* Code generate the whole fragment pipeline.
*
* The fragment pipeline consists of the following stages:
+ * - triangle edge in/out testing
+ * - scissor test
* - stipple (TBI)
* - early depth test
* - fragment shader
* - depth/stencil test (stencil TBI)
* - blending
*
- * This file has only the glue to assembly the fragment pipeline. The actual
+ * This file has only the glue to assemble the fragment pipeline. The actual
* plumbing of converting Gallium state into LLVM IR is done elsewhere, in the
* lp_bld_*.[ch] files, and in a complete generic and reusable way. Here we
* muster the LLVM JIT execution engine to create a function that follows an
#include <limits.h>
#include "pipe/p_defines.h"
+#include "util/u_inlines.h"
#include "util/u_memory.h"
#include "util/u_format.h"
-#include "util/u_debug_dump.h"
-#include "pipe/internal/p_winsys_screen.h"
+#include "util/u_dump.h"
+#include "os/os_time.h"
#include "pipe/p_shader_tokens.h"
#include "draw/draw_context.h"
#include "tgsi/tgsi_dump.h"
#include "tgsi/tgsi_scan.h"
#include "tgsi/tgsi_parse.h"
-#include "lp_bld_type.h"
-#include "lp_bld_const.h"
-#include "lp_bld_conv.h"
-#include "lp_bld_intr.h"
-#include "lp_bld_logic.h"
-#include "lp_bld_depth.h"
-#include "lp_bld_interp.h"
-#include "lp_bld_tgsi.h"
+#include "gallivm/lp_bld_type.h"
+#include "gallivm/lp_bld_const.h"
+#include "gallivm/lp_bld_conv.h"
+#include "gallivm/lp_bld_intr.h"
+#include "gallivm/lp_bld_logic.h"
+#include "gallivm/lp_bld_tgsi.h"
+#include "gallivm/lp_bld_swizzle.h"
+#include "gallivm/lp_bld_flow.h"
+#include "gallivm/lp_bld_debug.h"
+
#include "lp_bld_alpha.h"
#include "lp_bld_blend.h"
-#include "lp_bld_swizzle.h"
-#include "lp_bld_flow.h"
-#include "lp_bld_debug.h"
-#include "lp_screen.h"
+#include "lp_bld_depth.h"
+#include "lp_bld_interp.h"
#include "lp_context.h"
-#include "lp_buffer.h"
+#include "lp_debug.h"
+#include "lp_perf.h"
+#include "lp_screen.h"
#include "lp_setup.h"
#include "lp_state.h"
#include "lp_tex_sample.h"
-#include "lp_debug.h"
+
+
+#include <llvm-c/Analysis.h>
static const unsigned char quad_offset_x[4] = {0, 1, 0, 1};
/**
- * Generate the depth test.
+ * Generate the depth /stencil test code.
