[Mesa-dev] [PATCH v3 20/24] anv/cmd_buffer: Rework aux tracking
Nanley Chery
nanleychery at gmail.com
Wed Feb 7 17:59:58 UTC 2018
On Tue, Feb 06, 2018 at 05:57:26PM -0800, Jason Ekstrand wrote:
> On Tue, Feb 6, 2018 at 5:09 PM, Nanley Chery <nanleychery at gmail.com> wrote:
>
> > On Mon, Feb 05, 2018 at 06:16:26PM -0800, Jason Ekstrand wrote:
> > > This commit completely reworks aux tracking. This includes a number of
> > > somewhat distinct changes:
> > >
> > > 1) Since we are no longer fast-clearing multiple slices, we only need
> > > to track one fast clear color and one fast clear type.
> > >
> > > 2) We store two bits for fast clear instead of one to let us
> > > distinguish between zero and non-zero fast clear colors. This is
> > > needed so that we can do full resolves when transitioning to
> > > PRESENT_SRC_KHR with gen9 CCS images where we allow zero clear
> > > values in all sorts of places we wouldn't normally.
> > >
> > > 3) We now track compression state as a boolean separate from fast clear
> > > type and this is tracked on a per-slice granularity.
> > >
> > > The previous scheme had some issues when it came to individual slices of
> > > a multi-LOD images. In particular, we only tracked "needs resolve"
> > > per-LOD but you could do a vkCmdPipelineBarrier that would only resolve
> > > a portion of the image and would set "needs resolve" to false anyway.
> > > Also, any transition from an undefined layout would reset the clear
> > > color for the entire LOD regardless of whether or not there was some
> > > clear color on some other slice.
> > >
> > > As far as full/partial resolves go, he assumptions of the previous
> > > scheme held because the one case where we do need a full resolve when
> > > CCS_E is enabled is for window-system images. Since we only ever
> > > allowed X-tiled window-system images, CCS was entirely disabled on gen9+
> > > and we never got CCS_E. With the advent of Y-tiled window-system
> > > buffers, we now need to properly support doing a full resolve of images
> > > marked CCS_E.
> > >
> > > v2 (Jason Ekstrand):
> > > - Fix an bug in the compressed flag offset calculation
> > > - Treat 3D images as multi-slice for the purposes of resolve tracking
> > >
> > > Reviewed-by: Topi Pohjolainen <topi.pohjolainen at intel.com>
> > > ---
> > > src/intel/vulkan/anv_blorp.c | 3 +-
> > > src/intel/vulkan/anv_image.c | 100 ++++++-----
> > > src/intel/vulkan/anv_private.h | 60 ++++---
> > > src/intel/vulkan/genX_cmd_buffer.c | 340 +++++++++++++++++++++++++++---
> > -------
> > > 4 files changed, 345 insertions(+), 158 deletions(-)
> > >
> > > diff --git a/src/intel/vulkan/anv_blorp.c b/src/intel/vulkan/anv_blorp.c
> > > index 497ae6f..fc3b717 100644
> > > --- a/src/intel/vulkan/anv_blorp.c
> > > +++ b/src/intel/vulkan/anv_blorp.c
> > > @@ -1758,8 +1758,7 @@ anv_image_ccs_op(struct anv_cmd_buffer *cmd_buffer,
> > > * particular value and don't care about format or clear value.
> > > */
> > > const struct anv_address clear_color_addr =
> > > - anv_image_get_clear_color_addr(cmd_buffer->device, image,
> > > - aspect, level);
> > > + anv_image_get_clear_color_addr(cmd_buffer->device, image,
> > aspect);
> > > surf.clear_color_addr = anv_to_blorp_address(clear_color_addr);
> > > }
> > >
> > > diff --git a/src/intel/vulkan/anv_image.c b/src/intel/vulkan/anv_image.c
> > > index 11942d0..011e952 100644
> > > --- a/src/intel/vulkan/anv_image.c
> > > +++ b/src/intel/vulkan/anv_image.c
> > > @@ -190,46 +190,54 @@ all_formats_ccs_e_compatible(const struct
> > gen_device_info *devinfo,
> > > * fast-clear values in non-trivial cases (e.g., outside of a render
> > pass in
> > > * which a fast clear has occurred).
> > > *
> > > - * For the purpose of discoverability, the algorithm used to manage
> > this buffer
> > > - * is described here. A clear value in this buffer is updated when a
> > fast clear
> > > - * is performed on a subresource. One of two synchronization operations
> > is
> > > - * performed in order for a following memory access to use the
> > fast-clear
> > > - * value:
> > > - * a. Copy the value from the buffer to the surface state object
> > used for
> > > - * reading. This is done implicitly when the value is the clear
> > value
> > > - * predetermined to be the default in other surface state
> > objects. This
> > > - * is currently only done explicitly for the operation below.
> > > - * b. Do (a) and use the surface state object to resolve the
> > subresource.
> > > - * This is only done during layout transitions for decent
> > performance.
> > > + * In order to avoid having multiple clear colors for a single plane of
> > an
> > > + * image (hence a single RENDER_SURFACE_STATE), we only allow
> > fast-clears on
> > > + * the first slice (level 0, layer 0). At the time of our testing (Jan
> > 17,
> > > + * 2018), there were no known applications which would benefit from
> > fast-
> > > + * clearing more than just the first slice.
> > > *
> > > - * With the above scheme, we can fast-clear whenever the hardware
> > allows except
> > > - * for two cases in which synchronization becomes impossible or
> > undesirable:
> > > - * * The subresource is in the GENERAL layout and is cleared to a
> > value
> > > - * other than the special default value.
