Fix #131091: GL: Weird Lines appearing in Gizmo Overlays

This happened because NVidia GPUs require higher alignment
for SSBO binds than for vertex inputs.

This is related to	#131103 which fixed it for vulkan.

Add a common capability option for that.

source/blender/gpu/GPU_capabilities.hh

@@ -34,6 +34,9 @@ int GPU_max_shader_storage_buffer_bindings();
 int GPU_max_compute_shader_storage_blocks();
 int GPU_max_samplers();
 size_t GPU_max_storage_buffer_size();
+/* Used when binding subrange of SSBOs. In bytes.
+ * The start of the range must be aligned with this value. */
+size_t GPU_storage_buffer_alignment();
 
 int GPU_extensions_len();
 const char *GPU_extension_get(int i);

source/blender/gpu/intern/gpu_capabilities.cc

@@ -227,6 +227,11 @@ size_t GPU_max_storage_buffer_size()
   return GCaps.max_storage_buffer_size;
 }
 
+size_t GPU_storage_buffer_alignment()
+{
+  return GCaps.storage_buffer_alignment;
+}
+
 /** \} */
 
 /* -------------------------------------------------------------------- */

source/blender/gpu/intern/gpu_capabilities_private.hh

@@ -40,6 +40,7 @@ struct GPUCapabilities {
   int max_shader_storage_buffer_bindings = 0;
   int max_compute_shader_storage_blocks = 0;
   size_t max_storage_buffer_size = 0;
+  size_t storage_buffer_alignment = 256;
   int extensions_len = 0;
   const char *(*extension_get)(int);
 

source/blender/gpu/metal/mtl_backend.mm

@@ -498,6 +498,7 @@ void MTLBackend::capabilities_init(MTLContext *ctx)
    * Can use argument buffers if a higher limit is required. */
   GCaps.max_shader_storage_buffer_bindings = 14;
   GCaps.max_storage_buffer_size = size_t(ctx->device.maxBufferLength);
+  GCaps.storage_buffer_alignment = 256; /* TODO(fclem): But also unused. */
 
   GCaps.max_work_group_count[0] = 65535;
   GCaps.max_work_group_count[1] = 65535;

source/blender/gpu/opengl/gl_backend.cc

@@ -583,6 +583,9 @@ void GLBackend::capabilities_init()
   int64_t max_ssbo_size;
   glGetInteger64v(GL_MAX_SHADER_STORAGE_BLOCK_SIZE, &max_ssbo_size);
   GCaps.max_storage_buffer_size = size_t(max_ssbo_size);
+  GLint ssbo_alignment;
+  glGetIntegerv(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT, &ssbo_alignment);
+  GCaps.storage_buffer_alignment = size_t(ssbo_alignment);
 
   GCaps.transform_feedback_support = true;
   GCaps.texture_view_support = epoxy_gl_version() >= 43 ||

source/blender/gpu/opengl/gl_immediate.cc

@@ -8,6 +8,8 @@
  * Mimics old style opengl immediate mode drawing.
  */
 
+#include "GPU_capabilities.hh"
+
 #include "gpu_context_private.hh"
 #include "gpu_shader_private.hh"
 #include "gpu_vertex_format_private.hh"
@@ -96,9 +98,14 @@ uchar *GLImmediate::begin()
     recreate_buffer = true;
   }
 
-  /* ensure vertex data is aligned */
-  /* Might waste a little space, but it's safe. */
-  const uint pre_padding = padding(buffer_offset(), vertex_format.stride);
+  uint vert_alignment = vertex_format.stride;
+  if (unwrap(this->shader)->is_polyline) {
+    /* Polyline needs to bind the buffer as SSBO.
+     * The start of the range needs to match the SSBO alignment requirements. */
+    vert_alignment = max_uu(vert_alignment, GPU_storage_buffer_alignment());
+  }
+  /* Ensure vertex data is aligned. Might waste a little space, but it's safe. */
+  const uint pre_padding = padding(buffer_offset(), vert_alignment);
 
   if (!recreate_buffer && ((bytes_needed + pre_padding) <= available_bytes)) {
     buffer_offset() += pre_padding;

source/blender/gpu/vulkan/vk_backend.cc

@@ -573,6 +573,7 @@ void VKBackend::capabilities_init(VKDevice &device)
   GCaps.max_shader_storage_buffer_bindings = GCaps.max_compute_shader_storage_blocks = min_uu(
       limits.maxPerStageDescriptorStorageBuffers, INT_MAX);
   GCaps.max_storage_buffer_size = size_t(limits.maxStorageBufferRange);
+  GCaps.storage_buffer_alignment = limits.minStorageBufferOffsetAlignment;
 
   GCaps.max_parallel_compilations = BLI_system_thread_count();
   GCaps.mem_stats_support = true;

source/blender/gpu/vulkan/vk_immediate.cc

@@ -8,11 +8,13 @@
  * Mimics old style OpenGL immediate mode drawing.
  */
 
-#include "vk_immediate.hh"
+#include "GPU_capabilities.hh"
+
 #include "vk_backend.hh"
 #include "vk_context.hh"
 #include "vk_data_conversion.hh"
 #include "vk_framebuffer.hh"
+#include "vk_immediate.hh"
 #include "vk_state_manager.hh"
 
 #include "CLG_log.h"
@@ -51,8 +53,7 @@ uchar *VKImmediate::begin()
   vertex_format_converter.init(&vertex_format, workarounds);
   const size_t bytes_needed = vertex_buffer_size(&vertex_format_converter.device_format_get(),
                                                  vertex_len);
-  VkDeviceSize offset_alignment =
-      device.physical_device_properties_get().limits.minStorageBufferOffsetAlignment;
+  size_t offset_alignment = GPU_storage_buffer_alignment();
   VKBuffer &buffer = ensure_space(bytes_needed, offset_alignment);
 
   /* Apply alignment when allocating new sub buffer, to reduce signed/unsigned data conversion