#include "util/u_format.h"
#include "util/u_dump.h"
#include "util/u_string.h"
-#include "util/u_simple_list.h"
+#include "util/simple_list.h"
#include "util/u_dual_blend.h"
#include "os/os_time.h"
#include "pipe/p_shader_tokens.h"
for (i = 0; i < fs_type.length / 4; i++) {
unsigned j = 2 * (i % 2) + (i / 2) * 8;
- bits[4*i + 0] = LLVMConstInt(i32t, 1 << (j + 0), 0);
- bits[4*i + 1] = LLVMConstInt(i32t, 1 << (j + 1), 0);
- bits[4*i + 2] = LLVMConstInt(i32t, 1 << (j + 4), 0);
- bits[4*i + 3] = LLVMConstInt(i32t, 1 << (j + 5), 0);
+ bits[4*i + 0] = LLVMConstInt(i32t, 1ULL << (j + 0), 0);
+ bits[4*i + 1] = LLVMConstInt(i32t, 1ULL << (j + 1), 0);
+ bits[4*i + 2] = LLVMConstInt(i32t, 1ULL << (j + 4), 0);
+ bits[4*i + 3] = LLVMConstInt(i32t, 1ULL << (j + 5), 0);
}
mask = LLVMBuildAnd(builder, mask, LLVMConstVector(bits, fs_type.length), "");
{
const struct util_format_description *zs_format_desc = NULL;
const struct tgsi_token *tokens = shader->base.tokens;
- LLVMTypeRef vec_type;
+ struct lp_type int_type = lp_int_type(type);
+ LLVMTypeRef vec_type, int_vec_type;
LLVMValueRef mask_ptr, mask_val;
LLVMValueRef consts_ptr, num_consts_ptr;
LLVMValueRef z;
zs_format_desc = util_format_description(key->zsbuf_format);
assert(zs_format_desc);
- if (!shader->info.base.writes_z) {
+ if (!shader->info.base.writes_z && !shader->info.base.writes_stencil) {
if (key->alpha.enabled ||
key->blend.alpha_to_coverage ||
shader->info.base.uses_kill) {
depth_mode = 0;
}
+ vec_type = lp_build_vec_type(gallivm, type);
+ int_vec_type = lp_build_vec_type(gallivm, int_type);
stencil_refs[0] = lp_jit_context_stencil_ref_front_value(gallivm, context_ptr);
stencil_refs[1] = lp_jit_context_stencil_ref_back_value(gallivm, context_ptr);
-
- vec_type = lp_build_vec_type(gallivm, type);
+ /* convert scalar stencil refs into vectors */
+ stencil_refs[0] = lp_build_broadcast(gallivm, int_vec_type, stencil_refs[0]);
+ stencil_refs[1] = lp_build_broadcast(gallivm, int_vec_type, stencil_refs[1]);
consts_ptr = lp_jit_context_constants(gallivm, context_ptr);
num_consts_ptr = lp_jit_context_num_constants(gallivm, context_ptr);
lp_build_tgsi_soa(gallivm, tokens, type, &mask,
consts_ptr, num_consts_ptr, &system_values,
interp->inputs,
- outputs, sampler, &shader->info.base, NULL);
+ outputs, context_ptr, thread_data_ptr,
+ sampler, &shader->info.base, NULL);
/* Alpha test */
if (key->alpha.enabled) {
int pos0 = find_output_by_semantic(&shader->info.base,
TGSI_SEMANTIC_POSITION,
0);
-
+ int s_out = find_output_by_semantic(&shader->info.base,
+ TGSI_SEMANTIC_STENCIL,
+ 0);
if (pos0 != -1 && outputs[pos0][2]) {
z = LLVMBuildLoad(builder, outputs[pos0][2], "output.z");
}
}
+ if (s_out != -1 && outputs[s_out][1]) {
+ /* there's only one value, and spec says to discard additional bits */
+ LLVMValueRef s_max_mask = lp_build_const_int_vec(gallivm, int_type, 255);
+ stencil_refs[0] = LLVMBuildLoad(builder, outputs[s_out][1], "output.s");
+ stencil_refs[0] = LLVMBuildBitCast(builder, stencil_refs[0], int_vec_type, "");
+ stencil_refs[0] = LLVMBuildAnd(builder, stencil_refs[0], s_max_mask, "");
+ stencil_refs[1] = stencil_refs[0];
+ }
+
lp_build_depth_stencil_load_swizzled(gallivm, type,
zs_format_desc, key->resource_1d,
depth_ptr, depth_stride,
*
* A format which has irregular channel sizes such as R3_G3_B2 or R5_G6_B5.
