#include "brw_eu.h"
#include "brw_state.h"
-struct brw_blorp_clear_prog_key
+#define FILE_DEBUG_FLAG DEBUG_BLORP
+
+struct brw_blorp_const_color_prog_key
{
bool use_simd16_replicated_data;
bool pad[3];
};
-class brw_blorp_clear_params : public brw_blorp_params
+/**
+ * Parameters for a blorp operation where the fragment shader outputs a
+ * constant color. This is used for both fast color clears and color
+ * resolves.
+ */
+class brw_blorp_const_color_params : public brw_blorp_params
+{
+public:
+ virtual uint32_t get_wm_prog(struct brw_context *brw,
+ brw_blorp_prog_data **prog_data) const;
+
+protected:
+ brw_blorp_const_color_prog_key wm_prog_key;
+};
+
+class brw_blorp_clear_params : public brw_blorp_const_color_params
{
public:
brw_blorp_clear_params(struct brw_context *brw,
struct gl_framebuffer *fb,
struct gl_renderbuffer *rb,
- GLubyte *color_mask);
+ GLubyte *color_mask,
+ bool partial_clear);
+};
- virtual uint32_t get_wm_prog(struct brw_context *brw,
- brw_blorp_prog_data **prog_data) const;
-private:
- brw_blorp_clear_prog_key wm_prog_key;
+/**
+ * Parameters for a blorp operation that performs a "render target resolve".
+ * This is used to resolve pending fast clear pixels before a color buffer is
+ * used for texturing, ReadPixels, or scanout.
+ */
+class brw_blorp_rt_resolve_params : public brw_blorp_const_color_params
+{
+public:
+ brw_blorp_rt_resolve_params(struct brw_context *brw,
+ struct intel_mipmap_tree *mt);
};
-class brw_blorp_clear_program
+
+class brw_blorp_const_color_program
{
public:
- brw_blorp_clear_program(struct brw_context *brw,
- const brw_blorp_clear_prog_key *key);
- ~brw_blorp_clear_program();
+ brw_blorp_const_color_program(struct brw_context *brw,
+ const brw_blorp_const_color_prog_key *key);
+ ~brw_blorp_const_color_program();
const GLuint *compile(struct brw_context *brw, GLuint *program_size);
void *mem_ctx;
struct brw_context *brw;
- const brw_blorp_clear_prog_key *key;
+ const brw_blorp_const_color_prog_key *key;
struct brw_compile func;
/* Thread dispatch header */
GLuint base_mrf;
};
-brw_blorp_clear_program::brw_blorp_clear_program(
+brw_blorp_const_color_program::brw_blorp_const_color_program(
struct brw_context *brw,
- const brw_blorp_clear_prog_key *key)
+ const brw_blorp_const_color_prog_key *key)
: mem_ctx(ralloc_context(NULL)),
brw(brw),
- key(key)
+ key(key),
+ R0(),
+ R1(),
+ clear_rgba(),
+ base_mrf(0)
{
brw_init_compile(brw, &func, mem_ctx);
}
-brw_blorp_clear_program::~brw_blorp_clear_program()
+brw_blorp_const_color_program::~brw_blorp_const_color_program()
{
ralloc_free(mem_ctx);
}
+
+/**
+ * Determine if fast color clear supports the given clear color.
+ *
+ * Fast color clear can only clear to color values of 1.0 or 0.0. At the
+ * moment we only support floating point, unorm, and snorm buffers.
+ */
+static bool
+is_color_fast_clear_compatible(struct brw_context *brw,
+ gl_format format,
+ const union gl_color_union *color)
+{
+ if (_mesa_is_format_integer_color(format))
+ return false;
+
+ for (int i = 0; i < 4; i++) {
+ if (color->f[i] != 0.0 && color->f[i] != 1.0) {
+ perf_debug("Clear color unsupported by fast color clear. "
+ "Falling back to slow clear.\n");
+ return false;
+ }
+ }
+ return true;
+}
+
+
+/**
+ * Convert the given color to a bitfield suitable for ORing into DWORD 7 of
+ * SURFACE_STATE.
+ */
+static uint32_t
+compute_fast_clear_color_bits(const union gl_color_union *color)
+{
+ uint32_t bits = 0;
+ for (int i = 0; i < 4; i++) {
+ if (color->f[i] != 0.0)
+ bits |= 1 << (GEN7_SURFACE_CLEAR_COLOR_SHIFT + (3 - i));
+ }
+ return bits;
+}
+
+
brw_blorp_clear_params::brw_blorp_clear_params(struct brw_context *brw,
struct gl_framebuffer *fb,
struct gl_renderbuffer *rb,
- GLubyte *color_mask)
+ GLubyte *color_mask,
+ bool partial_clear)
{
- struct intel_context *intel = &brw->intel;
- struct gl_context *ctx = &intel->ctx;
+ struct gl_context *ctx = &brw->ctx;
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
dst.set(brw, irb->mt, irb->mt_level, irb->mt_layer);
wm_prog_key.use_simd16_replicated_data = false;
}
}
+
+ /* If we can do this as a fast color clear, do so. */
+ if (irb->mt->mcs_state != INTEL_MCS_STATE_NONE && !partial_clear &&
+ wm_prog_key.use_simd16_replicated_data &&
+ is_color_fast_clear_compatible(brw, format, &ctx->Color.ClearColor)) {
+ memset(push_consts, 0xff, 4*sizeof(float));
+ fast_clear_op = GEN7_FAST_CLEAR_OP_FAST_CLEAR;
+
+ /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
+ * Target(s)", beneath the "Fast Color Clear" bullet (p327):
+ *
+ * Clear pass must have a clear rectangle that must follow alignment
+ * rules in terms of pixels and lines as shown in the table
+ * below. Further, the clear-rectangle height and width must be
+ * multiple of the following dimensions. If the height and width of
+ * the render target being cleared do not meet these requirements,
+ * an MCS buffer can be created such that it follows the requirement
+ * and covers the RT.