> > > + * The fast clear portion of the image is laid out in the following
> > order:
> > > *
> > > - * Performing a synchronization operation in order to read from the
> > > - * subresource is undesirable in this case. Firstly, b) is not an
> > option
> > > - * because a layout transition isn't required between a write and
> > read of
> > > - * an image in the GENERAL layout. Secondly, it's undesirable to
> > do a)
> > > - * explicitly because it would require large infrastructural
> > changes. The
> > > - * Vulkan API supports us in deciding not to optimize this layout
> > by
> > > - * stating that using this layout may cause suboptimal
> > performance. NOTE:
> > > - * the auxiliary buffer must always be enabled to support a)
> > implicitly.
> > > + * * 1 or 4 dwords (depending on hardware generation) for the clear
> > color
> > > + * * 1 dword for the anv_fast_clear_type of the clear color
> > > + * * On gen9+, 1 dword per level and layer of the image (3D levels
> > count
> > > + * multiple layers) in level-major order for compression state.
> > > *
> > > + * For the purpose of discoverability, the algorithm used to manage
> > > + * compression and fast-clears is described here:
> > > *
> > > - * * For the given miplevel, only some of the layers are cleared at
> > once.
> > > + * * On a transition from UNDEFINED or PREINITIALIZED to a defined
> > layout,
> > > + * all of the values in the fast clear portion of the image are
> > initialized
> > > + * to default values.
> > > *
> > > - * If the user clears each layer to a different value, then tries
> > to
> > > - * render to multiple layers at once, we have no ability to
> > perform a
> > > - * synchronization operation in between. a) is not helpful because
> > the
> > > - * object can only hold one clear value. b) is not an option
> > because a
> > > - * layout transition isn't required in this case.
> > > + * * On fast-clear, the clear value is written into surface state and
> > also
> > > + * into the buffer and the fast clear type is set appropriately.
> > Both
> > > + * setting the fast-clear value in the buffer and setting the
> > fast-clear
> > > + * type happen from the GPU using MI commands.
> >
> > There's no mention about setting the compression dwords during
> > fast-clear operations or on render target writes.
> >
>
> I've added another bullet:
>
> * * Whenever a render or blorp operation is performed with CCS_E, we call
> * anv_cmd_buffer_mark_image_written to set the compression state to 1.
>
>
That'll work. I think we should replace
anv_cmd_buffer_mark_image_written with
genX(cmd_buffer_mark_image_written) though, since the former's a wrapper
that only gets called for blorp operations.
> > > + *
> > > + * * On pipeline barrier transitions, the worst-case transition is
> > computed
> > > + * from the image layouts. The command streamer inspects the fast
> > clear
> > > + * type and compression state dwords and constructs a predicate. The
> > > + * worst-case resolve is performed with the given predicate and the
> > fast
> > > + * clear and compression state is set accordingly.
> > > + *
> > > + * See anv_layout_to_aux_usage and anv_layout_to_fast_clear_type
> > functions for
> > > + * details on exactly what is allowed in what layouts.
> > > + *
> > > + * On gen7-9, we do not have a concept of indirect clear colors in
> > hardware.
> > > + * In order to deal with this, we have to do some clear color
> > management.
> > > + *
> > > + * * For LOAD_OP_LOAD at the top of a renderpass, we have to copy the
> > clear
> > > + * value from the buffer into the surface state with MI commands.
> > > + *
> > > + * * For any blorp operations, we pass the address to the clear value
> > into
> > > + * blorp and it knows to copy the clear color.
> > > */
> > > static void
> > > -add_fast_clear_state_buffer(struct anv_image *image,
> > > - VkImageAspectFlagBits aspect,
> > > - uint32_t plane,
> > > - const struct anv_device *device)
> > > +add_aux_state_tracking_buffer(struct anv_image *image,
> > > + VkImageAspectFlagBits aspect,
> > > + uint32_t plane,
> > > + const struct anv_device *device)
> >
> > The comment referencing this function in genX_cmd_buffer.c needs to be
> > updated.
> >
>
> I don't see that comment anywhere.
>
>
Maybe you delete it in a future series? This is the one I have in mind:
/* We only allow fast clears in the GENERAL layout if the auxiliary
* buffer is always enabled and the fast-clear value is all 0's. See
* add_fast_clear_state_buffer() for more information.
*/
> > > {
> > > assert(image && device);
> > > assert(image->planes[plane].aux_surface.isl.size > 0 &&
> > > @@ -251,20 +259,24 @@ add_fast_clear_state_buffer(struct anv_image
> > *image,
> > > (image->planes[plane].offset + image->planes[plane].size));
> > > }
> > >
> > > - const unsigned entry_size = anv_fast_clear_state_entry_size(device);
> > > - /* There's no padding between entries, so ensure that they're always
> > a
> > > - * multiple of 32 bits in order to enable GPU memcpy operations.
