#include "draw_gs.h"
#include "gallivm/lp_bld_arit.h"
+#include "gallivm/lp_bld_arit_overflow.h"
#include "gallivm/lp_bld_logic.h"
#include "gallivm/lp_bld_const.h"
#include "gallivm/lp_bld_swizzle.h"
#include "util/u_math.h"
#include "util/u_pointer.h"
#include "util/u_string.h"
-#include "util/u_simple_list.h"
+#include "util/simple_list.h"
#define DEBUG_STORE 0
LLVMValueRef input;
};
-static INLINE const struct draw_gs_llvm_iface *
+static inline const struct draw_gs_llvm_iface *
draw_gs_llvm_iface(const struct lp_build_tgsi_gs_iface *iface)
{
return (const struct draw_gs_llvm_iface *)iface;
}
+/**
+ * Create LLVM type for draw_vertex_buffer.
+ */
+static LLVMTypeRef
+create_jit_dvbuffer_type(struct gallivm_state *gallivm,
+ const char *struct_name)
+{
+ LLVMTargetDataRef target = gallivm->target;
+ LLVMTypeRef dvbuffer_type;
+ LLVMTypeRef elem_types[DRAW_JIT_DVBUFFER_NUM_FIELDS];
+ LLVMTypeRef int32_type = LLVMInt32TypeInContext(gallivm->context);
+
+ elem_types[DRAW_JIT_DVBUFFER_MAP] =
+ LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 8), 0);
+ elem_types[DRAW_JIT_DVBUFFER_SIZE] = int32_type;
+
+ dvbuffer_type = LLVMStructTypeInContext(gallivm->context, elem_types,
+ Elements(elem_types), 0);
+
+ (void) target; /* silence unused var warning for non-debug build */
+ LP_CHECK_MEMBER_OFFSET(struct draw_vertex_buffer, map,
+ target, dvbuffer_type,
+ DRAW_JIT_DVBUFFER_MAP);
+ LP_CHECK_MEMBER_OFFSET(struct draw_vertex_buffer, size,
+ target, dvbuffer_type,
+ DRAW_JIT_DVBUFFER_SIZE);
+
+ return dvbuffer_type;
+}
+
/**
* Create LLVM type for struct draw_jit_texture
*/
texture_type = LLVMStructTypeInContext(gallivm->context, elem_types,
Elements(elem_types), 0);
-#if HAVE_LLVM < 0x0300
- LLVMAddTypeName(gallivm->module, struct_name, texture_type);
-
- /* Make sure the target's struct layout cache doesn't return
- * stale/invalid data.
- */
- LLVMInvalidateStructLayout(gallivm->target, texture_type);
-#endif
-
+ (void) target; /* silence unused var warning for non-debug build */
LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, width,
target, texture_type,
DRAW_JIT_TEXTURE_WIDTH);
sampler_type = LLVMStructTypeInContext(gallivm->context, elem_types,
Elements(elem_types), 0);
-#if HAVE_LLVM < 0x0300
- LLVMAddTypeName(gallivm->module, struct_name, sampler_type);
-
- /* Make sure the target's struct layout cache doesn't return
- * stale/invalid data.
- */
- LLVMInvalidateStructLayout(gallivm->target, sampler_type);
-#endif
-
+ (void) target; /* silence unused var warning for non-debug build */
LP_CHECK_MEMBER_OFFSET(struct draw_jit_sampler, min_lod,
target, sampler_type,
DRAW_JIT_SAMPLER_MIN_LOD);
{
LLVMTargetDataRef target = gallivm->target;
LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
+ LLVMTypeRef int_type = LLVMInt32TypeInContext(gallivm->context);
LLVMTypeRef elem_types[DRAW_JIT_CTX_NUM_FIELDS];
LLVMTypeRef context_type;
elem_types[0] = LLVMArrayType(LLVMPointerType(float_type, 0), /* vs_constants */
LP_MAX_TGSI_CONST_BUFFERS);
- elem_types[1] = LLVMPointerType(LLVMArrayType(LLVMArrayType(float_type, 4),
+ elem_types[1] = LLVMArrayType(int_type, /* num_vs_constants */
+ LP_MAX_TGSI_CONST_BUFFERS);
+ elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(float_type, 4),
DRAW_TOTAL_CLIP_PLANES), 0);
- elem_types[2] = LLVMPointerType(float_type, 0); /* viewport */
- elem_types[3] = LLVMArrayType(texture_type,
+ elem_types[3] = LLVMPointerType(float_type, 0); /* viewports */
+ elem_types[4] = LLVMArrayType(texture_type,
PIPE_MAX_SHADER_SAMPLER_VIEWS); /* textures */
- elem_types[4] = LLVMArrayType(sampler_type,
+ elem_types[5] = LLVMArrayType(sampler_type,
PIPE_MAX_SAMPLERS); /* samplers */
context_type = LLVMStructTypeInContext(gallivm->context, elem_types,
Elements(elem_types), 0);
-#if HAVE_LLVM < 0x0300
- LLVMAddTypeName(gallivm->module, struct_name, context_type);
-
- LLVMInvalidateStructLayout(gallivm->target, context_type);
-#endif
+ (void) target; /* silence unused var warning for non-debug build */
LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, vs_constants,
target, context_type, DRAW_JIT_CTX_CONSTANTS);
+ LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, num_vs_constants,
+ target, context_type, DRAW_JIT_CTX_NUM_CONSTANTS);
LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, planes,
target, context_type, DRAW_JIT_CTX_PLANES);
- LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, viewport,
+ LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, viewports,
target, context_type, DRAW_JIT_CTX_VIEWPORT);
LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures,
target, context_type,
elem_types[0] = LLVMArrayType(LLVMPointerType(float_type, 0), /* constants */
LP_MAX_TGSI_CONST_BUFFERS);
- elem_types[1] = LLVMPointerType(LLVMArrayType(LLVMArrayType(float_type, 4),
+ elem_types[1] = LLVMArrayType(int_type, /* num_constants */
+ LP_MAX_TGSI_CONST_BUFFERS);
+ elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(float_type, 4),
DRAW_TOTAL_CLIP_PLANES), 0);
- elem_types[2] = LLVMPointerType(float_type, 0); /* viewport */
+ elem_types[3] = LLVMPointerType(float_type, 0); /* viewports */