*/
static void
-generate_depth(LLVMBuilderRef builder,
- const struct lp_fragment_shader_variant_key *key,
- struct lp_type src_type,
- struct lp_build_mask_context *mask,
- LLVMValueRef src,
- LLVMValueRef dst_ptr)
+generate_depth_stencil(LLVMBuilderRef builder,
+ const struct lp_fragment_shader_variant_key *key,
+ struct lp_type src_type,
+ struct lp_build_mask_context *mask,
+ LLVMValueRef stencil_refs[2],
+ LLVMValueRef src,
+ LLVMValueRef dst_ptr,
+ LLVMValueRef facing)
{
const struct util_format_description *format_desc;
struct lp_type dst_type;
- if(!key->depth.enabled)
+ if (!key->depth.enabled && !key->stencil[0].enabled && !key->stencil[1].enabled)
return;
format_desc = util_format_description(key->zsbuf_format);
assert(dst_type.width == src_type.width);
assert(dst_type.length == src_type.length);
+ /* Convert fragment Z from float to integer */
lp_build_conv(builder, src_type, dst_type, &src, 1, &src, 1);
dst_ptr = LLVMBuildBitCast(builder,
dst_ptr,
LLVMPointerType(lp_build_vec_type(dst_type), 0), "");
-
- lp_build_depth_test(builder,
- &key->depth,
- dst_type,
- format_desc,
- mask,
- src,
- dst_ptr);
+ lp_build_depth_stencil_test(builder,
+ &key->depth,
+ key->stencil,
+ dst_type,
+ format_desc,
+ mask,
+ stencil_refs,
+ src,
+ dst_ptr,
+ facing);
}
LLVMConstInt(LLVMInt32Type(), INT_MIN, 0),
"");
- in_out_mask = lp_build_int_const_scalar(i32_type, ~0);
+ in_out_mask = lp_build_const_int_vec(i32_type, ~0);
lp_build_flow_scope_declare(flow, &in_out_mask);
}
+static LLVMValueRef
+build_int32_vec_const(int value)
+{
+ struct lp_type i32_type;
+
+ memset(&i32_type, 0, sizeof i32_type);
+ i32_type.floating = FALSE; /* values are integers */
+ i32_type.sign = TRUE; /* values are signed */
+ i32_type.norm = FALSE; /* values are not normalized */
+ i32_type.width = 32; /* 32-bit int values */
+ i32_type.length = 4; /* 4 elements per vector */
+ return lp_build_const_int_vec(i32_type, value);
+}
+
+
+
/**
* Generate the fragment shader, depth/stencil test, and alpha tests.
* \param i which quad in the tile, in range [0,3]
+ * \param do_tri_test if 1, do triangle edge in/out testing
*/
static void
generate_fs(struct llvmpipe_context *lp,
LLVMValueRef *pmask,
LLVMValueRef (*color)[4],
LLVMValueRef depth_ptr,
+ LLVMValueRef facing,
+ unsigned do_tri_test,
LLVMValueRef c0,
LLVMValueRef c1,
LLVMValueRef c2,
LLVMValueRef consts_ptr;
LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][NUM_CHANNELS];
LLVMValueRef z = interp->pos[2];
+ LLVMValueRef stencil_refs[2];
struct lp_build_flow_context *flow;
struct lp_build_mask_context mask;
- boolean early_depth_test;
+ boolean early_depth_stencil_test;
unsigned attrib;
unsigned chan;
unsigned cbuf;
assert(i < 4);
+ stencil_refs[0] = lp_jit_context_stencil_ref_front_value(builder, context_ptr);
+ stencil_refs[1] = lp_jit_context_stencil_ref_back_value(builder, context_ptr);
+
elem_type = lp_build_elem_type(type);
vec_type = lp_build_vec_type(type);
int_vec_type = lp_build_int_vec_type(type);
lp_build_flow_scope_declare(flow, &z);
/* do triangle edge testing */
- generate_tri_edge_mask(builder, i, pmask,
