#include "anv_private.h"
#include "gen8_pack.h"
+#include "gen9_pack.h"
+
+static const uint8_t
+anv_surftype(const struct anv_image *image, VkImageViewType view_type)
+{
+ switch (view_type) {
+ default:
+ unreachable("bad VkImageViewType");
+ case VK_IMAGE_VIEW_TYPE_1D:
+ case VK_IMAGE_VIEW_TYPE_1D_ARRAY:
+ assert(image->type == VK_IMAGE_TYPE_1D);
+ return SURFTYPE_1D;
+ case VK_IMAGE_VIEW_TYPE_CUBE:
+ case VK_IMAGE_VIEW_TYPE_CUBE_ARRAY:
+ assert(image->type == VK_IMAGE_TYPE_2D);
+ return SURFTYPE_CUBE;
+ case VK_IMAGE_VIEW_TYPE_2D:
+ case VK_IMAGE_VIEW_TYPE_2D_ARRAY:
+ assert(image->type == VK_IMAGE_TYPE_2D);
+ return SURFTYPE_2D;
+ case VK_IMAGE_VIEW_TYPE_3D:
+ assert(image->type == VK_IMAGE_TYPE_3D);
+ return SURFTYPE_3D;
+ }
+}
void
-gen8_fill_buffer_surface_state(void *state, const struct anv_format *format,
- uint32_t offset, uint32_t range, uint32_t stride)
+genX(fill_buffer_surface_state)(void *state, const struct anv_format *format,
+ uint32_t offset, uint32_t range, uint32_t stride)
{
uint32_t num_elements = range / stride;
- struct GEN8_RENDER_SURFACE_STATE surface_state = {
+ struct GENX(RENDER_SURFACE_STATE) surface_state = {
.SurfaceType = SURFTYPE_BUFFER,
.SurfaceArray = false,
.SurfaceFormat = format->surface_format,
.TileMode = LINEAR,
.SamplerL2BypassModeDisable = true,
.RenderCacheReadWriteMode = WriteOnlyCache,
- .MemoryObjectControlState = GEN8_MOCS,
+ .MemoryObjectControlState = GENX(MOCS),
.Height = ((num_elements - 1) >> 7) & 0x3fff,
.Width = (num_elements - 1) & 0x7f,
.Depth = ((num_elements - 1) >> 21) & 0x3f,
.SurfaceBaseAddress = { NULL, offset },
};
- GEN8_RENDER_SURFACE_STATE_pack(NULL, state, &surface_state);
+ GENX(RENDER_SURFACE_STATE_pack)(NULL, state, &surface_state);
}
static const uint8_t anv_halign[] = {
};
static struct anv_state
-gen8_alloc_surface_state(struct anv_device *device,
- struct anv_cmd_buffer *cmd_buffer)
+alloc_surface_state(struct anv_device *device,
+ struct anv_cmd_buffer *cmd_buffer)
{
if (cmd_buffer) {
return anv_cmd_buffer_alloc_surface_state(cmd_buffer);
}
}
+static const uint32_t vk_to_gen_swizzle_map[] = {
+ [VK_COMPONENT_SWIZZLE_ZERO] = SCS_ZERO,
+ [VK_COMPONENT_SWIZZLE_ONE] = SCS_ONE,
+ [VK_COMPONENT_SWIZZLE_R] = SCS_RED,
+ [VK_COMPONENT_SWIZZLE_G] = SCS_GREEN,
+ [VK_COMPONENT_SWIZZLE_B] = SCS_BLUE,
+ [VK_COMPONENT_SWIZZLE_A] = SCS_ALPHA
+};
+
+static inline uint32_t
+vk_to_gen_swizzle(VkComponentSwizzle swizzle, VkComponentSwizzle component)
+{
+ if (swizzle == VK_COMPONENT_SWIZZLE_IDENTITY)
+ return vk_to_gen_swizzle_map[component];
+ else
+ return vk_to_gen_swizzle_map[swizzle];
+}
+
+/**
+ * Get the values to pack into RENDER_SUFFACE_STATE.SurfaceHorizontalAlignment
+ * and SurfaceVerticalAlignment.
+ */
+static void
+get_halign_valign(const struct isl_surf *surf, uint32_t *halign, uint32_t *valign)
+{
+ #if ANV_GENx10 >= 90
+ /* In Skylake, RENDER_SUFFACE_STATE.SurfaceVerticalAlignment is in units
+ * of surface elements (not pixels nor samples). For compressed formats,
+ * a "surface element" is defined as a compression block. For example,
+ * if SurfaceVerticalAlignment is VALIGN_4 and SurfaceFormat is an ETC2
+ * format (ETC2 has a block height of 4), then the vertical alignment is
+ * 4 compression blocks or, equivalently, 16 pixels.