- elem_types[3] = LLVMArrayType(texture_type,
+ elem_types[4] = LLVMArrayType(texture_type,
PIPE_MAX_SHADER_SAMPLER_VIEWS); /* textures */
- elem_types[4] = LLVMArrayType(sampler_type,
+ elem_types[5] = LLVMArrayType(sampler_type,
PIPE_MAX_SAMPLERS); /* samplers */
- elem_types[5] = LLVMPointerType(LLVMPointerType(int_type, 0), 0);
- elem_types[6] = LLVMPointerType(LLVMVectorType(int_type,
- vector_length), 0);
+ elem_types[6] = LLVMPointerType(LLVMPointerType(int_type, 0), 0);
elem_types[7] = LLVMPointerType(LLVMVectorType(int_type,
vector_length), 0);
+ elem_types[8] = LLVMPointerType(LLVMVectorType(int_type,
+ vector_length), 0);
context_type = LLVMStructTypeInContext(gallivm->context, elem_types,
Elements(elem_types), 0);
-#if HAVE_LLVM < 0x0300
- LLVMAddTypeName(gallivm->module, struct_name, context_type);
-
- LLVMInvalidateStructLayout(gallivm->target, context_type);
-#endif
+ (void) target; /* silence unused var warning for non-debug build */
LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, constants,
target, context_type, DRAW_GS_JIT_CTX_CONSTANTS);
+ LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, num_constants,
+ target, context_type, DRAW_GS_JIT_CTX_NUM_CONSTANTS);
LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, planes,
target, context_type, DRAW_GS_JIT_CTX_PLANES);
- LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, viewport,
+ LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, viewports,
target, context_type, DRAW_GS_JIT_CTX_VIEWPORT);
LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, textures,
target, context_type,
* Create LLVM type for struct pipe_vertex_buffer
*/
static LLVMTypeRef
-create_jit_vertex_buffer_type(struct gallivm_state *gallivm, const char *struct_name)
+create_jit_vertex_buffer_type(struct gallivm_state *gallivm,
+ const char *struct_name)
{
LLVMTargetDataRef target = gallivm->target;
LLVMTypeRef elem_types[4];
elem_types[0] =
elem_types[1] = LLVMInt32TypeInContext(gallivm->context);
elem_types[2] =
- elem_types[3] = LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0); /* vs_constants */
+ elem_types[3] = LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0);
vb_type = LLVMStructTypeInContext(gallivm->context, elem_types,
Elements(elem_types), 0);
-#if HAVE_LLVM < 0x0300
- LLVMAddTypeName(gallivm->module, struct_name, vb_type);
-
- LLVMInvalidateStructLayout(gallivm->target, vb_type);
-#endif
+ (void) target; /* silence unused var warning for non-debug build */
LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, stride,
target, vb_type, 0);
LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, buffer_offset,
create_jit_vertex_header(struct gallivm_state *gallivm, int data_elems)
{
LLVMTargetDataRef target = gallivm->target;
- LLVMTypeRef elem_types[4];
+ LLVMTypeRef elem_types[3];
LLVMTypeRef vertex_header;
char struct_name[24];
util_snprintf(struct_name, 23, "vertex_header%d", data_elems);
elem_types[DRAW_JIT_VERTEX_VERTEX_ID] = LLVMIntTypeInContext(gallivm->context, 32);
- elem_types[DRAW_JIT_VERTEX_CLIP] = LLVMArrayType(LLVMFloatTypeInContext(gallivm->context), 4);
- elem_types[DRAW_JIT_VERTEX_PRE_CLIP_POS] = LLVMArrayType(LLVMFloatTypeInContext(gallivm->context), 4);
+ elem_types[DRAW_JIT_VERTEX_CLIP_POS] = LLVMArrayType(LLVMFloatTypeInContext(gallivm->context), 4);
elem_types[DRAW_JIT_VERTEX_DATA] = LLVMArrayType(elem_types[1], data_elems);
vertex_header = LLVMStructTypeInContext(gallivm->context, elem_types,
Elements(elem_types), 0);
-#if HAVE_LLVM < 0x0300
- LLVMAddTypeName(gallivm->module, struct_name, vertex_header);
-
- LLVMInvalidateStructLayout(gallivm->target, vertex_header);
-#endif
/* these are bit-fields and we can't take address of them
LP_CHECK_MEMBER_OFFSET(struct vertex_header, clipmask,
target, vertex_header,
DRAW_JIT_VERTEX_VERTEX_ID);
*/
- LP_CHECK_MEMBER_OFFSET(struct vertex_header, clip,
+ (void) target; /* silence unused var warning for non-debug build */
+ LP_CHECK_MEMBER_OFFSET(struct vertex_header, clip_pos,
target, vertex_header,
- DRAW_JIT_VERTEX_CLIP);
- LP_CHECK_MEMBER_OFFSET(struct vertex_header, pre_clip_pos,
- target, vertex_header,
- DRAW_JIT_VERTEX_PRE_CLIP_POS);
+ DRAW_JIT_VERTEX_CLIP_POS);
LP_CHECK_MEMBER_OFFSET(struct vertex_header, data,
target, vertex_header,
DRAW_JIT_VERTEX_DATA);
create_jit_types(struct draw_llvm_variant *variant)
{
struct gallivm_state *gallivm = variant->gallivm;
- LLVMTypeRef texture_type, sampler_type, context_type, buffer_type, vb_type;
+ LLVMTypeRef texture_type, sampler_type, context_type, buffer_type,
+ vb_type;
texture_type = create_jit_texture_type(gallivm, "texture");
sampler_type = create_jit_sampler_type(gallivm, "sampler");
"draw_jit_context");
variant->context_ptr_type = LLVMPointerType(context_type, 0);
- buffer_type = LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 8), 0);
+ buffer_type = create_jit_dvbuffer_type(gallivm, "draw_vertex_buffer");
variant->buffer_ptr_type = LLVMPointerType(buffer_type, 0);
-
+
vb_type = create_jit_vertex_buffer_type(gallivm, "pipe_vertex_buffer");
variant->vb_ptr_type = LLVMPointerType(vb_type, 0);
}
* Create per-context LLVM info.
