[Mesa-dev] [PATCH] radv: rework the TC-compat HTILE hardware bug with COND_EXEC
Bas Nieuwenhuizen
bas at basnieuwenhuizen.nl
Mon Dec 3 22:21:19 UTC 2018
On Mon, Dec 3, 2018 at 11:13 PM Samuel Pitoiset
<samuel.pitoiset at gmail.com> wrote:
>
> After investigating on this, it appears that COND_WRITE doesn't
> work correctly in some situations. I don't know exactly why does
> it fail to update DB_Z_INFO.ZRANGE_PRECISION, but as AMDVLK
> also uses COND_EXEC I think there is a reason.
>
> Now the driver stores a new metadata value in order to reflect
> the last fast depth clear state. If a TC-compat HTILE is fast cleared
> with 0.0f, we have to update ZRANGE_PRECISION to 0 in order to
> work around that hardware bug.
>
> This fixes rendering issues with The Forest and DXVK and doesn't
> seem to introduce any regressions.
>
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108914
> Fixes: 68dead112e7 ("radv: update the ZRANGE_PRECISION value for the TC-compat bug")
> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
> ---
> src/amd/vulkan/radv_cmd_buffer.c | 91 ++++++++++++++++++++++----------
> src/amd/vulkan/radv_image.c | 10 +++-
> src/amd/vulkan/radv_private.h | 2 +
> 3 files changed, 75 insertions(+), 28 deletions(-)
>
> diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
> index ccf762ead46..23909a0f7dd 100644
> --- a/src/amd/vulkan/radv_cmd_buffer.c
> +++ b/src/amd/vulkan/radv_cmd_buffer.c
> @@ -1067,7 +1067,7 @@ static void
> radv_update_zrange_precision(struct radv_cmd_buffer *cmd_buffer,
> struct radv_ds_buffer_info *ds,
> struct radv_image *image, VkImageLayout layout,
> - bool requires_cond_write)
> + bool requires_cond_exec)
> {
> uint32_t db_z_info = ds->db_z_info;
> uint32_t db_z_info_reg;
> @@ -1091,38 +1091,21 @@ radv_update_zrange_precision(struct radv_cmd_buffer *cmd_buffer,
> }
>
> /* When we don't know the last fast clear value we need to emit a
> - * conditional packet, otherwise we can update DB_Z_INFO directly.
> + * conditional packet that will eventually skip the following
> + * SET_CONTEXT_REG packet.
> */
> - if (requires_cond_write) {
> - radeon_emit(cmd_buffer->cs, PKT3(PKT3_COND_WRITE, 7, 0));
> -
> - const uint32_t write_space = 0 << 8; /* register */
> - const uint32_t poll_space = 1 << 4; /* memory */
> - const uint32_t function = 3 << 0; /* equal to the reference */
> - const uint32_t options = write_space | poll_space | function;
> - radeon_emit(cmd_buffer->cs, options);
> -
> - /* poll address - location of the depth clear value */
> + if (requires_cond_exec) {
> uint64_t va = radv_buffer_get_va(image->bo);
> - va += image->offset + image->clear_value_offset;
> -
> - /* In presence of stencil format, we have to adjust the base
> - * address because the first value is the stencil clear value.
> - */
> - if (vk_format_is_stencil(image->vk_format))
> - va += 4;
> + va += image->offset + image->tc_compat_zrange_offset;
>
> + radeon_emit(cmd_buffer->cs, PKT3(PKT3_COND_EXEC, 3, 0));
> radeon_emit(cmd_buffer->cs, va);
> radeon_emit(cmd_buffer->cs, va >> 32);
> -
> - radeon_emit(cmd_buffer->cs, fui(0.0f)); /* reference value */
> - radeon_emit(cmd_buffer->cs, (uint32_t)-1); /* comparison mask */
> - radeon_emit(cmd_buffer->cs, db_z_info_reg >> 2); /* write address low */
> - radeon_emit(cmd_buffer->cs, 0u); /* write address high */
> - radeon_emit(cmd_buffer->cs, db_z_info);
> - } else {
> - radeon_set_context_reg(cmd_buffer->cs, db_z_info_reg, db_z_info);
> + radeon_emit(cmd_buffer->cs, 0);
> + radeon_emit(cmd_buffer->cs, 3); /* SET_CONTEXT_REG size */
> }
> +
> + radeon_set_context_reg(cmd_buffer->cs, db_z_info_reg, db_z_info);
> }
>
> static void
> @@ -1269,6 +1252,45 @@ radv_set_ds_clear_metadata(struct radv_cmd_buffer *cmd_buffer,
> radeon_emit(cs, fui(ds_clear_value.depth));
> }
>
> +/**
> + * Update the TC-compat metadata value for this image.