+ */
+ struct isl_extent3d image_align_el
+ = isl_surf_get_image_alignment_el(surf);
+
+ *halign = anv_halign[image_align_el.width];
+ *valign = anv_valign[image_align_el.height];
+ #else
+ /* Pre-Skylake, RENDER_SUFFACE_STATE.SurfaceVerticalAlignment is in
+ * units of surface samples. For example, if SurfaceVerticalAlignment
+ * is VALIGN_4 and the surface is singlesampled, then for any surface
+ * format (compressed or not) the vertical alignment is
+ * 4 pixels.
+ */
+ struct isl_extent3d image_align_sa
+ = isl_surf_get_image_alignment_sa(surf);
+
+ *halign = anv_halign[image_align_sa.width];
+ *valign = anv_valign[image_align_sa.height];
+ #endif
+}
+
void
-gen8_image_view_init(struct anv_image_view *iview,
- struct anv_device *device,
- const VkImageViewCreateInfo* pCreateInfo,
- struct anv_cmd_buffer *cmd_buffer)
+genX(image_view_init)(struct anv_image_view *iview,
+ struct anv_device *device,
+ const VkImageViewCreateInfo* pCreateInfo,
+ struct anv_cmd_buffer *cmd_buffer)
{
ANV_FROM_HANDLE(anv_image, image, pCreateInfo->image);
* example, if Minimum Array Element is set to 1024 on a 2D surface,
* the range of this field is reduced to [0,1023].
*/
- depth = range->arraySize;
+ depth = range->layerCount;
/* From the Broadwell PRM >> RENDER_SURFACE_STATE::RenderTargetViewExtent:
*
unreachable(!"bad VkImageType");
}
- static const uint32_t vk_to_gen_swizzle[] = {
- [VK_CHANNEL_SWIZZLE_ZERO] = SCS_ZERO,
- [VK_CHANNEL_SWIZZLE_ONE] = SCS_ONE,
- [VK_CHANNEL_SWIZZLE_R] = SCS_RED,
- [VK_CHANNEL_SWIZZLE_G] = SCS_GREEN,
- [VK_CHANNEL_SWIZZLE_B] = SCS_BLUE,
- [VK_CHANNEL_SWIZZLE_A] = SCS_ALPHA
- };
-
static const uint8_t isl_to_gen_tiling[] = {
[ISL_TILING_LINEAR] = LINEAR,
[ISL_TILING_X] = XMAJOR,
- [ISL_TILING_Y] = YMAJOR,
+ [ISL_TILING_Y0] = YMAJOR,
[ISL_TILING_Yf] = YMAJOR,
[ISL_TILING_Ys] = YMAJOR,
[ISL_TILING_W] = WMAJOR,
};
- struct GEN8_RENDER_SURFACE_STATE surface_state = {
- .SurfaceType = image->surface_type,
+ uint32_t halign, valign;
+ get_halign_valign(&surface->isl, &halign, &valign);
+
+ struct GENX(RENDER_SURFACE_STATE) surface_state = {
+ .SurfaceType = anv_surftype(image, pCreateInfo->viewType),
.SurfaceArray = image->array_size > 1,
.SurfaceFormat = format_info->surface_format,
- .SurfaceVerticalAlignment = anv_valign[surface->v_align],
- .SurfaceHorizontalAlignment = anv_halign[surface->h_align],
- .TileMode = isl_to_gen_tiling[surface->tiling],
+ .SurfaceVerticalAlignment = valign,
+ .SurfaceHorizontalAlignment = halign,
+ .TileMode = isl_to_gen_tiling[surface->isl.tiling],
.VerticalLineStride = 0,
.VerticalLineStrideOffset = 0,
.SamplerL2BypassModeDisable = true,
.RenderCacheReadWriteMode = WriteOnlyCache,
- .MemoryObjectControlState = GEN8_MOCS,
+ .MemoryObjectControlState = GENX(MOCS),
/* The driver sets BaseMipLevel in SAMPLER_STATE, not here in
* RENDER_SURFACE_STATE. The Broadwell PRM says "it is illegal to have
*/
.BaseMipLevel = 0.0,
- .SurfaceQPitch = surface->qpitch >> 2,
+ .SurfaceQPitch = isl_surf_get_array_pitch_el_rows(&surface->isl) >> 2,
.Height = image->extent.height - 1,