*/
struct draw_llvm *
-draw_llvm_create(struct draw_context *draw)
+draw_llvm_create(struct draw_context *draw, LLVMContextRef context)
{
struct draw_llvm *llvm;
+ if (!lp_build_init())
+ return NULL;
+
llvm = CALLOC_STRUCT( draw_llvm );
if (!llvm)
return NULL;
- lp_build_init();
-
llvm->draw = draw;
+ llvm->context = context;
+ if (!llvm->context) {
+ llvm->context = LLVMContextCreate();
+ llvm->context_owned = true;
+ }
+ if (!llvm->context)
+ goto fail;
+
llvm->nr_variants = 0;
make_empty_list(&llvm->vs_variants_list);
make_empty_list(&llvm->gs_variants_list);
return llvm;
+
+fail:
+ draw_llvm_destroy(llvm);
+ return NULL;
}
void
draw_llvm_destroy(struct draw_llvm *llvm)
{
+ if (llvm->context_owned)
+ LLVMContextDispose(llvm->context);
+ llvm->context = NULL;
+
/* XXX free other draw_llvm data? */
FREE(llvm);
}
struct llvm_vertex_shader *shader =
llvm_vertex_shader(llvm->draw->vs.vertex_shader);
LLVMTypeRef vertex_header;
+ char module_name[64];
variant = MALLOC(sizeof *variant +
shader->variant_key_size -
sizeof variant->key);
- if (variant == NULL)
+ if (!variant)
return NULL;
variant->llvm = llvm;
+ variant->shader = shader;
- variant->gallivm = gallivm_create();
+ util_snprintf(module_name, sizeof(module_name), "draw_llvm_vs_variant%u",
+ variant->shader->variants_cached);
+
+ variant->gallivm = gallivm_create(module_name, llvm->context);
create_jit_types(variant);
memcpy(&variant->key, key, shader->variant_key_size);
+ if (gallivm_debug & (GALLIVM_DEBUG_TGSI | GALLIVM_DEBUG_IR)) {
+ tgsi_dump(llvm->draw->vs.vertex_shader->state.tokens, 0);
+ draw_llvm_dump_variant_key(&variant->key);
+ }
+
vertex_header = create_jit_vertex_header(variant->gallivm, num_inputs);
variant->vertex_header_ptr_type = LLVMPointerType(vertex_header, 0);
variant->jit_func_elts = (draw_jit_vert_func_elts)
gallivm_jit_function(variant->gallivm, variant->function_elts);
- variant->shader = shader;
+ gallivm_free_ir(variant->gallivm);
+
variant->list_item_global.base = variant;
variant->list_item_local.base = variant;
/*variant->no = */shader->variants_created++;
{
struct draw_llvm *llvm = variant->llvm;
const struct tgsi_token *tokens = llvm->draw->vs.vertex_shader->state.tokens;
- LLVMValueRef consts_ptr = draw_jit_context_vs_constants(variant->gallivm, context_ptr);
- struct lp_build_sampler_soa *sampler = 0;
-
- if (gallivm_debug & (GALLIVM_DEBUG_TGSI | GALLIVM_DEBUG_IR)) {
- tgsi_dump(tokens, 0);
- draw_llvm_dump_variant_key(&variant->key);
- }
-
- if (llvm->draw->num_sampler_views && llvm->draw->num_samplers)
- sampler = draw_sampler;
+ LLVMValueRef consts_ptr =
+ draw_jit_context_vs_constants(variant->gallivm, context_ptr);
+ LLVMValueRef num_consts_ptr =
+ draw_jit_context_num_vs_constants(variant->gallivm, context_ptr);
lp_build_tgsi_soa(variant->gallivm,
tokens,
vs_type,
NULL /*struct lp_build_mask_context *mask*/,
consts_ptr,
+ num_consts_ptr,
system_values,
- NULL /*pos*/,
inputs,
outputs,
- sampler,
+ context_ptr,
+ NULL,
+ draw_sampler,
&llvm->draw->vs.vertex_shader->info,
NULL);
LLVMBuildStore(builder, out, outputs[attrib][chan]);
}
break;
- case TGSI_SEMANTIC_FOG:
- if (chan == 1 || chan == 2)
- LLVMBuildStore(builder, bld.zero, outputs[attrib][chan]);
- else if (chan == 3)
- LLVMBuildStore(builder, bld.one, outputs[attrib][chan]);
- break;
}
}
}
}
}
-
static void
generate_fetch(struct gallivm_state *gallivm,
+ struct draw_context *draw,
LLVMValueRef vbuffers_ptr,
LLVMValueRef *res,
struct pipe_vertex_element *velem,
LLVMValueRef vbuf,
LLVMValueRef index,
- LLVMValueRef instance_id)
+ LLVMValueRef instance_id,
+ LLVMValueRef start_instance)
{
- const struct util_format_description *format_desc = util_format_description(velem->src_format);
+ const struct util_format_description *format_desc =
+ util_format_description(velem->src_format);
LLVMValueRef zero = LLVMConstNull(LLVMInt32TypeInContext(gallivm->context));
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef indices =
&indices, 1, "");
LLVMValueRef vb_stride = draw_jit_vbuffer_stride(gallivm, vbuf);
LLVMValueRef vb_buffer_offset = draw_jit_vbuffer_offset(gallivm, vbuf);
+ LLVMValueRef map_ptr = draw_jit_dvbuffer_map(gallivm, vbuffer_ptr);
+ LLVMValueRef buffer_size = draw_jit_dvbuffer_size(gallivm, vbuffer_ptr);
LLVMValueRef stride;
+ LLVMValueRef buffer_overflowed;
+ LLVMValueRef needed_buffer_size;
+ LLVMValueRef temp_ptr =
+ lp_build_alloca(gallivm,
+ lp_build_vec_type(gallivm, lp_float32_vec4_type()), "");
+ LLVMValueRef ofbit = NULL;
+ struct lp_build_if_state if_ctx;
if (velem->instance_divisor) {
- /* array index = instance_id / instance_divisor */
- index = LLVMBuildUDiv(builder, instance_id,
- lp_build_const_int32(gallivm, velem->instance_divisor),
- "instance_divisor");
+ /* Index is equal to the start instance plus the number of current
+ * instance divided by the divisor. In this case we compute it as:
+ * index = start_instance + (instance_id / divisor)
+ */
+ LLVMValueRef current_instance;
+ current_instance = LLVMBuildUDiv(builder, instance_id,
+ lp_build_const_int32(gallivm, velem->instance_divisor),
+ "instance_divisor");