> > > - */
> > > - assert(entry_size % 4 == 0);
> > > + /* Clear color and fast clear type */
> > > + unsigned state_size = device->isl_dev.ss.clear_value_size + 4;
> > >
> > > - const unsigned plane_state_size =
> > > - entry_size * anv_image_aux_levels(image, aspect);
> > > + /* We only need to track compression on CCS_E surfaces. */
> > > + if (image->planes[plane].aux_usage == ISL_AUX_USAGE_CCS_E) {
> > > + if (image->type == VK_IMAGE_TYPE_3D) {
> > > + for (uint32_t l = 0; l < image->levels; l++)
> > > + state_size += anv_minify(image->extent.depth, l) * 4;
> > > + } else {
> > > + state_size += image->levels * image->array_size * 4;
> > > + }
> > > + }
> > >
> > > image->planes[plane].fast_clear_state_offset =
> > > image->planes[plane].offset + image->planes[plane].size;
> > >
> > > - image->planes[plane].size += plane_state_size;
> > > - image->size += plane_state_size;
> > > + image->planes[plane].size += state_size;
> > > + image->size += state_size;
> > > }
> > >
> > > /**
> > > @@ -437,7 +449,7 @@ make_surface(const struct anv_device *dev,
> > > }
> > >
> > > add_surface(image, &image->planes[plane].aux_surface,
> > plane);
> > > - add_fast_clear_state_buffer(image, aspect, plane, dev);
> > > + add_aux_state_tracking_buffer(image, aspect, plane, dev);
> > >
> > > /* For images created without MUTABLE_FORMAT_BIT set, we
> > know that
> > > * they will always be used with the original format. In
> > > @@ -461,7 +473,7 @@ make_surface(const struct anv_device *dev,
> > > &image->planes[plane].aux_
> > surface.isl);
> > > if (ok) {
> > > add_surface(image, &image->planes[plane].aux_surface, plane);
> > > - add_fast_clear_state_buffer(image, aspect, plane, dev);
> > > + add_aux_state_tracking_buffer(image, aspect, plane, dev);
> > > image->planes[plane].aux_usage = ISL_AUX_USAGE_MCS;
> > > }
> > > }
> > > diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_
> > private.h
> > > index 5f82702..d38dd9e 100644
> > > --- a/src/intel/vulkan/anv_private.h
> > > +++ b/src/intel/vulkan/anv_private.h
> > > @@ -2533,50 +2533,58 @@ anv_image_aux_layers(const struct anv_image *
> > const image,
> > > }
> > > }
> > >
> > > -static inline unsigned
> > > -anv_fast_clear_state_entry_size(const struct anv_device *device)
> > > -{
> > > - assert(device);
> > > - /* Entry contents:
> > > - * +--------------------------------------------+
> > > - * | clear value dword(s) | needs resolve dword |
> > > - * +--------------------------------------------+
> > > - */
> > > -
> > > - /* Ensure that the needs resolve dword is in fact dword-aligned to
> > enable
> > > - * GPU memcpy operations.
> > > - */
> > > - assert(device->isl_dev.ss.clear_value_size % 4 == 0);
> > > - return device->isl_dev.ss.clear_value_size + 4;
> > > -}
> > > -
> > > static inline struct anv_address
> > > anv_image_get_clear_color_addr(const struct anv_device *device,
> > > const struct anv_image *image,
> > > - VkImageAspectFlagBits aspect,
> > > - unsigned level)
> > > + VkImageAspectFlagBits aspect)
> > > {
> > > + assert(image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
> > > +
> > > uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
> > > return (struct anv_address) {
> > > .bo = image->planes[plane].bo,
> > > .offset = image->planes[plane].bo_offset +
> > > - image->planes[plane].fast_clear_state_offset +
> > > - anv_fast_clear_state_entry_size(device) * level,
> > > + image->planes[plane].fast_clear_state_offset,
> > > };
> > > }
> > >
> > > static inline struct anv_address
> > > -anv_image_get_needs_resolve_addr(const struct anv_device *device,
> > > - const struct anv_image *image,
> > > - VkImageAspectFlagBits aspect,
> > > - unsigned level)
> > > +anv_image_get_fast_clear_type_addr(const struct anv_device *device,
> > > + const struct anv_image *image,
> > > + VkImageAspectFlagBits aspect)
> > > {
> > > struct anv_address addr =
> > > - anv_image_get_clear_color_addr(device, image, aspect, level);
> > > + anv_image_get_clear_color_addr(device, image, aspect);
> > > addr.offset += device->isl_dev.ss.clear_value_size;
> > > return addr;
> > > }
> > >
> > > +static inline struct anv_address
> > > +anv_image_get_compression_state_addr(const struct anv_device *device,
> > > + const struct anv_image *image,
> > > + VkImageAspectFlagBits aspect,
> > > + uint32_t level, uint32_t
> > array_layer)
> > > +{
> > > + assert(level < anv_image_aux_levels(image, aspect));
> > > + assert(array_layer < anv_image_aux_layers(image, aspect, level));
> > > + UNUSED uint32_t plane = anv_image_aspect_to_plane(image->aspects,
> > aspect);
> > > + assert(image->planes[plane].aux_usage == ISL_AUX_USAGE_CCS_E);
> > > +
> > > + struct anv_address addr =
> > > + anv_image_get_fast_clear_type_addr(device, image, aspect);
> > > + addr.offset += 4; /* Go past the fast clear type */
> > > +
> > > + if (image->type == VK_IMAGE_TYPE_3D) {
> > > + for (uint32_t l = 0; l < level; l++)
> > > + addr.offset += anv_minify(image->extent.depth, l) * 4;
> > > + } else {
> > > + addr.offset += level * image->array_size * 4;
> > > + }
> > > + addr.offset += array_layer * 4;
> > > +
> > > + return addr;
> > > +}
> > > +
> > > /* Returns true if a HiZ-enabled depth buffer can be sampled from. */
> > > static inline bool
> > > anv_can_sample_with_hiz(const struct gen_device_info * const devinfo,
> > > diff --git a/src/intel/vulkan/genX_cmd_buffer.c
> > b/src/intel/vulkan/genX_cmd_buffer.c
> > > index e1a4d95..4c75e0c 100644
> > > --- a/src/intel/vulkan/genX_cmd_buffer.c
> > > +++ b/src/intel/vulkan/genX_cmd_buffer.c
> > > @@ -407,27 +407,45 @@ transition_depth_buffer(struct anv_cmd_buffer
> > *cmd_buffer,
> > > #define MI_PREDICATE_SRC0 0x2400
> > > #define MI_PREDICATE_SRC1 0x2408
> > >
> > > -/* Manages the state of an color image subresource to ensure resolves
> > are
> > > - * performed properly.