.Width = image->extent.width - 1,
.Depth = depth - 1,
- .SurfacePitch = surface->stride - 1,
+ .SurfacePitch = surface->isl.row_pitch - 1,
.RenderTargetViewExtent = rt_view_extent - 1,
.MinimumArrayElement = range->baseArrayLayer,
.NumberofMultisamples = MULTISAMPLECOUNT_1,
.GreenClearColor = 0,
.BlueClearColor = 0,
.AlphaClearColor = 0,
- .ShaderChannelSelectRed = vk_to_gen_swizzle[pCreateInfo->channels.r],
- .ShaderChannelSelectGreen = vk_to_gen_swizzle[pCreateInfo->channels.g],
- .ShaderChannelSelectBlue = vk_to_gen_swizzle[pCreateInfo->channels.b],
- .ShaderChannelSelectAlpha = vk_to_gen_swizzle[pCreateInfo->channels.a],
+ .ShaderChannelSelectRed = vk_to_gen_swizzle(pCreateInfo->components.r,
+ VK_COMPONENT_SWIZZLE_R),
+ .ShaderChannelSelectGreen = vk_to_gen_swizzle(pCreateInfo->components.g,
+ VK_COMPONENT_SWIZZLE_G),
+ .ShaderChannelSelectBlue = vk_to_gen_swizzle(pCreateInfo->components.b,
+ VK_COMPONENT_SWIZZLE_B),
+ .ShaderChannelSelectAlpha = vk_to_gen_swizzle(pCreateInfo->components.a,
+ VK_COMPONENT_SWIZZLE_A),
.ResourceMinLOD = 0.0,
.SurfaceBaseAddress = { NULL, iview->offset },
};
if (image->needs_nonrt_surface_state) {
iview->nonrt_surface_state =
- gen8_alloc_surface_state(device, cmd_buffer);
+ alloc_surface_state(device, cmd_buffer);
/* For non render target surfaces, the hardware interprets field
* MIPCount/LOD as MIPCount. The range of levels accessible by the
* sampler engine is [SurfaceMinLOD, SurfaceMinLOD + MIPCountLOD].
*/
surface_state.SurfaceMinLOD = range->baseMipLevel;
- surface_state.MIPCountLOD = range->mipLevels - 1;
+ surface_state.MIPCountLOD = range->levelCount - 1;
- GEN8_RENDER_SURFACE_STATE_pack(NULL, iview->nonrt_surface_state.map,
- &surface_state);
+ GENX(RENDER_SURFACE_STATE_pack)(NULL, iview->nonrt_surface_state.map,
+ &surface_state);
+ if (!device->info.has_llc)
+ anv_state_clflush(iview->nonrt_surface_state);
}
if (image->needs_color_rt_surface_state) {
iview->color_rt_surface_state =
- gen8_alloc_surface_state(device, cmd_buffer);
+ alloc_surface_state(device, cmd_buffer);
/* For render target surfaces, the hardware interprets field
* MIPCount/LOD as LOD. The Broadwell PRM says:
surface_state.MIPCountLOD = range->baseMipLevel;
surface_state.SurfaceMinLOD = 0;
- GEN8_RENDER_SURFACE_STATE_pack(NULL, iview->color_rt_surface_state.map,
- &surface_state);
+ GENX(RENDER_SURFACE_STATE_pack)(NULL, iview->color_rt_surface_state.map,
+ &surface_state);
+ if (!device->info.has_llc)
+ anv_state_clflush(iview->color_rt_surface_state);
+ }
+
+ if (image->needs_storage_surface_state) {
+ iview->storage_surface_state =
+ alloc_surface_state(device, cmd_buffer);
+
+ surface_state.SurfaceFormat =
+ isl_lower_storage_image_format(&device->isl_dev,
+ format_info->surface_format);
+
+ surface_state.SurfaceMinLOD = range->baseMipLevel;
+ surface_state.MIPCountLOD = range->levelCount - 1;
+
+ GENX(RENDER_SURFACE_STATE_pack)(NULL, iview->storage_surface_state.map,
+ &surface_state);
}
}
-VkResult gen8_CreateSampler(
+VkResult genX(CreateSampler)(
VkDevice _device,