+ index = lp_build_uadd_overflow(gallivm, start_instance,
+ current_instance, &ofbit);
}
- stride = LLVMBuildMul(builder, vb_stride, index, "");
-
- vbuffer_ptr = LLVMBuildLoad(builder, vbuffer_ptr, "vbuffer");
-
- stride = LLVMBuildAdd(builder, stride,
- vb_buffer_offset,
- "");
- stride = LLVMBuildAdd(builder, stride,
- lp_build_const_int32(gallivm, velem->src_offset),
- "");
+ stride = lp_build_umul_overflow(gallivm, vb_stride, index, &ofbit);
+ stride = lp_build_uadd_overflow(gallivm, stride, vb_buffer_offset, &ofbit);
+ stride = lp_build_uadd_overflow(
+ gallivm, stride,
+ lp_build_const_int32(gallivm, velem->src_offset), &ofbit);
+ needed_buffer_size = lp_build_uadd_overflow(
+ gallivm, stride,
+ lp_build_const_int32(gallivm,
+ util_format_get_blocksize(velem->src_format)),
+ &ofbit);
+
+ buffer_overflowed = LLVMBuildICmp(builder, LLVMIntUGT,
+ needed_buffer_size, buffer_size,
+ "buffer_overflowed");
+ buffer_overflowed = LLVMBuildOr(builder, buffer_overflowed, ofbit, "");
+#if 0
+ lp_build_printf(gallivm, "vbuf index = %u, vb_stride is %u\n",
+ index, vb_stride);
+ lp_build_printf(gallivm, " vb_buffer_offset = %u, src_offset is %u\n",
+ vb_buffer_offset,
+ lp_build_const_int32(gallivm, velem->src_offset));
+ lp_build_print_value(gallivm, " blocksize = ",
+ lp_build_const_int32(
+ gallivm,
+ util_format_get_blocksize(velem->src_format)));
+ lp_build_printf(gallivm, " instance_id = %u\n", instance_id);
+ lp_build_printf(gallivm, " stride = %u\n", stride);
+ lp_build_printf(gallivm, " buffer size = %u\n", buffer_size);
+ lp_build_printf(gallivm, " needed_buffer_size = %u\n", needed_buffer_size);
+ lp_build_print_value(gallivm, " buffer overflowed = ", buffer_overflowed);
+#endif
-/* lp_build_printf(gallivm, "vbuf index = %d, stride is %d\n", indices, stride);*/
- vbuffer_ptr = LLVMBuildGEP(builder, vbuffer_ptr, &stride, 1, "");
+ lp_build_if(&if_ctx, gallivm, buffer_overflowed);
+ {
+ LLVMValueRef val =
+ lp_build_const_vec(gallivm, lp_float32_vec4_type(), 0);
+ LLVMBuildStore(builder, val, temp_ptr);
+ }
+ lp_build_else(&if_ctx);
+ {
+ LLVMValueRef val;
+ map_ptr = LLVMBuildGEP(builder, map_ptr, &stride, 1, "");
+
+ val = lp_build_fetch_rgba_aos(gallivm,
+ format_desc,
+ lp_float32_vec4_type(),
+ FALSE,
+ map_ptr,
+ zero, zero, zero,
+ NULL);
+ LLVMBuildStore(builder, val, temp_ptr);
+ }
+ lp_build_endif(&if_ctx);
- *res = lp_build_fetch_rgba_aos(gallivm,
- format_desc,
- lp_float32_vec4_type(),
- vbuffer_ptr,
- zero, zero, zero);
+ *res = LLVMBuildLoad(builder, temp_ptr, "aos");
}
static void
#endif
/* Unaligned store due to the vertex header */
- lp_set_store_alignment(LLVMBuildStore(builder, value, data_ptr), sizeof(float));
+ LLVMSetAlignment(LLVMBuildStore(builder, value, data_ptr), sizeof(float));
}
/**
* struct vertex_header {
* unsigned clipmask:DRAW_TOTAL_CLIP_PLANES;
* unsigned edgeflag:1;
- * unsigned have_clipdist:1;
+ * unsigned pad:1;
* unsigned vertex_id:16;
* [...]
* }
* {
* return (x >> 16) | // vertex_id
* ((x & 0x3fff) << 18) | // clipmask
- * ((x & 0x4000) << 3) | // have_clipdist
+ * ((x & 0x4000) << 3) | // pad
* ((x & 0x8000) << 1); // edgeflag
* }
*/
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef vertex_id;
LLVMValueRef clipmask;
- LLVMValueRef have_clipdist;
+ LLVMValueRef pad;
LLVMValueRef edgeflag;
vertex_id = LLVMBuildLShr(builder, mask, lp_build_const_int32(gallivm, 16), "");
clipmask = LLVMBuildAnd(builder, mask, lp_build_const_int32(gallivm, 0x3fff), "");
clipmask = LLVMBuildShl(builder, clipmask, lp_build_const_int32(gallivm, 18), "");
- have_clipdist = LLVMBuildAnd(builder, mask, lp_build_const_int32(gallivm, 0x4000), "");
- have_clipdist = LLVMBuildShl(builder, have_clipdist, lp_build_const_int32(gallivm, 3), "");
+ if (0) {
+ pad = LLVMBuildAnd(builder, mask, lp_build_const_int32(gallivm, 0x4000), "");
+ pad = LLVMBuildShl(builder, pad, lp_build_const_int32(gallivm, 3), "");
+ }
edgeflag = LLVMBuildAnd(builder, mask, lp_build_const_int32(gallivm, 0x8000), "");
edgeflag = LLVMBuildShl(builder, edgeflag, lp_build_const_int32(gallivm, 1), "");
mask = LLVMBuildOr(builder, vertex_id, clipmask, "");
- mask = LLVMBuildOr(builder, mask, have_clipdist, "");
+ if (0) {
+ mask = LLVMBuildOr(builder, mask, pad, "");
+ }
mask = LLVMBuildOr(builder, mask, edgeflag, "");
#endif
return mask;
int attrib,
int num_outputs,
LLVMValueRef clipmask,
- boolean have_clipdist)
+ boolean need_edgeflag)
{
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef attr_index = lp_build_const_int32(gallivm, attrib);
LLVMValueRef inds[LP_MAX_VECTOR_WIDTH / 32];
+ LLVMValueRef linear_inds[LP_MAX_VECTOR_WIDTH / 32];
LLVMValueRef io_ptrs[LP_MAX_VECTOR_WIDTH / 32];
int vector_length = soa_type.length;
int i;
debug_assert(TGSI_NUM_CHANNELS == 4);
for (i = 0; i < vector_length; i++) {
+ linear_inds[i] = lp_build_const_int32(gallivm, i);
if (indices) {
inds[i] = indices[i];
} else {
- inds[i] = lp_build_const_int32(gallivm, i);
+ inds[i] = linear_inds[i];
}
io_ptrs[i] = LLVMBuildGEP(builder, io_ptr, &inds[i], 1, "");
}
* code here. See struct vertex_header in draw_private.h.