> > > - */
> > > static void
> > > -genX(set_image_needs_resolve)(struct anv_cmd_buffer *cmd_buffer,
> > > - const struct anv_image *image,
> > > - VkImageAspectFlagBits aspect,
> > > - unsigned level, bool needs_resolve)
> > > +set_image_fast_clear_state(struct anv_cmd_buffer *cmd_buffer,
> > > + const struct anv_image *image,
> > > + VkImageAspectFlagBits aspect,
> > > + enum anv_fast_clear_type fast_clear)
> > > {
> > > - assert(cmd_buffer && image);
> > > - assert(image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
> > > - assert(level < anv_image_aux_levels(image, aspect));
> > > -
> > > - /* The HW docs say that there is no way to guarantee the completion
> > of
> > > - * the following command. We use it nevertheless because it shows no
> > > - * issues in testing is currently being used in the GL driver.
> > > - */
> > > anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM), sdi) {
> > > - sdi.Address = anv_image_get_needs_resolve_
> > addr(cmd_buffer->device,
> > > - image, aspect,
> > level);
> > > - sdi.ImmediateData = needs_resolve;
> > > + sdi.Address = anv_image_get_fast_clear_type_
> > addr(cmd_buffer->device,
> > > + image, aspect);
> > > + sdi.ImmediateData = fast_clear;
> > > + }
> > > +}
> > > +
> > > +static void
> > > +set_image_compressed_bit(struct anv_cmd_buffer *cmd_buffer,
> > > + const struct anv_image *image,
> > > + VkImageAspectFlagBits aspect,
> > > + uint32_t level,
> > > + uint32_t base_layer, uint32_t layer_count,
> > > + bool compressed)
> > > +{
> > > + /* We only have CCS_E on gen9+ */
> > > + if (GEN_GEN < 9)
> > > + return;
> > > +
> >
> > I'm also fine with dropping this if statement.
> >
>
> Ok, we'll drop it.
>
>
> > > + uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
> > > +
> > > + /* We only have compression tracking for CCS_E */
> > > + if (image->planes[plane].aux_usage != ISL_AUX_USAGE_CCS_E)
> > > + return;
> > > +
> > > + for (uint32_t a = 0; a < layer_count; a++) {
> > > + uint32_t layer = base_layer + a;
> > > + anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM), sdi)
> > {
> > > + sdi.Address = anv_image_get_compression_
> > state_addr(cmd_buffer->device,
> > > + image,
> > aspect,
> > > + level,
> > layer);
> > > + sdi.ImmediateData = compressed ? UINT32_MAX : 0;
> > > + }
> > > }
> > > }
> > >
> > > @@ -451,32 +469,176 @@ mi_alu(uint32_t opcode, uint32_t operand1,
> > uint32_t operand2)
> > > #define CS_GPR(n) (0x2600 + (n) * 8)
> > >
> > > static void
> > > -genX(load_needs_resolve_predicate)(struct anv_cmd_buffer *cmd_buffer,
> > > - const struct anv_image *image,
> > > - VkImageAspectFlagBits aspect,
> > > - unsigned level)
> > > +anv_cmd_predicated_ccs_resolve(struct anv_cmd_buffer *cmd_buffer,
> > > + const struct anv_image *image,
> > > + VkImageAspectFlagBits aspect,
> > > + uint32_t level, uint32_t array_layer,
> > > + enum isl_aux_op resolve_op,
> > > + enum anv_fast_clear_type
> > fast_clear_supported)
> > > {
> > > - assert(cmd_buffer && image);
> > > - assert(image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
> > > - assert(level < anv_image_aux_levels(image, aspect));
> > > + struct anv_address fast_clear_type_addr =
> > > + anv_image_get_fast_clear_type_addr(cmd_buffer->device, image,
> > aspect);
> > > +
> > > +#if GEN_GEN >= 9
> > > + const uint32_t plane = anv_image_aspect_to_plane(image->aspects,
> > aspect);
> > > + const bool decompress =
> > > + resolve_op == ISL_AUX_OP_FULL_RESOLVE &&
> > > + image->planes[plane].aux_usage == ISL_AUX_USAGE_CCS_E;
> > > +
> > > + /* This function shouldn't get called if it isn't going to do
> > anything */
> > > + assert(decompress || fast_clear_supported < ANV_FAST_CLEAR_ANY);
> > > +
> > > + if (level == 0 && array_layer == 0) {
> > > + /* This is the complex case because we have to worry about
> > dealing with
> > > + * the fast clear color. Unfortunately, it's also the common
> > case.
> > > + */
> > > +
> > > + /* Poor-man's register allocation */
> > > + int next_reg = MI_ALU_REG0;
> > > + int pred_reg = -1;
> > > +
> > > + /* Needed for ALU operations */
> > > + uint32_t *dw;
> > > +
> > > + const int image_fc = next_reg++;
> > > + anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_MEM),
> > lrm) {
> > > + lrm.RegisterAddress = CS_GPR(image_fc);
> > > + lrm.MemoryAddress = fast_clear_type_addr;
> > > + }
> > > + emit_lri(&cmd_buffer->batch, CS_GPR(image_fc) + 4, 0);
> > > +
> > > + if (fast_clear_supported < ANV_FAST_CLEAR_ANY) {
> > > + /* We need to compute (fast_clear_supported <
> > image->fast_clear).
