GL_ARB_shader_bit_encoding DONE
GL_ARB_texture_rgb10_a2ui DONE (i965, r600)
GL_ARB_texture_swizzle DONE (same as EXT version) (i965, r300, r600, swrast)
-GL_ARB_timer_query DONE
+GL_ARB_timer_query DONE (i965)
GL_ARB_instanced_arrays DONE (i965, r300, r600)
GL_ARB_vertex_type_2_10_10_10_rev DONE (r600)
#include "main/imports.h"
#include "brw_context.h"
+#include "brw_defines.h"
#include "brw_state.h"
#include "intel_batchbuffer.h"
#include "intel_reg.h"
query->Base.Result += 1000 * ((results[1] >> 32) - (results[0] >> 32));
break;
+ case GL_TIMESTAMP:
+ if (intel->gen >= 6) {
+ /* Our timer is a clock that increments every 80ns (regardless of
+ * other clock scaling in the system). The timestamp register we can
+ * read for glGetTimestamp() masks out the top 32 bits, so we do that
+ * here too to let the two counters be compared against each other.
+ *
+ * If we just multiplied that 32 bits of data by 80, it would roll
+ * over at a non-power-of-two, so an application couldn't use
+ * GL_QUERY_COUNTER_BITS to handle rollover correctly. Instead, we
+ * report 36 bits and truncate at that (rolling over 5 times as often
+ * as the HW counter), and when the 32-bit counter rolls over, it
+ * happens to also be at a rollover in the reported value from near
+ * (1<<36) to 0.
+ *
+ * The low 32 bits rolls over in ~343 seconds. Our 36-bit result
+ * rolls over every ~69 seconds.
+ */
+ query->Base.Result = 80 * (results[1] & 0xffffffff);
+ query->Base.Result &= (1ull << 36) - 1;
+ } else {
+ query->Base.Result = 1000 * (results[1] >> 32);
+ }
+
+ break;
+
case GL_SAMPLES_PASSED_ARB:
/* Map and count the pixels from the current query BO */
for (i = query->first_index; i <= query->last_index; i++) {
struct brw_query_object *query = (struct brw_query_object *)q;
switch (query->Base.Target) {
+ case GL_TIMESTAMP:
+ drm_intel_bo_unreference(query->bo);
+ query->bo = drm_intel_bo_alloc(intel->bufmgr, "timer query",
+ 4096, 4096);
+ /* FALLTHROUGH */
+
case GL_TIME_ELAPSED_EXT:
write_timestamp(intel, query->bo, 1);
intel_batchbuffer_flush(intel);
brw->query.index++;
}
+static uint64_t
+brw_get_timestamp(struct gl_context *ctx)
+{
+ struct intel_context *intel = intel_context(ctx);
+ uint64_t result = 0;
+
+ drm_intel_reg_read(intel->bufmgr, TIMESTAMP, &result);
+
+ /* See logic in brw_queryobj_get_results() */
+ result = result >> 32;
+ result *= 80;
+ result &= (1ull << 36) - 1;
+
+ return result;
+}
+
void brw_init_queryobj_functions(struct dd_function_table *functions)
{
functions->NewQueryObject = brw_new_query_object;
functions->EndQuery = brw_end_query;
functions->CheckQuery = brw_check_query;
functions->WaitQuery = brw_wait_query;
+ functions->GetTimestamp = brw_get_timestamp;
}
#include "intel_chipset.h"
#include "intel_context.h"
#include "intel_extensions.h"
+#include "intel_reg.h"
#include "utils.h"
/**
if (intel->gen >= 5)
ctx->Extensions.EXT_timer_query = true;
+ if (intel->gen >= 6) {
+ uint64_t dummy;
+ /* Test if the kernel has the ioctl. */
+ if (drm_intel_reg_read(intel->bufmgr, TIMESTAMP, &dummy) == 0)
+ ctx->Extensions.ARB_timer_query = true;
+ }
+
if (intel->gen >= 4) {
ctx->Extensions.ARB_color_buffer_float = true;
ctx->Extensions.ARB_depth_buffer_float = true;