*/
assert(DRAW_TOTAL_CLIP_PLANES==14);
- /* initialize vertex id:16 = 0xffff, have_clipdist:1 = 0, edgeflag:1 = 1 */
- vertex_id_pad_edgeflag = (0xffff << 16) | (1 << DRAW_TOTAL_CLIP_PLANES);
- if (have_clipdist)
- vertex_id_pad_edgeflag |= 1 << (DRAW_TOTAL_CLIP_PLANES+1);
- val = lp_build_const_int_vec(gallivm, lp_int_type(soa_type), vertex_id_pad_edgeflag);
+ /* initialize vertex id:16 = 0xffff, pad:1 = 0, edgeflag:1 = 1 */
+ if (!need_edgeflag) {
+ vertex_id_pad_edgeflag = (0xffff << 16) | (1 << DRAW_TOTAL_CLIP_PLANES);
+ }
+ else {
+ vertex_id_pad_edgeflag = (0xffff << 16);
+ }
+ val = lp_build_const_int_vec(gallivm, lp_int_type(soa_type),
+ vertex_id_pad_edgeflag);
/* OR with the clipmask */
cliptmp = LLVMBuildOr(builder, val, clipmask, "");
for (i = 0; i < vector_length; i++) {
LLVMValueRef id_ptr = draw_jit_header_id(gallivm, io_ptrs[i]);
- val = LLVMBuildExtractElement(builder, cliptmp, inds[i], "");
+ val = LLVMBuildExtractElement(builder, cliptmp, linear_inds[i], "");
val = adjust_mask(gallivm, val);
- LLVMBuildStore(builder, val, id_ptr);
#if DEBUG_STORE
lp_build_printf(gallivm, "io = %p, index %d, clipmask = %x\n",
io_ptrs[i], inds[i], val);
#endif
+ LLVMBuildStore(builder, val, id_ptr);
}
}
LLVMValueRef clipmask,
int num_outputs,
struct lp_type soa_type,
- boolean have_clipdist)
+ boolean need_edgeflag)
{
LLVMBuilderRef builder = gallivm->builder;
unsigned chan, attrib, i;
LLVMConstInt(LLVMInt32TypeInContext(gallivm->context),
chan, 0));
lp_build_print_value(gallivm, "val = ", out);
+ {
+ LLVMValueRef iv =
+ LLVMBuildBitCast(builder, out, lp_build_int_vec_type(gallivm, soa_type), "");
+
+ lp_build_print_value(gallivm, " ival = ", iv);
+ }
#endif
soa[chan] = out;
}
aos,
attrib,
num_outputs,
- clipmask, have_clipdist);
+ clipmask,
+ need_edgeflag);
}
#if DEBUG_STORE
lp_build_printf(gallivm, " # storing end\n");
const struct lp_type vs_type,
LLVMValueRef io_ptr,
LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS],
- boolean pre_clip_pos, int idx)
+ int idx)
{
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef soa[4];
LLVMValueRef io_ptrs[LP_MAX_VECTOR_WIDTH / 32];
LLVMValueRef inds[LP_MAX_VECTOR_WIDTH / 32];
LLVMValueRef clip_ptrs[LP_MAX_VECTOR_WIDTH / 32];
+ LLVMTypeRef clip_ptr_type =
+ LLVMPointerType(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context),
+ 4), 0);
int i, j;
indices[0] =
soa[2] = LLVMBuildLoad(builder, outputs[idx][2], ""); /*z0 z1 .. zn*/
soa[3] = LLVMBuildLoad(builder, outputs[idx][3], ""); /*w0 w1 .. wn*/
- if (!pre_clip_pos) {
- for (i = 0; i < vs_type.length; i++) {
- clip_ptrs[i] = draw_jit_header_clip(gallivm, io_ptrs[i]);
- }
- } else {
- for (i = 0; i < vs_type.length; i++) {
- clip_ptrs[i] = draw_jit_header_pre_clip_pos(gallivm, io_ptrs[i]);
- }
+ for (i = 0; i < vs_type.length; i++) {
+ clip_ptrs[i] = draw_jit_header_clip_pos(gallivm, io_ptrs[i]);
}
lp_build_transpose_aos(gallivm, vs_type, soa, soa);
}
for (j = 0; j < vs_type.length; j++) {
- LLVMTypeRef clip_ptr_type = LLVMPointerType(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4), 0);
LLVMValueRef clip_ptr;
clip_ptr = LLVMBuildGEP(builder, clip_ptrs[j], indices, 2, "clipo");
clip_ptr = LLVMBuildPointerCast(builder, clip_ptr, clip_ptr_type, "");
/* Unaligned store */
- lp_set_store_alignment(LLVMBuildStore(builder, aos[j], clip_ptr), sizeof(float));
+ LLVMSetAlignment(LLVMBuildStore(builder, aos[j], clip_ptr), sizeof(float));
}
}
int i;
struct gallivm_state *gallivm = variant->gallivm;
struct lp_type f32_type = vs_type;
+ const unsigned pos = variant->llvm->draw->vs.position_output;
LLVMTypeRef vs_type_llvm = lp_build_vec_type(gallivm, vs_type);
- LLVMValueRef out3 = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 .. wn*/
+ LLVMValueRef out3 = LLVMBuildLoad(builder, outputs[pos][3], ""); /*w0 w1 .. wn*/
LLVMValueRef const1 = lp_build_const_vec(gallivm, f32_type, 1.0); /*1.0 1.0 1.0 1.0*/
- LLVMValueRef vp_ptr = draw_jit_context_viewport(gallivm, context_ptr);
+ LLVMValueRef vp_ptr = draw_jit_context_viewports(gallivm, context_ptr);
+
+ /* We treat pipe_viewport_state as a float array */
+ const int scale_index_offset = offsetof(struct pipe_viewport_state, scale) / sizeof(float);
+ const int trans_index_offset = offsetof(struct pipe_viewport_state, translate) / sizeof(float);
/* for 1/w convention*/
out3 = LLVMBuildFDiv(builder, const1, out3, "");