const VkSamplerCreateInfo* pCreateInfo,
+ const VkAllocationCallbacks* pAllocator,
VkSampler* pSampler)
{
ANV_FROM_HANDLE(anv_device, device, _device);
assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO);
- sampler = anv_device_alloc(device, sizeof(*sampler), 8,
- VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+ sampler = anv_alloc2(&device->alloc, pAllocator, sizeof(*sampler), 8,
+ VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (!sampler)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
static const uint32_t vk_to_gen_tex_filter[] = {
- [VK_TEX_FILTER_NEAREST] = MAPFILTER_NEAREST,
- [VK_TEX_FILTER_LINEAR] = MAPFILTER_LINEAR
+ [VK_FILTER_NEAREST] = MAPFILTER_NEAREST,
+ [VK_FILTER_LINEAR] = MAPFILTER_LINEAR
};
static const uint32_t vk_to_gen_mipmap_mode[] = {
- [VK_TEX_MIPMAP_MODE_BASE] = MIPFILTER_NONE,
- [VK_TEX_MIPMAP_MODE_NEAREST] = MIPFILTER_NEAREST,
- [VK_TEX_MIPMAP_MODE_LINEAR] = MIPFILTER_LINEAR
+ [VK_SAMPLER_MIPMAP_MODE_BASE] = MIPFILTER_NONE,
+ [VK_SAMPLER_MIPMAP_MODE_NEAREST] = MIPFILTER_NEAREST,
+ [VK_SAMPLER_MIPMAP_MODE_LINEAR] = MIPFILTER_LINEAR
};
static const uint32_t vk_to_gen_tex_address[] = {
- [VK_TEX_ADDRESS_MODE_WRAP] = TCM_WRAP,
- [VK_TEX_ADDRESS_MODE_MIRROR] = TCM_MIRROR,
- [VK_TEX_ADDRESS_MODE_CLAMP] = TCM_CLAMP,
- [VK_TEX_ADDRESS_MODE_MIRROR_ONCE] = TCM_MIRROR_ONCE,
- [VK_TEX_ADDRESS_MODE_CLAMP_BORDER] = TCM_CLAMP_BORDER,
+ [VK_SAMPLER_ADDRESS_MODE_REPEAT] = TCM_WRAP,
+ [VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT] = TCM_MIRROR,
+ [VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE] = TCM_CLAMP,
+ [VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE] = TCM_MIRROR_ONCE,
+ [VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER] = TCM_CLAMP_BORDER,
};
static const uint32_t vk_to_gen_compare_op[] = {
[VK_COMPARE_OP_NEVER] = PREFILTEROPNEVER,
[VK_COMPARE_OP_LESS] = PREFILTEROPLESS,
[VK_COMPARE_OP_EQUAL] = PREFILTEROPEQUAL,
- [VK_COMPARE_OP_LESS_EQUAL] = PREFILTEROPLEQUAL,
+ [VK_COMPARE_OP_LESS_OR_EQUAL] = PREFILTEROPLEQUAL,
[VK_COMPARE_OP_GREATER] = PREFILTEROPGREATER,
[VK_COMPARE_OP_NOT_EQUAL] = PREFILTEROPNOTEQUAL,
- [VK_COMPARE_OP_GREATER_EQUAL] = PREFILTEROPGEQUAL,
+ [VK_COMPARE_OP_GREATER_OR_EQUAL] = PREFILTEROPGEQUAL,
[VK_COMPARE_OP_ALWAYS] = PREFILTEROPALWAYS,
};
max_anisotropy = RATIO21;
}
- struct GEN8_SAMPLER_STATE sampler_state = {
+ struct GENX(SAMPLER_STATE) sampler_state = {
.SamplerDisable = false,
.TextureBorderColorMode = DX10OGL,
.LODPreClampMode = 0,
+#if ANV_GEN == 8
.BaseMipLevel = 0.0,
- .MipModeFilter = vk_to_gen_mipmap_mode[pCreateInfo->mipMode],
+#endif
+ .MipModeFilter = vk_to_gen_mipmap_mode[pCreateInfo->mipmapMode],
.MagModeFilter = mag_filter,
.MinModeFilter = min_filter,
.TextureLODBias = anv_clamp_f(pCreateInfo->mipLodBias, -16, 15.996),
.TCZAddressControlMode = vk_to_gen_tex_address[pCreateInfo->addressModeW],
};
- GEN8_SAMPLER_STATE_pack(NULL, sampler->state, &sampler_state);
+ GENX(SAMPLER_STATE_pack)(NULL, sampler->state, &sampler_state);
*pSampler = anv_sampler_to_handle(sampler);