+ *
+ * The alignment size in the table that follows is related to the
+ * alignment size returned by intel_get_non_msrt_mcs_alignment(), but
+ * with X alignment multiplied by 16 and Y alignment multiplied by 32.
+ */
+ unsigned x_align, y_align;
+ intel_get_non_msrt_mcs_alignment(brw, irb->mt, &x_align, &y_align);
+ x_align *= 16;
+ y_align *= 32;
+
+ if (brw->is_haswell && brw->gt == 3) {
+ /* From BSpec: 3D-Media-GPGPU Engine > 3D Pipeline > Pixel > Pixel
+ * Backend > MCS Buffer for Render Target(s) [DevIVB+]:
+ * [DevHSW:GT3]: Clear rectangle must be aligned to two times the
+ * number of pixels in the table shown below...
+ * x_align, y_align values computed above are the relevant entries
+ * in the referred table.
+ */
+ x0 = ROUND_DOWN_TO(x0, 2 * x_align);
+ y0 = ROUND_DOWN_TO(y0, 2 * y_align);
+ x1 = ALIGN(x1, 2 * x_align);
+ y1 = ALIGN(y1, 2 * y_align);
+ } else {
+ x0 = ROUND_DOWN_TO(x0, x_align);
+ y0 = ROUND_DOWN_TO(y0, y_align);
+ x1 = ALIGN(x1, x_align);
+ y1 = ALIGN(y1, y_align);
+ }
+
+ /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
+ * Target(s)", beneath the "Fast Color Clear" bullet (p327):
+ *
+ * In order to optimize the performance MCS buffer (when bound to 1X
+ * RT) clear similarly to MCS buffer clear for MSRT case, clear rect
+ * is required to be scaled by the following factors in the
+ * horizontal and vertical directions:
+ *
+ * The X and Y scale down factors in the table that follows are each
+ * equal to half the alignment value computed above.
+ */
+ unsigned x_scaledown = x_align / 2;
+ unsigned y_scaledown = y_align / 2;
+ x0 /= x_scaledown;
+ y0 /= y_scaledown;
+ x1 /= x_scaledown;
+ y1 /= y_scaledown;
+ }
+}
+
+
+brw_blorp_rt_resolve_params::brw_blorp_rt_resolve_params(
+ struct brw_context *brw,
+ struct intel_mipmap_tree *mt)
+{
+ dst.set(brw, mt, 0 /* level */, 0 /* layer */);
+
+ /* From the Ivy Bridge PRM, Vol2 Part1 11.9 "Render Target Resolve":
+ *
+ * A rectangle primitive must be scaled down by the following factors
+ * with respect to render target being resolved.
+ *
+ * The scaledown factors in the table that follows are related to the
+ * alignment size returned by intel_get_non_msrt_mcs_alignment(), but with
+ * X and Y alignment each divided by 2.
+ */
+ unsigned x_align, y_align;
+ intel_get_non_msrt_mcs_alignment(brw, mt, &x_align, &y_align);
+ unsigned x_scaledown = x_align / 2;
+ unsigned y_scaledown = y_align / 2;
+ x0 = y0 = 0;
+ x1 = ALIGN(mt->logical_width0, x_scaledown) / x_scaledown;
+ y1 = ALIGN(mt->logical_height0, y_scaledown) / y_scaledown;
+
+ fast_clear_op = GEN7_FAST_CLEAR_OP_RESOLVE;
+
+ /* Note: there is no need to initialize push constants because it doesn't
+ * matter what data gets dispatched to the render target. However, we must
+ * ensure that the fragment shader delivers the data using the "replicated
+ * color" message.