- c0, c1, c2, step0_ptr, step1_ptr, step2_ptr);
+ if (do_tri_test) {
+ generate_tri_edge_mask(builder, i, pmask,
+ c0, c1, c2, step0_ptr, step1_ptr, step2_ptr);
+ }
+ else {
+ *pmask = build_int32_vec_const(~0);
+ }
/* 'mask' will control execution based on quad's pixel alive/killed state */
lp_build_mask_begin(&mask, flow, type, *pmask);
lp_build_mask_update(&mask, smask);
}
- early_depth_test =
- key->depth.enabled &&
+ early_depth_stencil_test =
+ (key->depth.enabled || key->stencil[0].enabled) &&
!key->alpha.enabled &&
!shader->info.uses_kill &&
!shader->info.writes_z;
- if(early_depth_test)
- generate_depth(builder, key,
- type, &mask,
- z, depth_ptr);
+ if (early_depth_stencil_test)
+ generate_depth_stencil(builder, key,
+ type, &mask,
+ stencil_refs, z, depth_ptr, facing);
lp_build_tgsi_soa(builder, tokens, type, &mask,
consts_ptr, interp->pos, interp->inputs,
for (attrib = 0; attrib < shader->info.num_outputs; ++attrib) {
for(chan = 0; chan < NUM_CHANNELS; ++chan) {
if(outputs[attrib][chan]) {
- lp_build_name(outputs[attrib][chan], "output%u.%u.%c", i, attrib, "xyzw"[chan]);
+ LLVMValueRef out = LLVMBuildLoad(builder, outputs[attrib][chan], "");
+ lp_build_name(out, "output%u.%u.%c", i, attrib, "xyzw"[chan]);
switch (shader->info.output_semantic_name[attrib]) {
case TGSI_SEMANTIC_COLOR:
{
unsigned cbuf = shader->info.output_semantic_index[attrib];
- lp_build_name(outputs[attrib][chan], "color%u.%u.%c", i, attrib, "rgba"[chan]);
+ lp_build_name(out, "color%u.%u.%c", i, attrib, "rgba"[chan]);
/* Alpha test */
/* XXX: should the alpha reference value be passed separately? */
/* XXX: should only test the final assignment to alpha */
if(cbuf == 0 && chan == 3) {
- LLVMValueRef alpha = outputs[attrib][chan];
+ LLVMValueRef alpha = out;
LLVMValueRef alpha_ref_value;
alpha_ref_value = lp_jit_context_alpha_ref_value(builder, context_ptr);
alpha_ref_value = lp_build_broadcast(builder, vec_type, alpha_ref_value);
&mask, alpha, alpha_ref_value);
}
- color[cbuf][chan] = outputs[attrib][chan];
+ color[cbuf][chan] = out;
break;
}
case TGSI_SEMANTIC_POSITION:
if(chan == 2)
- z = outputs[attrib][chan];
+ z = out;
break;
}
}
}
}
- if(!early_depth_test)
- generate_depth(builder, key,
- type, &mask,
- z, depth_ptr);
+ if (!early_depth_stencil_test)
+ generate_depth_stencil(builder, key,
+ type, &mask,
+ stencil_refs, z, depth_ptr, facing);
lp_build_mask_end(&mask);
lp_build_blend_soa(builder, blend, type, src, dst, con, res);
for(chan = 0; chan < 4; ++chan) {
- if(blend->colormask & (1 << chan)) {
+ if(blend->rt[0].colormask & (1 << chan)) {
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0);
lp_build_name(res[chan], "res.%c", "rgba"[chan]);
res[chan] = lp_build_select(&bld, mask, res[chan], dst[chan]);
}
+/** casting function to avoid compiler warnings */
+static lp_jit_frag_func
+cast_voidptr_to_lp_jit_frag_func(void *p)
+{
+ union {
+ void *v;
+ lp_jit_frag_func f;
+ } tmp;
+ assert(sizeof(tmp.v) == sizeof(tmp.f));
+ tmp.v = p;
+ return tmp.f;
+}
+
+
/**
* Generate the runtime callable function for the whole fragment pipeline.
