/**************************************************************************
*
* Copyright 2009 VMware, Inc.
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2007 VMware, Inc.
* All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
const struct tgsi_token *tokens = shader->base.tokens;
LLVMTypeRef vec_type;
LLVMValueRef mask_ptr, mask_val;
- LLVMValueRef consts_ptr;
+ LLVMValueRef consts_ptr, num_consts_ptr;
LLVMValueRef z;
LLVMValueRef z_value, s_value;
LLVMValueRef z_fb, s_fb;
assert(zs_format_desc);
if (!shader->info.base.writes_z) {
- if (key->alpha.enabled || shader->info.base.uses_kill) {
+ if (key->alpha.enabled ||
+ key->blend.alpha_to_coverage ||
+ shader->info.base.uses_kill) {
/* With alpha test and kill, can do the depth test early
* and hopefully eliminate some quads. But need to do a
* special deferred depth write once the final mask value
vec_type = lp_build_vec_type(gallivm, type);
consts_ptr = lp_jit_context_constants(gallivm, context_ptr);
+ num_consts_ptr = lp_jit_context_num_constants(gallivm, context_ptr);
lp_build_for_loop_begin(&loop_state, gallivm,
lp_build_const_int32(gallivm, 0),
/* Build the actual shader */
lp_build_tgsi_soa(gallivm, tokens, type, &mask,
- consts_ptr, &system_values,
+ consts_ptr, num_consts_ptr, &system_values,
interp->inputs,
outputs, sampler, &shader->info.base, NULL);
}
}
+ /* Emulate Alpha to Coverage with Alpha test */
+ if (key->blend.alpha_to_coverage) {
+ int color0 = find_output_by_semantic(&shader->info.base,
+ TGSI_SEMANTIC_COLOR,
+ 0);
+
+ if (color0 != -1 && outputs[color0][3]) {
+ LLVMValueRef alpha = LLVMBuildLoad(builder, outputs[color0][3], "alpha");
+
+ lp_build_alpha_to_coverage(gallivm, type,
+ &mask, alpha,
+ (depth_mode & LATE_DEPTH_TEST) != 0);
+ }
+ }
+
/* Late Z test */
if (depth_mode & LATE_DEPTH_TEST) {
int pos0 = find_output_by_semantic(&shader->info.base,
0);
if (pos0 != -1 && outputs[pos0][2]) {
- LLVMValueRef viewport, min_depth, max_depth;
- LLVMValueRef viewport_index;
- struct lp_build_context f32_bld;
-
- assert(type.floating);
- lp_build_context_init(&f32_bld, gallivm, type);
+ z = LLVMBuildLoad(builder, outputs[pos0][2], "output.z");
/*
- * Assumes clamping of the viewport index will occur in setup/gs. Value
- * is passed through the rasterization stage via lp_rast_shader_inputs.
- *
- * See: draw_clamp_viewport_idx and lp_clamp_viewport_idx for clamping
- * semantics.
+ * Clamp according to ARB_depth_clamp semantics.
*/
- viewport_index = lp_jit_thread_data_raster_state_viewport_index(gallivm,
- thread_data_ptr);
+ if (key->depth_clamp) {
+ LLVMValueRef viewport, min_depth, max_depth;
+ LLVMValueRef viewport_index;
+ struct lp_build_context f32_bld;
- /*
- * Load the min and max depth from the lp_jit_context.viewports
- * array of lp_jit_viewport structures.
- */
- viewport = lp_llvm_viewport(context_ptr, gallivm, viewport_index);
+ assert(type.floating);
+ lp_build_context_init(&f32_bld, gallivm, type);
- /* viewports[viewport_index].min_depth */
- min_depth = LLVMBuildExtractElement(builder, viewport,
- lp_build_const_int32(gallivm, LP_JIT_VIEWPORT_MIN_DEPTH),
- "");
- min_depth = lp_build_broadcast_scalar(&f32_bld, min_depth);
+ /*
+ * Assumes clamping of the viewport index will occur in setup/gs. Value
+ * is passed through the rasterization stage via lp_rast_shader_inputs.
+ *
+ * See: draw_clamp_viewport_idx and lp_clamp_viewport_idx for clamping
+ * semantics.
