From f2c4c5493e4a8389f14567f9c4a75893d78ca511 Mon Sep 17 00:00:00 2001 From: Eric Engestrom Date: Thu, 18 Sep 2025 10:23:01 +0200 Subject: [PATCH] radv: add comment explaining why fp16 is disabled by default on gfx8 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Suggested-by: Timur Kristóf Part-of: --- src/amd/vulkan/radv_physical_device.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/src/amd/vulkan/radv_physical_device.c b/src/amd/vulkan/radv_physical_device.c index 51a84bb2791..2e481443027 100644 --- a/src/amd/vulkan/radv_physical_device.c +++ b/src/amd/vulkan/radv_physical_device.c @@ -877,6 +877,9 @@ radv_physical_device_get_features(const struct radv_physical_device *pdev, struc .storagePushConstant8 = true, .shaderBufferInt64Atomics = true, .shaderSharedInt64Atomics = true, + /* GFX8 supports fp16, but not double rate packed math. + * We don't enable that by default because it can sometimes hurt perf. + */ .shaderFloat16 = pdev->info.has_packed_math_16bit || (pdev->info.gfx_level == GFX8 && instance->drirc.features.expose_float16_gfx8), .shaderInt8 = true, @@ -1470,6 +1473,9 @@ radv_get_physical_device_properties(struct radv_physical_device *pdev) device_type = VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU; } + /* GFX8 supports fp16, but not double rate packed math. + * We don't enable that by default because it can sometimes hurt perf. + */ bool has_fp16 = pdev->info.has_packed_math_16bit || (pdev->info.gfx_level == GFX8 && instance->drirc.features.expose_float16_gfx8);