anv/query: Use snooping on !LLC platforms

Commit b2c97bc789198427043cd902bc76e194e7e81c7d which made us start
using a busy-wait for individual query results also messed up cache
flushing on !LLC platforms.  For one thing, I forgot the mfence after
the clflush so memory access wasn't properly getting fenced.  More
importantly, however, was that we were clflushing the whole query range
and then waiting for individual queries and then trying to read the
results without clflushing again.  Getting the clflushing both correct
and efficient is very subtle and painful.  Instead, let's side-step the
problem by just snooping.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

diff --git a/src/intel/vulkan/genX_query.c b/src/intel/vulkan/genX_query.c
index 235da8b8b6ec05ad5f10eb3450c16970e28f56a4..2c70b4f528e58d47976388f12c59fd66d3863bcc 100644 (file)
--- a/src/intel/vulkan/genX_query.c
+++ b/src/intel/vulkan/genX_query.c
@@ -90,6 +90,14 @@ VkResult genX(CreateQueryPool)(
    if (result != VK_SUCCESS)
       goto fail;
 
+   /* For query pools, we set the caching mode to I915_CACHING_CACHED.  On LLC
+    * platforms, this does nothing.  On non-LLC platforms, this means snooping
+    * which comes at a slight cost.  However, the buffers aren't big, won't be
+    * written frequently, and trying to handle the flushing manually without
+    * doing too much flushing is extremely painful.
+    */
+   anv_gem_set_caching(device, pool->bo.gem_handle, I915_CACHING_CACHED);
+
    pool->bo.map = anv_gem_mmap(device, pool->bo.gem_handle, 0, size, 0);
 
    *pQueryPool = anv_query_pool_to_handle(pool);
@@ -132,11 +140,8 @@ cpu_write_query_result(void *dst_slot, VkQueryResultFlags flags,
 }
 
 static bool
-query_is_available(struct anv_device *device, uint64_t *slot)
+query_is_available(uint64_t *slot)
 {
-   if (!device->info.has_llc)
-      __builtin_ia32_clflush(slot);
-
    return *(volatile uint64_t *)slot;
 }
 
@@ -145,7 +150,7 @@ wait_for_available(struct anv_device *device,
                    struct anv_query_pool *pool, uint64_t *slot)
 {
    while (true) {
-      if (query_is_available(device, slot))
+      if (query_is_available(slot))
          return VK_SUCCESS;
 
       int ret = anv_gem_busy(device, pool->bo.gem_handle);
@@ -159,7 +164,7 @@ wait_for_available(struct anv_device *device,
       } else {
          assert(ret == 0);
          /* The BO is no longer busy. */
-         if (query_is_available(device, slot)) {
+         if (query_is_available(slot)) {
             return VK_SUCCESS;
          } else {
             VkResult status = anv_device_query_status(device);
@@ -204,13 +209,6 @@ VkResult genX(GetQueryPoolResults)(
 
    void *data_end = pData + dataSize;
 
-   if (!device->info.has_llc) {
-      uint64_t offset = firstQuery * pool->stride;
-      uint64_t size = queryCount * pool->stride;
-      anv_invalidate_range(pool->bo.map + offset,
-                           MIN2(size, pool->bo.size - offset));
-   }
-
    VkResult status = VK_SUCCESS;
    for (uint32_t i = 0; i < queryCount; i++) {
       uint64_t *slot = pool->bo.map + (firstQuery + i) * pool->stride;