> > > + * We do this by subtracting and storing the carry bit.
> > > + */
> > > + const int fc_imm = next_reg++;
> > > + emit_lri(&cmd_buffer->batch, CS_GPR(fc_imm),
> > fast_clear_supported);
> > > + emit_lri(&cmd_buffer->batch, CS_GPR(fc_imm) + 4, 0);
> > > +
> > > + assert(pred_reg == -1);
> > > + pred_reg = next_reg++;
> > > +
> > > + dw = anv_batch_emitn(&cmd_buffer->batch, 5, GENX(MI_MATH));
> > > + dw[1] = mi_alu(MI_ALU_LOAD, MI_ALU_SRCA, fc_imm);
> > > + dw[2] = mi_alu(MI_ALU_LOAD, MI_ALU_SRCB, image_fc);
> > > + dw[3] = mi_alu(MI_ALU_SUB, 0, 0);
> > > + dw[4] = mi_alu(MI_ALU_STORE, pred_reg, MI_ALU_CF);
> > > + }
> > > +
> > > + if (decompress) {
> > > + /* If we're doing a full resolve, we need the compression
> > state */
> > > + struct anv_address compression_state_addr =
> > > + anv_image_get_compression_state_addr(cmd_buffer->device,
> > image,
> > > + aspect, level,
> > array_layer);
> > > + if (pred_reg == -1) {
> > > + pred_reg = next_reg++;
> > > + anv_batch_emit(&cmd_buffer->batch,
> > GENX(MI_LOAD_REGISTER_MEM), lrm) {
> > > + lrm.RegisterAddress = CS_GPR(pred_reg);
> > > + lrm.MemoryAddress = compression_state_addr;
> > > + }
> >
>
> In this case, we don't have a fast-clear predicate. This is because "if
> (fast_clear_supported < ANV_FAST_CLEAR_ANY)" failed so we didn't compute
> one. In this case, we treat the fast-clear predicate as being 0 and OR
> with 0 is just the original value.
>
>
>
> > > + } else {
> > > + /* OR the compression state into the predicate. The
> > compression
> > > + * state is already in 0/~0 form.
> > > + */
> >
> > The OR is a result of this function's structure and wouldn't be needed
> > if the control flow in this function was different right?
> >
>
> In this case, we have both a fast-clear predicate and a compression
> predicate. We have to OR them together.
>
> I'm not sure what you mean about control-flow in this function being
> different. Yes, I'm probably over-optimizing a bit to avoid unneeded CS
> operations.
>
>
By that, I'm referring to the organization of the if's and else's in
this function - though that still isn't very exact/clear. Sorry for the
confusion.
It seems like the OR is one more CS operation than neeeded. It's always
the case that if we have some fast-clear data, the compression state
will always be true.
Here's how I understand the states could look:
FC | COMP
==========
0 | 0
0 | 1
1 | 1
2 | 1
If that's the case we only have to check the compression_state when a
decompress is called for. If we moved the "if (decompress) {}" block of
the function higher we wouldn't need to OR the compression predicate
with the fast clear predicate.
> > > + const int image_comp = next_reg++;
> > > + anv_batch_emit(&cmd_buffer->batch,
> > GENX(MI_LOAD_REGISTER_MEM), lrm) {
> > > + lrm.RegisterAddress = CS_GPR(image_comp);
> > > + lrm.MemoryAddress = compression_state_addr;
> > > + }
> > > +
> > > + dw = anv_batch_emitn(&cmd_buffer->batch, 5, GENX(MI_MATH));
> > > + dw[1] = mi_alu(MI_ALU_LOAD, MI_ALU_SRCA, pred_reg);
> > > + dw[2] = mi_alu(MI_ALU_LOAD, MI_ALU_SRCB, image_comp);
> > > + dw[3] = mi_alu(MI_ALU_OR, 0, 0);
> > > + dw[4] = mi_alu(MI_ALU_STORE, pred_reg, MI_ALU_ACCU);
> > > + }
> > > +
> > > + anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM),
> > sdi) {
> > > + sdi.Address = compression_state_addr;
> > > + sdi.ImmediateData = 0;
> > > + }
> > > + }
> > >
> > > - const struct anv_address resolve_flag_addr =
> > > - anv_image_get_needs_resolve_addr(cmd_buffer->device,
> > > - image, aspect, level);
> > > + /* Store the predicate */
> > > + assert(pred_reg != -1);
> > > + emit_lrr(&cmd_buffer->batch, MI_PREDICATE_SRC0, CS_GPR(pred_reg));
> > >
> > > - /* Make the pending predicated resolve a no-op if one is not needed.
> > > - * predicate = do_resolve = resolve_flag != 0;
> > > + /* If the predicate is true, we want to write 0 to the fast clear
> > type
> > > + * and, if it's false, leave it alone. We can do this by writing
> > > + *
> > > + * clear_type = clear_type & ~predicate;
> > > + */
> > > + dw = anv_batch_emitn(&cmd_buffer->batch, 5, GENX(MI_MATH));
> > > + dw[1] = mi_alu(MI_ALU_LOAD, MI_ALU_SRCA, image_fc);
> > > + dw[2] = mi_alu(MI_ALU_LOADINV, MI_ALU_SRCB, pred_reg);
> > > + dw[3] = mi_alu(MI_ALU_AND, 0, 0);
> > > + dw[4] = mi_alu(MI_ALU_STORE, image_fc, MI_ALU_ACCU);
> > > +
> > > + anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_REGISTER_MEM),
> > srm) {
> > > + srm.RegisterAddress = CS_GPR(image_fc);
> > > + srm.MemoryAddress = fast_clear_type_addr;
> > > + }
> > > + } else if (decompress) {
> > > + /* We're trying to get rid of compression but we don't care about
> > fast
> > > + * clears so all we need is the compression predicate.