> + */
> +static void
> +radv_set_tc_compat_zrange_metadata(struct radv_cmd_buffer *cmd_buffer,
> + struct radv_image *image,
> + uint32_t value)
> +{
> + struct radeon_cmdbuf *cs = cmd_buffer->cs;
> + uint64_t va = radv_buffer_get_va(image->bo);
> + va += image->offset + image->tc_compat_zrange_offset;
> +
> + radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 3, 0));
> + radeon_emit(cs, S_370_DST_SEL(V_370_MEM_ASYNC) |
> + S_370_WR_CONFIRM(1) |
> + S_370_ENGINE_SEL(V_370_PFP));
> + radeon_emit(cs, va);
> + radeon_emit(cs, va >> 32);
> + radeon_emit(cs, value);
> +}
> +
> +static void
> +radv_update_tc_compat_zrange_metadata(struct radv_cmd_buffer *cmd_buffer,
> + struct radv_image *image,
> + VkClearDepthStencilValue ds_clear_value)
> +{
> + struct radeon_cmdbuf *cs = cmd_buffer->cs;
> + uint64_t va = radv_buffer_get_va(image->bo);
> + va += image->offset + image->tc_compat_zrange_offset;
> + uint32_t cond_val;
> +
> + /* Conditionally set DB_Z_INFO.ZRANGE_PRECISION to 0 when the last
> + * depth clear value is 0.0f.
> + */
> + cond_val = ds_clear_value.depth == 0.0f ? UINT_MAX : 0;
> +
> + radv_set_tc_compat_zrange_metadata(cmd_buffer, image, cond_val);
> +}
> +
> /**
> * Update the clear depth/stencil values for this image.
> */
> @@ -1282,6 +1304,12 @@ radv_update_ds_clear_metadata(struct radv_cmd_buffer *cmd_buffer,
>
> radv_set_ds_clear_metadata(cmd_buffer, image, ds_clear_value, aspects);
>
> + if (radv_image_is_tc_compat_htile(image) &&
> + (aspects & VK_IMAGE_ASPECT_DEPTH_BIT)) {
> + radv_update_tc_compat_zrange_metadata(cmd_buffer, image,
> + ds_clear_value);
> + }
> +
> radv_update_bound_fast_clear_ds(cmd_buffer, image, ds_clear_value,
> aspects);
> }
> @@ -4229,6 +4257,15 @@ static void radv_initialize_htile(struct radv_cmd_buffer *cmd_buffer,
> aspects |= VK_IMAGE_ASPECT_STENCIL_BIT;
>
> radv_set_ds_clear_metadata(cmd_buffer, image, value, aspects);
> +
> + if (radv_image_is_tc_compat_htile(image)) {
> + /* Initialize the TC-compat metada value to 0 because by
> + * default DB_Z_INFO.RANGE_PRECISION is set to 1, and we only
> + * need have to conditionally update its value when performing
> + * a fast depth clear.
> + */
> + radv_set_tc_compat_zrange_metadata(cmd_buffer, image, 0);
> + }
> }
>
> static void radv_handle_depth_image_transition(struct radv_cmd_buffer *cmd_buffer,
> diff --git a/src/amd/vulkan/radv_image.c b/src/amd/vulkan/radv_image.c
> index f447166d80c..9fcbaa3184c 100644
> --- a/src/amd/vulkan/radv_image.c
> +++ b/src/amd/vulkan/radv_image.c
> @@ -870,6 +870,14 @@ radv_image_alloc_htile(struct radv_image *image)
> /* + 8 for storing the clear values */
> image->clear_value_offset = image->htile_offset + image->surface.htile_size;
> image->size = image->clear_value_offset + 8;
> + if (radv_image_is_tc_compat_htile(image)) {
> + /* Metadata for the TC-compatible HTILE hardware bug which
> + * have to be fixed by updating ZRANGE_PRECISION when doing
> + * fast depth clears to 0.0f.
> + */
> + image->tc_compat_zrange_offset = image->clear_value_offset + 8;
> + image->size = image->clear_value_offset + 16;
> + }
> image->alignment = align64(image->alignment, image->surface.htile_alignment);
> }
>
> @@ -1014,8 +1022,8 @@ radv_image_create(VkDevice _device,
> /* Otherwise, try to enable HTILE for depth surfaces. */
> if (radv_image_can_enable_htile(image) &&
> !(device->instance->debug_flags & RADV_DEBUG_NO_HIZ)) {
> - radv_image_alloc_htile(image);
> image->tc_compatible_htile = image->surface.flags & RADEON_SURF_TC_COMPATIBLE_HTILE;
> + radv_image_alloc_htile(image);
> } else {
> image->surface.htile_size = 0;
> }
> diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
> index ac756f2c247..1ef5b215db7 100644
> --- a/src/amd/vulkan/radv_private.h
> +++ b/src/amd/vulkan/radv_private.h
> @@ -1503,6 +1503,8 @@ struct radv_image {
> uint64_t clear_value_offset;
> uint64_t fce_pred_offset;
>
> + uint64_t tc_compat_zrange_offset;
Can we document what the values put into this memory location mean?
Otherwise
Reviewed-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
> +
> /* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
> VkDeviceMemory owned_memory;
> };
> --
> 2.19.2
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
More information about the mesa-dev
mailing list