*/
-static INLINE boolean
+static inline boolean
is_arithmetic_format(const struct util_format_description *format_desc)
{
boolean arith = false;
* to floats for blending, and furthermore has "natural" packed AoS -> unpacked
* SoA conversion.
*/
-static INLINE boolean
+static inline boolean
format_expands_to_float_soa(const struct util_format_description *format_desc)
{
if (format_desc->format == PIPE_FORMAT_R11G11B10_FLOAT ||
*
* e.g. RGBA16F = 4x half-float and R3G3B2 = 1x byte
*/
-static INLINE void
+static inline void
lp_mem_type_from_format_desc(const struct util_format_description *format_desc,
struct lp_type* type)
{
*
* e.g. RGBA16F = 4x float, R3G3B2 = 3x byte
*/
-static INLINE void
+static inline void
lp_blend_type_from_format_desc(const struct util_format_description *format_desc,
struct lp_type* type)
{
*
* but we try to avoid division and multiplication through shifts.
*/
-static INLINE LLVMValueRef
+static inline LLVMValueRef
scale_bits(struct gallivm_state *gallivm,
int src_bits,
int dst_bits,
/**
* If RT is a smallfloat (needing denorms) format
*/
-static INLINE int
+static inline int
have_smallfloat_format(struct lp_type dst_type,
enum pipe_format format)
{
partial_mask |= !variant->opaque;
i32_zero = lp_build_const_int32(gallivm, 0);
-#if HAVE_LLVM < 0x0302
- /*
- * undef triggers a crash in LLVMBuildTrunc in convert_from_blend_type in some
- * cases (seen with r10g10b10a2, 128bit wide vectors) (only used for 1d case).
- */
- undef_src_val = lp_build_zero(gallivm, fs_type);
-#else
undef_src_val = lp_build_undef(gallivm, fs_type);
-#endif
row_type.length = fs_type.length;
vector_width = dst_type.floating ? lp_native_vector_width : lp_integer_vector_width;
lp_build_name(dady_ptr, "dady");
lp_build_name(color_ptr_ptr, "color_ptr_ptr");
lp_build_name(depth_ptr, "depth");
- lp_build_name(thread_data_ptr, "thread_data");
lp_build_name(mask_input, "mask_input");
+ lp_build_name(thread_data_ptr, "thread_data");
lp_build_name(stride_ptr, "stride_ptr");
lp_build_name(depth_stride, "depth_stride");
LLVMPositionBuilderAtEnd(builder, block);
/* code generated texture sampling */
- sampler = lp_llvm_sampler_soa_create(key->state, context_ptr);
+ sampler = lp_llvm_sampler_soa_create(key->state);
num_fs = 16 / fs_type.length; /* number of loops per 4x4 stamp */
/* for 1d resources only run "upper half" of stamp */
char module_name[64];
variant = CALLOC_STRUCT(lp_fragment_shader_variant);
- if(!variant)
+ if (!variant)
return NULL;
util_snprintf(module_name, sizeof(module_name), "fs%u_variant%u",
/**
* Return the blend factor equivalent to a destination alpha of one.
*/
-static INLINE unsigned
+static inline unsigned
force_dst_alpha_one(unsigned factor, boolean clamped_zero)
{
switch(factor) {