> > > + */
> > > + assert(resolve_op == ISL_AUX_OP_FULL_RESOLVE);
> > > + struct anv_address compression_state_addr =
> > > + anv_image_get_compression_state_addr(cmd_buffer->device,
> > image,
> > > + aspect, level,
> > array_layer);
> > > + anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_MEM),
> > lrm) {
> > > + lrm.RegisterAddress = MI_PREDICATE_SRC0;
> > > + lrm.MemoryAddress = compression_state_addr;
> > > + }
> > > + anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM), sdi)
> > {
> > > + sdi.Address = compression_state_addr;
> > > + sdi.ImmediateData = 0;
> > > + }
> > > + } else {
> > > + /* In this case, we're trying to do a partial resolve on a slice
> > that
> > > + * doesn't have clear color. There's nothing to do.
> > > + */
> > > + return;
> > > + }
> > > +
> > > +#else /* GEN_GEN <= 8 */
> > > + assert(resolve_op == ISL_AUX_OP_FULL_RESOLVE);
> > > + assert(fast_clear_supported != ANV_FAST_CLEAR_ANY);
> > > +
> > > + /* We don't support fast clears on anything other than the first
> > slice. */
> > > + if (level > 0 || array_layer > 0)
> > > + return;
> > > +
> > > + /* On gen8, we don't have a concept of default clear colors because
> > we
> > > + * can't sample from CCS surfaces. It's enough to just load the
> > fast clear
> > > + * state into the predicate register.
> > > */
> > > + anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_MEM), lrm)
> > {
> > > + lrm.RegisterAddress = MI_PREDICATE_SRC0;
> > > + lrm.MemoryAddress = fast_clear_type_addr;
> > > + }
> > > + anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM), sdi) {
> > > + sdi.Address = fast_clear_type_addr;
> > > + sdi.ImmediateData = 0;
> > > + }
> > > +#endif
> > > +
> > > + /* We use the first half of src0 for the actual predicate. Set the
> > second
> > > + * half of src0 and all of src1 to 0 as the predicate operation will
> > be
> > > + * doing an implicit src0 != src1.
> > > + */
> > > + emit_lri(&cmd_buffer->batch, MI_PREDICATE_SRC0 + 4, 0);
> > > emit_lri(&cmd_buffer->batch, MI_PREDICATE_SRC1 , 0);
> > > emit_lri(&cmd_buffer->batch, MI_PREDICATE_SRC1 + 4, 0);
> > > - emit_lri(&cmd_buffer->batch, MI_PREDICATE_SRC0 , 0);
> > > - emit_lrm(&cmd_buffer->batch, MI_PREDICATE_SRC0 + 4,
> > > - resolve_flag_addr.bo, resolve_flag_addr.offset);
> > > +
> > > anv_batch_emit(&cmd_buffer->batch, GENX(MI_PREDICATE), mip) {
> > > mip.LoadOperation = LOAD_LOADINV;
> > > mip.CombineOperation = COMBINE_SET;
> > > mip.CompareOperation = COMPARE_SRCS_EQUAL;
> > > }
> > > +
> > > + anv_image_ccs_op(cmd_buffer, image, aspect, level,
> > > + array_layer, 1, resolve_op, true);
> > > }
> > >
> > > void
> > > @@ -490,17 +652,30 @@ genX(cmd_buffer_mark_image_written)(struct
> > anv_cmd_buffer *cmd_buffer,
> > > {
> > > /* The aspect must be exactly one of the image aspects. */
> > > assert(_mesa_bitcount(aspect) == 1 && (aspect & image->aspects));
> > > +
> > > + /* The only compression types with more than just fast-clears are
> > MCS,
> > > + * CCS_E, and HiZ. With HiZ we just trust the layout and don't
> > actually
> > > + * track the current fast-clear and compression state. This leaves
> > us
> > > + * with just MCS and CCS_E.
> > > + */
> > > + if (aux_usage != ISL_AUX_USAGE_CCS_E &&
> > > + aux_usage != ISL_AUX_USAGE_MCS)
> > > + return;
> > > +
> > > + set_image_compressed_bit(cmd_buffer, image, aspect,
> > > + level, base_layer, layer_count, true);
> > > }
> > >
> > > static void
> > > -init_fast_clear_state_entry(struct anv_cmd_buffer *cmd_buffer,
> > > - const struct anv_image *image,
> > > - VkImageAspectFlagBits aspect,
> > > - unsigned level)
> > > +init_fast_clear_color(struct anv_cmd_buffer *cmd_buffer,
> > > + const struct anv_image *image,
> > > + VkImageAspectFlagBits aspect)
> > > {
> > > assert(cmd_buffer && image);
> > > assert(image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
> > > - assert(level < anv_image_aux_levels(image, aspect));
> > > +
> > > + set_image_fast_clear_state(cmd_buffer, image, aspect,
> > > + ANV_FAST_CLEAR_NONE);
> > >
> > > uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
> > > enum isl_aux_usage aux_usage = image->planes[plane].aux_usage;
> > > @@ -517,7 +692,7 @@ init_fast_clear_state_entry(struct anv_cmd_buffer
> > *cmd_buffer,
> > > * values in the clear value dword(s).
