Mesa (master): radv: enable TC-compat HTILE for D32_SFLOAT+MSAA on GFX10+

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: 81a6ee7a9b133039f7fd39e18b9b54707b62ff19
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=81a6ee7a9b133039f7fd39e18b9b54707b62ff19

Author: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Date:   Thu Dec 17 11:06:10 2020 +0100

radv: enable TC-compat HTILE for D32_SFLOAT+MSAA on GFX10+

This was disabled due to some depth/stencil resolve CTS failures
which are now fixed.

I figured that disabling TC-compat HTILE for D32_SFLOAT+MSAA reduced
performance in Control by -11% on Vega10. In fact, the game only uses
D32_SFLOAT for depth rendering.

This gives a huge boost in Control on Navi10 (eg. +17% in MSAA4x).
Note that the game is still slower than PRO without MSAA on Navi10,
but as fast (or even a bit faster) on Vega10.

I think TC-compat HILE could also be enabled for D32_SFLOAT_S8_UINT
but it needs more testing first.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8143>

---

 src/amd/vulkan/radv_image.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/src/amd/vulkan/radv_image.c b/src/amd/vulkan/radv_image.c
index cb67fb5d0f7..e0e58917065 100644
--- a/src/amd/vulkan/radv_image.c
+++ b/src/amd/vulkan/radv_image.c
@@ -94,12 +94,9 @@ radv_use_tc_compat_htile_for_image(struct radv_device *device,
 		return false;
 
 	/* FIXME: for some reason TC compat with 2/4/8 samples breaks some cts
-	 * tests - disable for now. On GFX10 D32_SFLOAT is affected as well.
+	 * tests - disable for now.
 	 */
-	if (pCreateInfo->samples >= 2 &&
-	    (format == VK_FORMAT_D32_SFLOAT_S8_UINT ||
-	     (format == VK_FORMAT_D32_SFLOAT &&
-	      device->physical_device->rad_info.chip_class >= GFX10)))
+	if (pCreateInfo->samples >= 2 && format == VK_FORMAT_D32_SFLOAT_S8_UINT)
 		return false;
 
 	/* GFX9 supports both 32-bit and 16-bit depth surfaces, while GFX8 only