- LLVMBuildStore(builder, out3, outputs[0][3]);
+ LLVMBuildStore(builder, out3, outputs[pos][3]);
/* Viewport Mapping */
for (i=0; i<3; i++) {
- LLVMValueRef out = LLVMBuildLoad(builder, outputs[0][i], ""); /*x0 x1 .. xn*/
+ LLVMValueRef out = LLVMBuildLoad(builder, outputs[pos][i], ""); /*x0 x1 .. xn*/
LLVMValueRef scale;
LLVMValueRef trans;
LLVMValueRef scale_i;
LLVMValueRef trans_i;
LLVMValueRef index;
- index = lp_build_const_int32(gallivm, i);
+ index = lp_build_const_int32(gallivm, i + scale_index_offset);
scale_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
- index = lp_build_const_int32(gallivm, i+4);
+ index = lp_build_const_int32(gallivm, i + trans_index_offset);
trans_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
scale = lp_build_broadcast(gallivm, vs_type_llvm,
out = LLVMBuildFAdd(builder, out, trans, "");
/* store transformed outputs */
- LLVMBuildStore(builder, out, outputs[0][i]);
+ LLVMBuildStore(builder, out, outputs[pos][i]);
}
}
struct gallivm_state *gallivm,
struct lp_type vs_type,
LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS],
- boolean clip_xy,
- boolean clip_z,
- boolean clip_user,
- boolean clip_halfz,
- unsigned ucp_enable,
+ struct draw_llvm_variant_key *key,
LLVMValueRef context_ptr,
boolean *have_clipdist)
{
LLVMValueRef plane1, planes, plane_ptr, sum;
struct lp_type f32_type = vs_type;
struct lp_type i32_type = lp_int_type(vs_type);
- const unsigned pos = draw_current_shader_position_output(llvm->draw);
- const unsigned cv = draw_current_shader_clipvertex_output(llvm->draw);
+ const unsigned pos = llvm->draw->vs.position_output;
+ const unsigned cv = llvm->draw->vs.clipvertex_output;
int num_written_clipdistance = llvm->draw->vs.vertex_shader->info.num_written_clipdistance;
- bool have_cd = false;
+ boolean have_cd = false;
+ boolean clip_user = key->clip_user;
+ unsigned ucp_enable = key->ucp_enable;
unsigned cd[2];
- cd[0] = draw_current_shader_clipdistance_output(llvm->draw, 0);
- cd[1] = draw_current_shader_clipdistance_output(llvm->draw, 1);
+ cd[0] = llvm->draw->vs.clipdistance_output[0];
+ cd[1] = llvm->draw->vs.clipdistance_output[1];
if (cd[0] != pos || cd[1] != pos)
have_cd = true;
+ if (num_written_clipdistance && !clip_user) {
+ clip_user = true;
+ ucp_enable = (1 << num_written_clipdistance) - 1;
+ }
+
mask = lp_build_const_int_vec(gallivm, i32_type, 0);
temp = lp_build_const_int_vec(gallivm, i32_type, 0);
zero = lp_build_const_vec(gallivm, f32_type, 0); /* 0.0f 0.0f 0.0f 0.0f */
cv_w = pos_w;
}
+ /*
+ * Be careful with the comparisons and NaNs (using llvm's unordered
+ * comparisons here).
+ */
/* Cliptest, for hardwired planes */
- if (clip_xy) {
+ /*
+ * XXX should take guardband into account (currently not in key).
+ * Otherwise might run the draw pipeline stages for nothing.
+ */
+ if (key->clip_xy) {
/* plane 1 */
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w);
temp = shift;
mask = LLVMBuildOr(builder, mask, test, "");
}
- if (clip_z) {
+ if (key->clip_z) {
temp = lp_build_const_int_vec(gallivm, i32_type, 16);
- if (clip_halfz) {
+ if (key->clip_halfz) {
/* plane 5 */
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, pos_z);
test = LLVMBuildAnd(builder, test, temp, "");
if (clip_user) {
LLVMValueRef planes_ptr = draw_jit_context_planes(gallivm, context_ptr);
LLVMValueRef indices[3];
+ LLVMValueRef is_nan_or_inf;
/* userclip planes */
while (ucp_enable) {
clipdist = LLVMBuildLoad(builder, outputs[cd[1]][i-4], "");
}
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, clipdist);
- temp = lp_build_const_int_vec(gallivm, i32_type, 1 << plane_idx);
+ is_nan_or_inf = lp_build_is_inf_or_nan(gallivm, vs_type, clipdist);
+ test = LLVMBuildOr(builder, test, is_nan_or_inf, "");
+ temp = lp_build_const_int_vec(gallivm, i32_type, 1LL << plane_idx);
test = LLVMBuildAnd(builder, test, temp, "");
mask = LLVMBuildOr(builder, mask, test, "");
} else {
sum = LLVMBuildFAdd(builder, sum, test, "");
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, sum);
- temp = lp_build_const_int_vec(gallivm, i32_type, 1 << plane_idx);
+ temp = lp_build_const_int_vec(gallivm, i32_type, 1LL << plane_idx);
test = LLVMBuildAnd(builder, test, temp, "");
mask = LLVMBuildOr(builder, mask, test, "");
}
}
}
+ if (key->need_edgeflags) {
+ /*
+ * This isn't really part of clipmask but stored the same in vertex
+ * header later, so do it here.