> > > */
> > > struct anv_address addr =
> > > - anv_image_get_clear_color_addr(cmd_buffer->device, image,
> > aspect, level);
> > > + anv_image_get_clear_color_addr(cmd_buffer->device, image,
> > aspect);
> > > unsigned i = 0;
> > > for (; i < cmd_buffer->device->isl_dev.ss.clear_value_size; i += 4)
> > {
> > > anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_DATA_IMM), sdi)
> > {
> > > @@ -558,19 +733,17 @@ genX(copy_fast_clear_dwords)(struct
> > anv_cmd_buffer *cmd_buffer,
> > > struct anv_state surface_state,
> > > const struct anv_image *image,
> > > VkImageAspectFlagBits aspect,
> > > - unsigned level,
> > > bool copy_from_surface_state)
> > > {
> > > assert(cmd_buffer && image);
> > > assert(image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
> > > - assert(level < anv_image_aux_levels(image, aspect));
> > >
> > > struct anv_bo *ss_bo =
> > > &cmd_buffer->device->surface_state_pool.block_pool.bo;
> > > uint32_t ss_clear_offset = surface_state.offset +
> > > cmd_buffer->device->isl_dev.ss.clear_value_offset;
> > > const struct anv_address entry_addr =
> > > - anv_image_get_clear_color_addr(cmd_buffer->device, image,
> > aspect, level);
> > > + anv_image_get_clear_color_addr(cmd_buffer->device, image,
> > aspect);
> > > unsigned copy_size = cmd_buffer->device->isl_dev.
> > ss.clear_value_size;
> > >
> > > if (copy_from_surface_state) {
> > > @@ -660,18 +833,6 @@ transition_color_buffer(struct anv_cmd_buffer
> > *cmd_buffer,
> > > if (base_layer >= anv_image_aux_layers(image, aspect, base_level))
> > > return;
> > >
> > > - /* A transition of a 3D subresource works on all slices at a time. */
> > > - if (image->type == VK_IMAGE_TYPE_3D) {
> > > - base_layer = 0;
> > > - layer_count = anv_minify(image->extent.depth, base_level);
> > > - }
> > > -
> > > - /* We're interested in the subresource range subset that has aux
> > data. */
> > > - level_count = MIN2(level_count, anv_image_aux_levels(image, aspect)
> > - base_level);
> > > - layer_count = MIN2(layer_count,
> > > - anv_image_aux_layers(image, aspect, base_level) -
> > base_layer);
> > > - last_level_num = base_level + level_count;
> > > -
> > > assert(image->tiling == VK_IMAGE_TILING_OPTIMAL);
> > >
> > > if (initial_layout == VK_IMAGE_LAYOUT_UNDEFINED ||
> > > @@ -684,8 +845,8 @@ transition_color_buffer(struct anv_cmd_buffer
> > *cmd_buffer,
> > > *
> > > * Initialize the relevant clear buffer entries.
> > > */
> > > - for (unsigned level = base_level; level < last_level_num; level++)
> > > - init_fast_clear_state_entry(cmd_buffer, image, aspect, level);
> > > + if (base_level == 0 && base_layer == 0)
> > > + init_fast_clear_color(cmd_buffer, image, aspect);
> > >
> > > /* Initialize the aux buffers to enable correct rendering. In
> > order to
> > > * ensure that things such as storage images work correctly, aux
> > buffers
> > > @@ -723,13 +884,18 @@ transition_color_buffer(struct anv_cmd_buffer
> > *cmd_buffer,
> > > if (image->samples == 1) {
> > > for (uint32_t l = 0; l < level_count; l++) {
> > > const uint32_t level = base_level + l;
> > > - const uint32_t level_layer_count =
> > > + uint32_t level_layer_count =
> > > MIN2(layer_count, anv_image_aux_layers(image, aspect,
> > level));
> > > +
> > > anv_image_ccs_op(cmd_buffer, image, aspect, level,
> > > base_layer, level_layer_count,
> > > ISL_AUX_OP_AMBIGUATE, false);
> >
> > On gen8, we're doing extra ambiguates when the level is greater than 0.
> >
>
> Only if we never render to it with CCS_D enabled on miplevels > 0. I'm not
> sure if that's guaranteed by the current code.
>
You're right. Could you put a perf_warn for rendering with CCS_D enabled
on miplevels > 0 for gen8 in the patch,
"anv: Only fast clear single-slice images"?