+ */
+ viewport_index = lp_jit_thread_data_raster_state_viewport_index(gallivm,
+ thread_data_ptr);
- /* viewports[viewport_index].max_depth */
- max_depth = LLVMBuildExtractElement(builder, viewport,
- lp_build_const_int32(gallivm, LP_JIT_VIEWPORT_MAX_DEPTH),
- "");
- max_depth = lp_build_broadcast_scalar(&f32_bld, max_depth);
+ /*
+ * Load the min and max depth from the lp_jit_context.viewports
+ * array of lp_jit_viewport structures.
+ */
+ viewport = lp_llvm_viewport(context_ptr, gallivm, viewport_index);
- z = LLVMBuildLoad(builder, outputs[pos0][2], "output.z");
+ /* viewports[viewport_index].min_depth */
+ min_depth = LLVMBuildExtractElement(builder, viewport,
+ lp_build_const_int32(gallivm, LP_JIT_VIEWPORT_MIN_DEPTH),
+ "");
+ min_depth = lp_build_broadcast_scalar(&f32_bld, min_depth);
- /*
- * Clamp to the min and max depth values for the given viewport.
- */
- z = lp_build_clamp(&f32_bld, z, min_depth, max_depth);
+ /* viewports[viewport_index].max_depth */
+ max_depth = LLVMBuildExtractElement(builder, viewport,
+ lp_build_const_int32(gallivm, LP_JIT_VIEWPORT_MAX_DEPTH),
+ "");
+ max_depth = lp_build_broadcast_scalar(&f32_bld, max_depth);
+
+ /*
+ * Clamp to the min and max depth values for the given viewport.
+ */
+ z = lp_build_clamp(&f32_bld, z, min_depth, max_depth);
+ }
}
lp_build_depth_stencil_load_swizzled(gallivm, type,
return result;
}
+/**
+ * If RT is a smallfloat (needing denorms) format
+ */
+static INLINE int
+have_smallfloat_format(struct lp_type dst_type,
+ enum pipe_format format)
+{
+ return ((dst_type.floating && dst_type.width != 32) ||
+ /* due to format handling hacks this format doesn't have floating set
+ * here (and actually has width set to 32 too) so special case this. */
+ (format == PIPE_FORMAT_R11G11B10_FLOAT));
+}
+
/**
* Convert from memory format to blending format
const boolean is_1d = variant->key.resource_1d;
unsigned num_fullblock_fs = is_1d ? 2 * num_fs : num_fs;
+ LLVMValueRef fpstate = 0;
+
+ /* Get type from output format */
+ lp_blend_type_from_format_desc(out_format_desc, &row_type);
+ lp_mem_type_from_format_desc(out_format_desc, &dst_type);
+
+ /*
+ * Technically this code should go into lp_build_smallfloat_to_float
+ * and lp_build_float_to_smallfloat but due to the
+ * http://llvm.org/bugs/show_bug.cgi?id=6393
+ * llvm reorders the mxcsr intrinsics in a way that breaks the code.
+ * So the ordering is important here and there shouldn't be any
+ * llvm ir instrunctions in this function before
+ * this, otherwise half-float format conversions won't work
+ * (again due to llvm bug #6393).
+ */
+ if (have_smallfloat_format(dst_type, out_format)) {
+ /* We need to make sure that denorms are ok for half float
+ conversions */
+ fpstate = lp_build_fpstate_get(gallivm);
+ lp_build_fpstate_set_denorms_zero(gallivm, FALSE);
+ }
mask_type = lp_int32_vec4_type();
mask_type.length = fs_type.length;
undef_src_val = lp_build_undef(gallivm, fs_type);
#endif
-
- /* Get type from output format */
- lp_blend_type_from_format_desc(out_format_desc, &row_type);
- lp_mem_type_from_format_desc(out_format_desc, &dst_type);
-
row_type.length = fs_type.length;
vector_width = dst_type.floating ? lp_native_vector_width : lp_integer_vector_width;
dst, dst_type, dst_count, dst_alignment);
}
+ if (have_smallfloat_format(dst_type, out_format)) {
+ lp_build_fpstate_set(gallivm, fpstate);
+ }
+
if (do_branch) {
lp_build_mask_end(&mask_ctx);
}
/* Loop over color outputs / color buffers to do blending.