* Note that the function which we generate operates on a block of 16
static void
generate_fragment(struct llvmpipe_context *lp,
struct lp_fragment_shader *shader,
- struct lp_fragment_shader_variant *variant)
+ struct lp_fragment_shader_variant *variant,
+ unsigned do_tri_test)
{
struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
const struct lp_fragment_shader_variant_key *key = &variant->key;
LLVMTypeRef fs_int_vec_type;
LLVMTypeRef blend_vec_type;
LLVMTypeRef blend_int_vec_type;
- LLVMTypeRef arg_types[14];
+ LLVMTypeRef arg_types[15];
LLVMTypeRef func_type;
LLVMTypeRef int32_vec4_type = lp_build_int32_vec4_type();
LLVMValueRef context_ptr;
LLVMValueRef blend_mask;
LLVMValueRef blend_in_color[NUM_CHANNELS];
LLVMValueRef function;
+ LLVMValueRef facing;
unsigned num_fs;
unsigned i;
unsigned chan;
arg_types[0] = screen->context_ptr_type; /* context */
arg_types[1] = LLVMInt32Type(); /* x */
arg_types[2] = LLVMInt32Type(); /* y */
- arg_types[3] = LLVMPointerType(fs_elem_type, 0); /* a0 */
- arg_types[4] = LLVMPointerType(fs_elem_type, 0); /* dadx */
- arg_types[5] = LLVMPointerType(fs_elem_type, 0); /* dady */
- arg_types[6] = LLVMPointerType(LLVMPointerType(blend_vec_type, 0), 0); /* color */
- arg_types[7] = LLVMPointerType(fs_int_vec_type, 0); /* depth */
- arg_types[8] = LLVMInt32Type(); /* c0 */
- arg_types[9] = LLVMInt32Type(); /* c1 */
- arg_types[10] = LLVMInt32Type(); /* c2 */
+ arg_types[3] = LLVMFloatType(); /* facing */
+ arg_types[4] = LLVMPointerType(fs_elem_type, 0); /* a0 */
+ arg_types[5] = LLVMPointerType(fs_elem_type, 0); /* dadx */
+ arg_types[6] = LLVMPointerType(fs_elem_type, 0); /* dady */
+ arg_types[7] = LLVMPointerType(LLVMPointerType(blend_vec_type, 0), 0); /* color */
+ arg_types[8] = LLVMPointerType(fs_int_vec_type, 0); /* depth */
+ arg_types[9] = LLVMInt32Type(); /* c0 */
+ arg_types[10] = LLVMInt32Type(); /* c1 */
+ arg_types[11] = LLVMInt32Type(); /* c2 */
/* Note: the step arrays are built as int32[16] but we interpret
* them here as int32_vec4[4].
*/
- arg_types[11] = LLVMPointerType(int32_vec4_type, 0);/* step0 */
- arg_types[12] = LLVMPointerType(int32_vec4_type, 0);/* step1 */
- arg_types[13] = LLVMPointerType(int32_vec4_type, 0);/* step2 */
+ arg_types[12] = LLVMPointerType(int32_vec4_type, 0);/* step0 */
+ arg_types[13] = LLVMPointerType(int32_vec4_type, 0);/* step1 */
+ arg_types[14] = LLVMPointerType(int32_vec4_type, 0);/* step2 */
func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
function = LLVMAddFunction(screen->module, "shader", func_type);
LLVMSetFunctionCallConv(function, LLVMCCallConv);
- variant->function = function;
+ variant->function[do_tri_test] = function;
/* XXX: need to propagate noalias down into color param now we are
context_ptr = LLVMGetParam(function, 0);
x = LLVMGetParam(function, 1);
y = LLVMGetParam(function, 2);
- a0_ptr = LLVMGetParam(function, 3);
- dadx_ptr = LLVMGetParam(function, 4);
- dady_ptr = LLVMGetParam(function, 5);
- color_ptr_ptr = LLVMGetParam(function, 6);
- depth_ptr = LLVMGetParam(function, 7);
- c0 = LLVMGetParam(function, 8);
- c1 = LLVMGetParam(function, 9);
- c2 = LLVMGetParam(function, 10);
- step0_ptr = LLVMGetParam(function, 11);
- step1_ptr = LLVMGetParam(function, 12);
- step2_ptr = LLVMGetParam(function, 13);
+ facing = LLVMGetParam(function, 3);
+ a0_ptr = LLVMGetParam(function, 4);
+ dadx_ptr = LLVMGetParam(function, 5);
+ dady_ptr = LLVMGetParam(function, 6);
+ color_ptr_ptr = LLVMGetParam(function, 7);
+ depth_ptr = LLVMGetParam(function, 8);
+ c0 = LLVMGetParam(function, 9);
+ c1 = LLVMGetParam(function, 10);
+ c2 = LLVMGetParam(function, 11);
+ step0_ptr = LLVMGetParam(function, 12);
+ step1_ptr = LLVMGetParam(function, 13);
+ step2_ptr = LLVMGetParam(function, 14);
lp_build_name(context_ptr, "context");
lp_build_name(x, "x");
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
LLVMValueRef out_color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS];
LLVMValueRef depth_ptr_i;
- int cbuf;
if(i != 0)
lp_build_interp_soa_update(&interp, i);
&fs_mask[i], /* output */
out_color,
depth_ptr_i,
+ facing,
+ do_tri_test,
c0, c1, c2,
step0_ptr, step1_ptr, step2_ptr);
/*
* Translate the LLVM IR into machine code.