+ */
+ unsigned edge_attr = llvm->draw->vs.edgeflag_output;
+ LLVMValueRef one = lp_build_const_vec(gallivm, f32_type, 1.0);
+ LLVMValueRef edgeflag = LLVMBuildLoad(builder, outputs[edge_attr][0], "");
+ test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_EQUAL, one, edgeflag);
+ temp = lp_build_const_int_vec(gallivm, i32_type,
+ 1LL << DRAW_TOTAL_CLIP_PLANES);
+ test = LLVMBuildAnd(builder, test, temp, "");
+ mask = LLVMBuildOr(builder, mask, test, "");
+ }
return mask;
}
static LLVMValueRef
clipmask_booli32(struct gallivm_state *gallivm,
const struct lp_type vs_type,
- LLVMValueRef clipmask_bool_ptr)
+ LLVMValueRef clipmask_bool_ptr,
+ boolean edgeflag_in_clipmask)
{
LLVMBuilderRef builder = gallivm->builder;
LLVMTypeRef int32_type = LLVMInt32TypeInContext(gallivm->context);
int i;
/*
- * Can do this with log2(vector length) pack instructions and one extract
- * (as we don't actually need a or) with sse2 which would be way better.
+ * We need to invert the edgeflag bit from the clipmask here
+ * (because the result is really if we want to run the pipeline or not
+ * and we (may) need it if edgeflag was 0).
+ */
+ if (edgeflag_in_clipmask) {
+ struct lp_type i32_type = lp_int_type(vs_type);
+ LLVMValueRef edge = lp_build_const_int_vec(gallivm, i32_type,
+ 1LL << DRAW_TOTAL_CLIP_PLANES);
+ clipmask_bool = LLVMBuildXor(builder, clipmask_bool, edge, "");
+ }
+ /*
+ * Could do much better with just cmp/movmskps.
*/
for (i=0; i < vs_type.length; i++) {
temp = LLVMBuildExtractElement(builder, clipmask_bool,
static LLVMValueRef
draw_gs_llvm_fetch_input(const struct lp_build_tgsi_gs_iface *gs_iface,
struct lp_build_tgsi_context * bld_base,
+ boolean is_vindex_indirect,
LLVMValueRef vertex_index,
+ boolean is_aindex_indirect,
LLVMValueRef attrib_index,
LLVMValueRef swizzle_index)
{
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef indices[3];
LLVMValueRef res;
+ struct lp_type type = bld_base->base.type;
+
+ if (is_vindex_indirect || is_aindex_indirect) {
+ int i;
+ res = bld_base->base.zero;
+ for (i = 0; i < type.length; ++i) {
+ LLVMValueRef idx = lp_build_const_int32(gallivm, i);
+ LLVMValueRef vert_chan_index = vertex_index;
+ LLVMValueRef attr_chan_index = attrib_index;
+ LLVMValueRef channel_vec, value;
+
+ if (is_vindex_indirect) {
+ vert_chan_index = LLVMBuildExtractElement(builder,
+ vertex_index, idx, "");
+ }
+ if (is_aindex_indirect) {
+ attr_chan_index = LLVMBuildExtractElement(builder,
+ attrib_index, idx, "");
+ }
- indices[0] = vertex_index;
- indices[1] = attrib_index;
- indices[2] = swizzle_index;
-
- res = LLVMBuildGEP(builder, gs->input, indices, 3, "");
- res = LLVMBuildLoad(builder, res, "");
+ indices[0] = vert_chan_index;
+ indices[1] = attr_chan_index;
+ indices[2] = swizzle_index;
+
+ channel_vec = LLVMBuildGEP(builder, gs->input, indices, 3, "");
+ channel_vec = LLVMBuildLoad(builder, channel_vec, "");
+ value = LLVMBuildExtractElement(builder, channel_vec, idx, "");
+
+ res = LLVMBuildInsertElement(builder, res, value, idx, "");
+ }
+ } else {
+ indices[0] = vertex_index;
+ indices[1] = attrib_index;
+ indices[2] = swizzle_index;
+
+ res = LLVMBuildGEP(builder, gs->input, indices, 3, "");
+ res = LLVMBuildLoad(builder, res, "");
+ }
return res;
}
LLVMValueRef clipmask = lp_build_const_int_vec(gallivm,
lp_int_type(gs_type), 0);
LLVMValueRef indices[LP_MAX_VECTOR_LENGTH];
- LLVMValueRef max_output_vertices =
- lp_build_const_int32(gallivm, variant->shader->base.max_output_vertices);
+ LLVMValueRef next_prim_offset =
+ lp_build_const_int32(gallivm, variant->shader->base.primitive_boundary);
LLVMValueRef io = variant->io_ptr;
unsigned i;
const struct tgsi_shader_info *gs_info = &variant->shader->base.info;
LLVMValueRef ind = lp_build_const_int32(gallivm, i);
LLVMValueRef currently_emitted =
LLVMBuildExtractElement(builder, emitted_vertices_vec, ind, "");
- indices[i] = LLVMBuildMul(builder, ind, max_output_vertices, "");
+ indices[i] = LLVMBuildMul(builder, ind, next_prim_offset, "");
indices[i] = LLVMBuildAdd(builder, indices[i], currently_emitted, "");
}
struct gallivm_state *gallivm = variant->gallivm;
LLVMContextRef context = gallivm->context;
LLVMTypeRef int32_type = LLVMInt32TypeInContext(context);
- LLVMTypeRef arg_types[8];
+ LLVMTypeRef arg_types[11];
+ unsigned num_arg_types =
+ elts ? Elements(arg_types) : Elements(arg_types) - 1;
LLVMTypeRef func_type;
LLVMValueRef context_ptr;
LLVMBasicBlockRef block;
LLVMBuilderRef builder;
+ char func_name[64];
struct lp_type vs_type;
LLVMValueRef end, start;
- LLVMValueRef count, fetch_elts, fetch_count;
+ LLVMValueRef count, fetch_elts, fetch_elt_max, fetch_count;
+ LLVMValueRef vertex_id_offset, start_instance;
LLVMValueRef stride, step, io_itr;
LLVMValueRef io_ptr, vbuffers_ptr, vb_ptr;
LLVMValueRef zero = lp_build_const_int32(gallivm, 0);
LLVMValueRef fetch_max;
struct lp_build_sampler_soa *sampler = 0;
LLVMValueRef ret, clipmask_bool_ptr;
- const struct draw_geometry_shader *gs = draw->gs.geometry_shader;
+ struct draw_llvm_variant_key *key = &variant->key;
/* If geometry shader is present we need to skip both the viewport
* transformation and clipping otherwise the inputs to the geometry
* shader will be incorrect.