+ */
+ use_wm_prog = true;
+ memset(&wm_prog_key, 0, sizeof(wm_prog_key));
+ wm_prog_key.use_simd16_replicated_data = true;
}
+
uint32_t
-brw_blorp_clear_params::get_wm_prog(struct brw_context *brw,
- brw_blorp_prog_data **prog_data) const
+brw_blorp_const_color_params::get_wm_prog(struct brw_context *brw,
+ brw_blorp_prog_data **prog_data)
+ const
{
- uint32_t prog_offset;
- if (!brw_search_cache(&brw->cache, BRW_BLORP_CLEAR_PROG,
+ uint32_t prog_offset = 0;
+ if (!brw_search_cache(&brw->cache, BRW_BLORP_CONST_COLOR_PROG,
&this->wm_prog_key, sizeof(this->wm_prog_key),
&prog_offset, prog_data)) {
- brw_blorp_clear_program prog(brw, &this->wm_prog_key);
+ brw_blorp_const_color_program prog(brw, &this->wm_prog_key);
GLuint program_size;
const GLuint *program = prog.compile(brw, &program_size);
- brw_upload_cache(&brw->cache, BRW_BLORP_CLEAR_PROG,
+ brw_upload_cache(&brw->cache, BRW_BLORP_CONST_COLOR_PROG,
&this->wm_prog_key, sizeof(this->wm_prog_key),
program, program_size,
&prog.prog_data, sizeof(prog.prog_data),
}
void
-brw_blorp_clear_program::alloc_regs()
+brw_blorp_const_color_program::alloc_regs()
{
int reg = 0;
this->R0 = retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW);
}
const GLuint *
-brw_blorp_clear_program::compile(struct brw_context *brw,
- GLuint *program_size)
+brw_blorp_const_color_program::compile(struct brw_context *brw,
+ GLuint *program_size)
{
/* Set up prog_data */
memset(&prog_data, 0, sizeof(prog_data));
extern "C" {
bool
-brw_blorp_clear_color(struct intel_context *intel, struct gl_framebuffer *fb)
+brw_blorp_clear_color(struct brw_context *brw, struct gl_framebuffer *fb,
+ bool partial_clear)
{
- struct gl_context *ctx = &intel->ctx;
- struct brw_context *brw = brw_context(ctx);
+ struct gl_context *ctx = &brw->ctx;
/* The constant color clear code doesn't work for multisampled surfaces, so
* we need to support falling back to other clear mechanisms.
for (unsigned buf = 0; buf < ctx->DrawBuffer->_NumColorDrawBuffers; buf++) {
struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[buf];
+ struct intel_renderbuffer *irb = intel_renderbuffer(rb);
/* If this is an ES2 context or GL_ARB_ES2_compatibility is supported,
* the framebuffer can be complete with some attachments missing. In
if (rb == NULL)
continue;
- brw_blorp_clear_params params(brw, fb, rb, ctx->Color.ColorMask[buf]);
- brw_blorp_exec(intel, ¶ms);
+ brw_blorp_clear_params params(brw, fb, rb, ctx->Color.ColorMask[buf],
+ partial_clear);
+
+ bool is_fast_clear =
+ (params.fast_clear_op == GEN7_FAST_CLEAR_OP_FAST_CLEAR);
+ if (is_fast_clear) {
+ /* Record the clear color in the miptree so that it will be
+ * programmed in SURFACE_STATE by later rendering and resolve
+ * operations.
+ */
+ uint32_t new_color_value =
+ compute_fast_clear_color_bits(&ctx->Color.ClearColor);
+ if (irb->mt->fast_clear_color_value != new_color_value) {
+ irb->mt->fast_clear_color_value = new_color_value;
+ brw->state.dirty.brw |= BRW_NEW_SURFACES;
+ }
+
+ /* If the buffer is already in INTEL_MCS_STATE_CLEAR, the clear is
+ * redundant and can be skipped.
+ */
+ if (irb->mt->mcs_state == INTEL_MCS_STATE_CLEAR)
+ continue;
+
+ /* If the MCS buffer hasn't been allocated yet, we need to allocate
+ * it now.
+ */
+ if (!irb->mt->mcs_mt) {
+ if (!intel_miptree_alloc_non_msrt_mcs(brw, irb->mt)) {
+ /* MCS allocation failed--probably this will only happen in
+ * out-of-memory conditions. But in any case, try to recover
+ * by falling back to a non-blorp clear technique.
+ */
+ return false;
+ }
+ brw->state.dirty.brw |= BRW_NEW_SURFACES;
+ }
+ }
+
+ DBG("%s to mt %p level %d layer %d\n", __FUNCTION__,
+ irb->mt, irb->mt_level, irb->mt_layer);
+
+ brw_blorp_exec(brw, ¶ms);
+
+ if (is_fast_clear) {
+ /* Now that the fast clear has occurred, put the buffer in
+ * INTEL_MCS_STATE_CLEAR so that we won't waste time doing redundant
+ * clears.
+ */
+ irb->mt->mcs_state = INTEL_MCS_STATE_CLEAR;
+ }
}
return true;
}
+void
+brw_blorp_resolve_color(struct brw_context *brw, struct intel_mipmap_tree *mt)
+{
+ DBG("%s to mt %p\n", __FUNCTION__, mt);
+
+ brw_blorp_rt_resolve_params params(brw, mt);
+ brw_blorp_exec(brw, ¶ms);
+ mt->mcs_state = INTEL_MCS_STATE_RESOLVED;
+}
+
} /* extern "C" */