*/
- variant->jit_function = (lp_jit_frag_func)LLVMGetPointerToGlobal(screen->engine, function);
+ {
+ void *f = LLVMGetPointerToGlobal(screen->engine, function);
+
+ variant->jit_function[do_tri_test] = cast_voidptr_to_lp_jit_frag_func(f);
- if (LP_DEBUG & DEBUG_ASM)
- lp_disassemble(variant->jit_function);
+ if (LP_DEBUG & DEBUG_ASM)
+ lp_disassemble(f);
+ }
}
tgsi_dump(shader->base.tokens, 0);
if(key->depth.enabled) {
- debug_printf("depth.format = %s\n", pf_name(key->zsbuf_format));
- debug_printf("depth.func = %s\n", debug_dump_func(key->depth.func, TRUE));
+ debug_printf("depth.format = %s\n", util_format_name(key->zsbuf_format));
+ debug_printf("depth.func = %s\n", util_dump_func(key->depth.func, TRUE));
debug_printf("depth.writemask = %u\n", key->depth.writemask);
}
if(key->alpha.enabled) {
- debug_printf("alpha.func = %s\n", debug_dump_func(key->alpha.func, TRUE));
+ debug_printf("alpha.func = %s\n", util_dump_func(key->alpha.func, TRUE));
debug_printf("alpha.ref_value = %f\n", key->alpha.ref_value);
}
if(key->blend.logicop_enable) {
debug_printf("blend.logicop_func = %u\n", key->blend.logicop_func);
}
- else if(key->blend.blend_enable) {
- debug_printf("blend.rgb_func = %s\n", debug_dump_blend_func (key->blend.rgb_func, TRUE));
- debug_printf("rgb_src_factor = %s\n", debug_dump_blend_factor(key->blend.rgb_src_factor, TRUE));
- debug_printf("rgb_dst_factor = %s\n", debug_dump_blend_factor(key->blend.rgb_dst_factor, TRUE));
- debug_printf("alpha_func = %s\n", debug_dump_blend_func (key->blend.alpha_func, TRUE));
- debug_printf("alpha_src_factor = %s\n", debug_dump_blend_factor(key->blend.alpha_src_factor, TRUE));
- debug_printf("alpha_dst_factor = %s\n", debug_dump_blend_factor(key->blend.alpha_dst_factor, TRUE));
+ else if(key->blend.rt[0].blend_enable) {
+ debug_printf("blend.rgb_func = %s\n", util_dump_blend_func (key->blend.rt[0].rgb_func, TRUE));
+ debug_printf("rgb_src_factor = %s\n", util_dump_blend_factor(key->blend.rt[0].rgb_src_factor, TRUE));
+ debug_printf("rgb_dst_factor = %s\n", util_dump_blend_factor(key->blend.rt[0].rgb_dst_factor, TRUE));
+ debug_printf("alpha_func = %s\n", util_dump_blend_func (key->blend.rt[0].alpha_func, TRUE));
+ debug_printf("alpha_src_factor = %s\n", util_dump_blend_factor(key->blend.rt[0].alpha_src_factor, TRUE));
+ debug_printf("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.colormask);
+ debug_printf("blend.colormask = 0x%x\n", key->blend.rt[0].colormask);
for(i = 0; i < PIPE_MAX_SAMPLERS; ++i) {
if(key->sampler[i].format) {
debug_printf("sampler[%u] = \n", i);
debug_printf(" .format = %s\n",
- pf_name(key->sampler[i].format));
+ util_format_name(key->sampler[i].format));
debug_printf(" .target = %s\n",
- debug_dump_tex_target(key->sampler[i].target, TRUE));
+ util_dump_tex_target(key->sampler[i].target, TRUE));