*/
for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
- LLVMValueRef color_ptr;
- LLVMValueRef stride;
- LLVMValueRef index = lp_build_const_int32(gallivm, cbuf);
+ if (key->cbuf_format[cbuf] != PIPE_FORMAT_NONE) {
+ LLVMValueRef color_ptr;
+ LLVMValueRef stride;
+ LLVMValueRef index = lp_build_const_int32(gallivm, cbuf);
- boolean do_branch = ((key->depth.enabled
- || key->stencil[0].enabled
- || key->alpha.enabled)
- && !shader->info.base.uses_kill);
+ boolean do_branch = ((key->depth.enabled
+ || key->stencil[0].enabled
+ || key->alpha.enabled)
+ && !shader->info.base.uses_kill);
- color_ptr = LLVMBuildLoad(builder,
- LLVMBuildGEP(builder, color_ptr_ptr, &index, 1, ""),
- "");
+ color_ptr = LLVMBuildLoad(builder,
+ LLVMBuildGEP(builder, color_ptr_ptr,
+ &index, 1, ""),
+ "");
- lp_build_name(color_ptr, "color_ptr%d", cbuf);
+ lp_build_name(color_ptr, "color_ptr%d", cbuf);
- stride = LLVMBuildLoad(builder,
- LLVMBuildGEP(builder, stride_ptr, &index, 1, ""),
- "");
+ stride = LLVMBuildLoad(builder,
+ LLVMBuildGEP(builder, stride_ptr, &index, 1, ""),
+ "");
- generate_unswizzled_blend(gallivm, cbuf, variant, key->cbuf_format[cbuf],
- num_fs, fs_type, fs_mask, fs_out_color,
- context_ptr, color_ptr, stride, partial_mask, do_branch);
+ generate_unswizzled_blend(gallivm, cbuf, variant,
+ key->cbuf_format[cbuf],
+ num_fs, fs_type, fs_mask, fs_out_color,
+ context_ptr, color_ptr, stride,
+ partial_mask, do_branch);
+ }
}
LLVMBuildRetVoid(builder);
debug_printf("blend.alpha_dst_factor = %s\n", util_dump_blend_factor(key->blend.rt[0].alpha_dst_factor, TRUE));
}
debug_printf("blend.colormask = 0x%x\n", key->blend.rt[0].colormask);
+ if (key->blend.alpha_to_coverage) {
+ debug_printf("blend.alpha_to_coverage is enabled\n");
+ }
for (i = 0; i < key->nr_samplers; ++i) {
const struct lp_static_sampler_state *sampler = &key->state[i].sampler_state;
debug_printf("sampler[%u] = \n", i);
fullcolormask &&
!key->stencil[0].enabled &&
!key->alpha.enabled &&
+ !key->blend.alpha_to_coverage &&
!key->depth.enabled &&
!shader->info.base.uses_kill
? TRUE : FALSE;
}
}
+ /*
+ * Propagate the depth clamp setting from the rasterizer state.
+ * depth_clip == 0 implies depth clamping is enabled.
+ *
+ * When clip_halfz is enabled, then always clamp the depth values.
+ */
+ if (lp->rasterizer->clip_halfz) {
+ key->depth_clamp = 1;
+ } else {
+ key->depth_clamp = (lp->rasterizer->depth_clip == 0) ? 1 : 0;
+ }
+
/* alpha test only applies if render buffer 0 is non-integer (or does not exist) */
if (!lp->framebuffer.nr_cbufs ||
+ !lp->framebuffer.cbufs[0] ||
!util_format_is_pure_integer(lp->framebuffer.cbufs[0]->format)) {
key->alpha.enabled = lp->depth_stencil->alpha.enabled;
}
}
for (i = 0; i < lp->framebuffer.nr_cbufs; i++) {
- enum pipe_format format = lp->framebuffer.cbufs[i]->format;
struct pipe_rt_blend_state *blend_rt = &key->blend.rt[i];
- const struct util_format_description *format_desc;
- key->cbuf_format[i] = format;
+ if (lp->framebuffer.cbufs[i]) {
+ enum pipe_format format = lp->framebuffer.cbufs[i]->format;
+ const struct util_format_description *format_desc;
- /*
- * Figure out if this is a 1d resource. Note that OpenGL allows crazy
- * mixing of 2d textures with height 1 and 1d textures, so make sure
- * we pick 1d if any cbuf or zsbuf is 1d.
