llvmpipe: refuse float16 formats for sampling/render in favor of float32

This should be faster, and the state tracker will automatically fall back
to the 32-bit formats.

1 files changed, 15 insertions, 0 deletions

diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index 1e65a91fc6..1b07a98c21 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -246,6 +246,21 @@ llvmpipe_is_format_supported( struct pipe_screen *_screen,
    if (sample_count > 1)
       return FALSE;
 
+   if (bind & (PIPE_BIND_RENDER_TARGET | PIPE_BIND_SAMPLER_VIEW)) {
+      /* Reject 16-bit floating point formats because conversion is not native.
+       * The state tracker will fall back to 32-bit ones.
+       * TODO: extend this interface to allow us to return a "ok, but slow" answer
+       * TODO: fully enable this on (unreleased) CPUs supporting CVT16
+       */
+      unsigned i;
+      for(i = 0; i < format_desc->nr_channels; ++i)
+      {
+         if(format_desc->channel[0].type == UTIL_FORMAT_TYPE_FLOAT
+               && format_desc->channel[0].size < 32)
+            return FALSE;
+      }
+   }
+
    if (bind & PIPE_BIND_RENDER_TARGET) {
       if (format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS)
          return FALSE;