debug_printf(" .pot = %u %u %u\n",
key->sampler[i].pot_width,
key->sampler[i].pot_height,
key->sampler[i].pot_depth);
debug_printf(" .wrap = %s %s %s\n",
- debug_dump_tex_wrap(key->sampler[i].wrap_s, TRUE),
- debug_dump_tex_wrap(key->sampler[i].wrap_t, TRUE),
- debug_dump_tex_wrap(key->sampler[i].wrap_r, TRUE));
+ util_dump_tex_wrap(key->sampler[i].wrap_s, TRUE),
+ util_dump_tex_wrap(key->sampler[i].wrap_t, TRUE),
+ util_dump_tex_wrap(key->sampler[i].wrap_r, TRUE));
debug_printf(" .min_img_filter = %s\n",
- debug_dump_tex_filter(key->sampler[i].min_img_filter, TRUE));
+ util_dump_tex_filter(key->sampler[i].min_img_filter, TRUE));
debug_printf(" .min_mip_filter = %s\n",
- debug_dump_tex_mipfilter(key->sampler[i].min_mip_filter, TRUE));
+ util_dump_tex_mipfilter(key->sampler[i].min_mip_filter, TRUE));
debug_printf(" .mag_img_filter = %s\n",
- debug_dump_tex_filter(key->sampler[i].mag_img_filter, TRUE));
+ util_dump_tex_filter(key->sampler[i].mag_img_filter, TRUE));
if(key->sampler[i].compare_mode != PIPE_TEX_COMPARE_NONE)
- debug_printf(" .compare_func = %s\n", debug_dump_func(key->sampler[i].compare_func, TRUE));
+ debug_printf(" .compare_func = %s\n", util_dump_func(key->sampler[i].compare_func, TRUE));
debug_printf(" .normalized_coords = %u\n", key->sampler[i].normalized_coords);
- debug_printf(" .prefilter = %u\n", key->sampler[i].prefilter);
}
}
}
variant->shader = shader;
memcpy(&variant->key, key, sizeof *key);
- generate_fragment(lp, shader, variant);
+ generate_fragment(lp, shader, variant, 0);
+ generate_fragment(lp, shader, variant, 1);
/* insert new variant into linked list */
variant->next = shader->variants;
assert(fs != llvmpipe->fs);
(void) llvmpipe;
+ /*
+ * XXX: we need to flush the context until we have some sort of reference
+ * counting in fragment shaders as they may still be binned
+ */
+ draw_flush(llvmpipe->draw);
+ lp_setup_flush(llvmpipe->setup, 0);
+
variant = shader->variants;
while(variant) {
struct lp_fragment_shader_variant *next = variant->next;
+ unsigned i;
- if(variant->function) {
- if(variant->jit_function)
- LLVMFreeMachineCodeForFunction(screen->engine, variant->function);
- LLVMDeleteFunction(variant->function);
+ for (i = 0; i < Elements(variant->function); i++) {
+ if (variant->function[i]) {
+ if (variant->jit_function[i])
+ LLVMFreeMachineCodeForFunction(screen->engine,
+ variant->function[i]);
+ LLVMDeleteFunction(variant->function[i]);
+ }
}
FREE(variant);
void
llvmpipe_set_constant_buffer(struct pipe_context *pipe,
uint shader, uint index,
- const struct pipe_constant_buffer *constants)
+ struct pipe_resource *constants)
{
struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
- struct pipe_buffer *buffer = constants ? constants->buffer : NULL;
- unsigned size = buffer ? buffer->size : 0;