-Nanley
> --Jason
>
>
> > -Nanley
> >
> > > - genX(set_image_needs_resolve)(cmd_buffer, image,
> > > - aspect, level, false);
> > > +
> > > + if (image->planes[plane].aux_usage == ISL_AUX_USAGE_CCS_E)
> > {
> > > + set_image_compressed_bit(cmd_buffer, image, aspect,
> > > + level, base_layer,
> > level_layer_count,
> > > + false);
> > > + }
> > > }
> > > } else {
> > > if (image->samples == 4 || image->samples == 16) {
> > > @@ -812,19 +978,17 @@ transition_color_buffer(struct anv_cmd_buffer
> > *cmd_buffer,
> > > cmd_buffer->state.pending_pipe_bits |=
> > > ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT | ANV_PIPE_CS_STALL_BIT;
> > >
> > > - for (uint32_t level = base_level; level < last_level_num; level++) {
> > > + for (uint32_t l = 0; l < level_count; l++) {
> > > + uint32_t level = base_level + l;
> > > + uint32_t level_layer_count =
> > > + MIN2(layer_count, anv_image_aux_layers(image, aspect, level));
> > >
> > > - /* The number of layers changes at each 3D miplevel. */
> > > - if (image->type == VK_IMAGE_TYPE_3D) {
> > > - layer_count = MIN2(layer_count, anv_image_aux_layers(image,
> > aspect, level));
> > > + for (uint32_t a = 0; a < level_layer_count; a++) {
> > > + uint32_t array_layer = base_layer + a;
> > > + anv_cmd_predicated_ccs_resolve(cmd_buffer, image, aspect,
> > > + level, array_layer, resolve_op,
> > > + final_fast_clear);
> > > }
> > > -
> > > - genX(load_needs_resolve_predicate)(cmd_buffer, image, aspect,
> > level);
> > > -
> > > - anv_image_ccs_op(cmd_buffer, image, aspect, level,
> > > - base_layer, layer_count, resolve_op, true);
> > > -
> > > - genX(set_image_needs_resolve)(cmd_buffer, image, aspect, level,
> > false);
> > > }
> > >
> > > cmd_buffer->state.pending_pipe_bits |=
> > > @@ -1488,12 +1652,20 @@ void genX(CmdPipelineBarrier)(
> > > anv_image_expand_aspects(image, range->aspectMask);
> > > uint32_t aspect_bit;
> > >
> > > + uint32_t base_layer, layer_count;
> > > + if (image->type == VK_IMAGE_TYPE_3D) {
> > > + base_layer = 0;
> > > + layer_count = anv_minify(image->extent.depth,
> > range->baseMipLevel);
> > > + } else {
> > > + base_layer = range->baseArrayLayer;
> > > + layer_count = anv_get_layerCount(image, range);
> > > + }
> > > +
> > > anv_foreach_image_aspect_bit(aspect_bit, image,
> > color_aspects) {
> > > transition_color_buffer(cmd_buffer, image, 1UL <<
> > aspect_bit,
> > > range->baseMipLevel,
> > > anv_get_levelCount(image, range),
> > > - range->baseArrayLayer,
> > > - anv_get_layerCount(image, range),
> > > + base_layer, layer_count,
> > > pImageMemoryBarriers[i].oldLayout,
> > > pImageMemoryBarriers[i].newLayout);
> > > }
> > > @@ -3203,28 +3375,26 @@ cmd_buffer_subpass_sync_fast_clear_values(struct
> > anv_cmd_buffer *cmd_buffer)
> > > genX(copy_fast_clear_dwords)(cmd_buffer,
> > att_state->color.state,
> > > iview->image,
> > > VK_IMAGE_ASPECT_COLOR_BIT,
> > > - iview->planes[0].isl.base_level,
> > > true /* copy from ss */);
> > >
> > > /* Fast-clears impact whether or not a resolve will be
> > necessary. */
> > > - if (iview->image->planes[0].aux_usage == ISL_AUX_USAGE_CCS_E
> > &&
> > > - att_state->clear_color_is_zero) {
> > > + if (att_state->clear_color_is_zero) {
> > > /* This image always has the auxiliary buffer enabled. We
> > can mark
> > > * the subresource as not needing a resolve because the
> > clear color
> > > * will match what's in every RENDER_SURFACE_STATE object
> > when it's
> > > * being used for sampling.
> > > */
> > > - genX(set_image_needs_resolve)(cmd_buffer, iview->image,
> > > - VK_IMAGE_ASPECT_COLOR_BIT,
> > > - iview->planes[0].isl.base_
> > level,
> > > - false);
> > > + set_image_fast_clear_state(cmd_buffer, iview->image,
> > > + VK_IMAGE_ASPECT_COLOR_BIT,
> > > + ANV_FAST_CLEAR_DEFAULT_VALUE);
> > > } else {
> > > - genX(set_image_needs_resolve)(cmd_buffer, iview->image,
> > > - VK_IMAGE_ASPECT_COLOR_BIT,
> > > - iview->planes[0].isl.base_
> > level,
> > > - true);
> > > + set_image_fast_clear_state(cmd_buffer, iview->image,
> > > + VK_IMAGE_ASPECT_COLOR_BIT,
> > > + ANV_FAST_CLEAR_ANY);
> > > }
> > > - } else if (rp_att->load_op == VK_ATTACHMENT_LOAD_OP_LOAD) {
> > > + } else if (rp_att->load_op == VK_ATTACHMENT_LOAD_OP_LOAD &&
> > > + iview->planes[0].isl.base_level == 0 &&
> > > + iview->planes[0].isl.base_array_layer == 0) {
> > > /* The attachment may have been fast-cleared in a previous
> > render
> > > * pass and the value is needed now. Update the surface
> > state(s).
> > > *
> > > @@ -3233,7 +3403,6 @@ cmd_buffer_subpass_sync_fast_clear_values(struct
> > anv_cmd_buffer *cmd_buffer)
> > > genX(copy_fast_clear_dwords)(cmd_buffer,
> > att_state->color.state,
> > > iview->image,
> > > VK_IMAGE_ASPECT_COLOR_BIT,
> > > - iview->planes[0].isl.base_level,
> > > false /* copy to ss */);
> > >
> > > if (need_input_attachment_state(rp_att) &&
> > > @@ -3241,7 +3410,6 @@ cmd_buffer_subpass_sync_fast_clear_values(struct
> > anv_cmd_buffer *cmd_buffer)
> > > genX(copy_fast_clear_dwords)(cmd_buffer,
> > att_state->input.state,
> > > iview->image,
> > > VK_IMAGE_ASPECT_COLOR_BIT,
> > > - iview->planes[0].isl.base_
> > level,
> > > false /* copy to ss */);
> > > }
> > > }
> > > --
> > > 2.5.0.400.gff86faf
> > >
> > > _______________________________________________
> > > 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