) && !defined(__CYGWIN__)
+    DWORD_PTR m = mask[0];
  
-@@ -155,7 +197,7 @@ util_get_L3_for_pinned_thread(thrd_t thr
+@@ -242,7 +266,7 @@ util_set_current_thread_affinity(const u
  static inline int64_t
- u_thread_get_time_nano(thrd_t thread)
+ util_thread_get_time_nano(thrd_t thread)
  {
 -#if defined(HAVE_PTHREAD) && !defined(__APPLE__) && !defined(__HAIKU__)
 +#if defined(HAVE_PTHREAD) && !defined(__APPLE__) && !defined(__HAIKU__) && !defined(__sun)