- const void *data = buffer ? llvmpipe_buffer(buffer)->data : NULL;
+ unsigned size = constants ? constants->width0 : 0;
+ const void *data = constants ? llvmpipe_resource_data(constants) : NULL;
assert(shader < PIPE_SHADER_TYPES);
assert(index == 0);
- if(llvmpipe->constants[shader].buffer == buffer)
+ if(llvmpipe->constants[shader] == constants)
return;
draw_flush(llvmpipe->draw);
/* note: reference counting */
- pipe_buffer_reference(&llvmpipe->constants[shader].buffer, buffer);
+ pipe_resource_reference(&llvmpipe->constants[shader], constants);
if(shader == PIPE_SHADER_VERTEX) {
- draw_set_mapped_constant_buffer(llvmpipe->draw, PIPE_SHADER_VERTEX,
+ draw_set_mapped_constant_buffer(llvmpipe->draw, PIPE_SHADER_VERTEX, 0,
data, size);
}
memset(key, 0, sizeof *key);
- if(lp->framebuffer.zsbuf &&
- lp->depth_stencil->depth.enabled) {
- key->zsbuf_format = lp->framebuffer.zsbuf->format;
- memcpy(&key->depth, &lp->depth_stencil->depth, sizeof key->depth);
+ if (lp->framebuffer.zsbuf) {
+ if (lp->depth_stencil->depth.enabled) {
+ key->zsbuf_format = lp->framebuffer.zsbuf->format;
+ memcpy(&key->depth, &lp->depth_stencil->depth, sizeof key->depth);
+ }
+ if (lp->depth_stencil->stencil[0].enabled) {
+ key->zsbuf_format = lp->framebuffer.zsbuf->format;
+ memcpy(&key->stencil, &lp->depth_stencil->stencil, sizeof key->stencil);
+ }
}
key->alpha.enabled = lp->depth_stencil->alpha.enabled;
enum util_format_swizzle swizzle = format_desc->swizzle[chan];
if(swizzle <= UTIL_FORMAT_SWIZZLE_W)
- key->cbuf_blend[i].colormask |= (1 << chan);
+ key->blend.rt[0].colormask |= (1 << chan);
}
}
for(i = 0; i < PIPE_MAX_SAMPLERS; ++i)
if(shader->info.file_mask[TGSI_FILE_SAMPLER] & (1 << i))
- lp_sampler_static_state(&key->sampler[i], lp->texture[i], lp->sampler[i]);
+ lp_sampler_static_state(&key->sampler[i], lp->fragment_sampler_views[i], lp->sampler[i]);
}
variant = variant->next;
}
- if(!variant)
+ if (!variant) {
+ int64_t t0, t1;
+ int64_t dt;
+ t0 = os_time_get();
+
variant = generate_variant(lp, shader, &key);
+ t1 = os_time_get();
+ dt = t1 - t0;
+ LP_COUNT_ADD(llvm_compile_time, dt);
+ LP_COUNT_ADD(nr_llvm_compiles, 2); /* emit vs. omit in/out test */
+ }
+
shader->current = variant;
/* TODO: put this in the variant */
/* TODO: most of these can be relaxed, in particular the colormask */
opaque = !key.blend.logicop_enable &&
- !key.blend.blend_enable &&
- key.blend.colormask == 0xf &&
+ !key.blend.rt[0].blend_enable &&
+ key.blend.rt[0].colormask == 0xf &&
+ !key.stencil[0].enabled &&
!key.alpha.enabled &&
!key.depth.enabled &&
!key.scissor &&
!shader->info.uses_kill
? TRUE : FALSE;
- lp_setup_set_fs_function(lp->setup,
- shader->current->jit_function,
- opaque);
+ lp_setup_set_fs_functions(lp->setup,
+ shader->current->jit_function[RAST_WHOLE],
+ shader->current->jit_function[RAST_EDGE_TEST],
+ opaque);
}