- */
- if (llvmpipe_resource_is_1d(lp->framebuffer.cbufs[0]->texture)) {
- key->resource_1d = TRUE;
- }
+ key->cbuf_format[i] = format;
- format_desc = util_format_description(format);
- assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_RGB ||
- format_desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB);
+ /*
+ * Figure out if this is a 1d resource. Note that OpenGL allows crazy
+ * mixing of 2d textures with height 1 and 1d textures, so make sure
+ * we pick 1d if any cbuf or zsbuf is 1d.
+ */
+ if (llvmpipe_resource_is_1d(lp->framebuffer.cbufs[i]->texture)) {
+ key->resource_1d = TRUE;
+ }
- /*
- * Mask out color channels not present in the color buffer.
- */
- blend_rt->colormask &= util_format_colormask(format_desc);
+ format_desc = util_format_description(format);
+ assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_RGB ||
+ format_desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB);
- /*
- * Disable blend for integer formats.
- */
- if (util_format_is_pure_integer(format)) {
- blend_rt->blend_enable = 0;
- }
+ /*
+ * Mask out color channels not present in the color buffer.
+ */
+ blend_rt->colormask &= util_format_colormask(format_desc);
- /*
- * Our swizzled render tiles always have an alpha channel, but the linear
- * render target format often does not, so force here the dst alpha to be
- * one.
- *
- * This is not a mere optimization. Wrong results will be produced if the
- * dst alpha is used, the dst format does not have alpha, and the previous
- * rendering was not flushed from the swizzled to linear buffer. For
- * example, NonPowTwo DCT.
- *
- * TODO: This should be generalized to all channels for better
- * performance, but only alpha causes correctness issues.
- *
- * Also, force rgb/alpha func/factors match, to make AoS blending easier.
- */
- if (format_desc->swizzle[3] > UTIL_FORMAT_SWIZZLE_W ||
- format_desc->swizzle[3] == format_desc->swizzle[0]) {
- /* Doesn't cover mixed snorm/unorm but can't render to them anyway */
- boolean clamped_zero = !util_format_is_float(format) &&
- !util_format_is_snorm(format);
- blend_rt->rgb_src_factor = force_dst_alpha_one(blend_rt->rgb_src_factor,
- clamped_zero);
- blend_rt->rgb_dst_factor = force_dst_alpha_one(blend_rt->rgb_dst_factor,
- clamped_zero);
- blend_rt->alpha_func = blend_rt->rgb_func;
- blend_rt->alpha_src_factor = blend_rt->rgb_src_factor;
- blend_rt->alpha_dst_factor = blend_rt->rgb_dst_factor;
+ /*
+ * Disable blend for integer formats.
+ */
+ if (util_format_is_pure_integer(format)) {
+ blend_rt->blend_enable = 0;
+ }
+
+ /*
+ * Our swizzled render tiles always have an alpha channel, but the
+ * linear render target format often does not, so force here the dst
+ * alpha to be one.
+ *
+ * This is not a mere optimization. Wrong results will be produced if
+ * the dst alpha is used, the dst format does not have alpha, and the
+ * previous rendering was not flushed from the swizzled to linear
+ * buffer. For example, NonPowTwo DCT.
+ *
+ * TODO: This should be generalized to all channels for better
+ * performance, but only alpha causes correctness issues.
+ *
+ * Also, force rgb/alpha func/factors match, to make AoS blending
+ * easier.
+ */
+ if (format_desc->swizzle[3] > UTIL_FORMAT_SWIZZLE_W ||
+ format_desc->swizzle[3] == format_desc->swizzle[0]) {
+ /* Doesn't cover mixed snorm/unorm but can't render to them anyway */
+ boolean clamped_zero = !util_format_is_float(format) &&
+ !util_format_is_snorm(format);
+ blend_rt->rgb_src_factor =
+ force_dst_alpha_one(blend_rt->rgb_src_factor, clamped_zero);
+ blend_rt->rgb_dst_factor =
+ force_dst_alpha_one(blend_rt->rgb_dst_factor, clamped_zero);
+ blend_rt->alpha_func = blend_rt->rgb_func;
+ blend_rt->alpha_src_factor = blend_rt->rgb_src_factor;
+ blend_rt->alpha_dst_factor = blend_rt->rgb_dst_factor;
+ }
+ }
+ else {
+ /* no color buffer for this fragment output */
+ key->cbuf_format[i] = PIPE_FORMAT_NONE;
+ blend_rt->colormask = 0x0;
+ blend_rt->blend_enable = 0;
}
}