+ * The code can't handle vp transform when vs writes vp index neither
+ * (though this would be fixable here, but couldn't just broadcast
+ * the values).
*/
- const boolean bypass_viewport = gs || variant->key.bypass_viewport;
- const boolean enable_cliptest = !gs && (variant->key.clip_xy ||
- variant->key.clip_z ||
- variant->key.clip_user);
+ const boolean bypass_viewport = key->has_gs || key->bypass_viewport ||
+ llvm->draw->vs.vertex_shader->info.writes_viewport_index;
+ const boolean enable_cliptest = !key->has_gs && (key->clip_xy ||
+ key->clip_z ||
+ key->clip_user ||
+ key->need_edgeflags);
LLVMValueRef variant_func;
- const unsigned pos = draw_current_shader_position_output(llvm->draw);
- const unsigned cv = draw_current_shader_clipvertex_output(llvm->draw);
+ const unsigned pos = llvm->draw->vs.position_output;
+ const unsigned cv = llvm->draw->vs.clipvertex_output;
boolean have_clipdist = FALSE;
struct lp_bld_tgsi_system_values system_values;
memset(&system_values, 0, sizeof(system_values));
- arg_types[0] = get_context_ptr_type(variant); /* context */
- arg_types[1] = get_vertex_header_ptr_type(variant); /* vertex_header */
- arg_types[2] = get_buffer_ptr_type(variant); /* vbuffers */
- if (elts)
- arg_types[3] = LLVMPointerType(int32_type, 0);/* fetch_elts * */
- else
- arg_types[3] = int32_type; /* start */
- arg_types[4] = int32_type; /* fetch_count / count */
- arg_types[5] = int32_type; /* stride */
- arg_types[6] = get_vb_ptr_type(variant); /* pipe_vertex_buffer's */
- arg_types[7] = int32_type; /* instance_id */
+ util_snprintf(func_name, sizeof(func_name), "draw_llvm_vs_variant%u_%s",
+ variant->shader->variants_cached, elts ? "elts" : "linear");
- func_type = LLVMFunctionType(int32_type, arg_types, Elements(arg_types), 0);
+ i = 0;
+ arg_types[i++] = get_context_ptr_type(variant); /* context */
+ arg_types[i++] = get_vertex_header_ptr_type(variant); /* vertex_header */
+ arg_types[i++] = get_buffer_ptr_type(variant); /* vbuffers */
+ if (elts) {
+ arg_types[i++] = LLVMPointerType(int32_type, 0);/* fetch_elts */
+ arg_types[i++] = int32_type; /* fetch_elt_max */
+ } else
+ arg_types[i++] = int32_type; /* start */
+ arg_types[i++] = int32_type; /* fetch_count / count */
+ arg_types[i++] = int32_type; /* stride */
+ arg_types[i++] = get_vb_ptr_type(variant); /* pipe_vertex_buffer's */
+ arg_types[i++] = int32_type; /* instance_id */
+ arg_types[i++] = int32_type; /* vertex_id_offset */
+ arg_types[i++] = int32_type; /* start_instance */
+
+ func_type = LLVMFunctionType(int32_type, arg_types, num_arg_types, 0);
- variant_func = LLVMAddFunction(gallivm->module,
- elts ? "draw_llvm_shader_elts" : "draw_llvm_shader",
- func_type);
+ variant_func = LLVMAddFunction(gallivm->module, func_name, func_type);
if (elts)
variant->function_elts = variant_func;
variant->function = variant_func;
LLVMSetFunctionCallConv(variant_func, LLVMCCallConv);
- for (i = 0; i < Elements(arg_types); ++i)
+ for (i = 0; i < num_arg_types; ++i)
if (LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
LLVMAddAttribute(LLVMGetParam(variant_func, i),
LLVMNoAliasAttribute);
context_ptr = LLVMGetParam(variant_func, 0);
io_ptr = LLVMGetParam(variant_func, 1);
vbuffers_ptr = LLVMGetParam(variant_func, 2);
- stride = LLVMGetParam(variant_func, 5);
- vb_ptr = LLVMGetParam(variant_func, 6);
- system_values.instance_id = LLVMGetParam(variant_func, 7);
+ /*
+ * XXX: stride is actually unused. The stride we use is strictly calculated
+ * from the number of outputs (including the draw_extra outputs).
+ * Should probably fix some day (we need a new vs just because of extra
+ * outputs which the generated vs won't touch).
+ */
+ stride = LLVMGetParam(variant_func, 5 + (elts ? 1 : 0));
+ vb_ptr = LLVMGetParam(variant_func, 6 + (elts ? 1 : 0));
+ system_values.instance_id = LLVMGetParam(variant_func, 7 + (elts ? 1 : 0));
+ vertex_id_offset = LLVMGetParam(variant_func, 8 + (elts ? 1 : 0));
+ start_instance = LLVMGetParam(variant_func, 9 + (elts ? 1 : 0));
lp_build_name(context_ptr, "context");
lp_build_name(io_ptr, "io");
lp_build_name(stride, "stride");
lp_build_name(vb_ptr, "vb");
lp_build_name(system_values.instance_id, "instance_id");
+ lp_build_name(vertex_id_offset, "vertex_id_offset");
+ lp_build_name(start_instance, "start_instance");
if (elts) {
- fetch_elts = LLVMGetParam(variant_func, 3);
- fetch_count = LLVMGetParam(variant_func, 4);
+ fetch_elts = LLVMGetParam(variant_func, 3);
+ fetch_elt_max = LLVMGetParam(variant_func, 4);
+ fetch_count = LLVMGetParam(variant_func, 5);
lp_build_name(fetch_elts, "fetch_elts");
+ lp_build_name(fetch_elt_max, "fetch_elt_max");
lp_build_name(fetch_count, "fetch_count");
start = count = NULL;
}
LLVMBuildStore(builder, lp_build_zero(gallivm, lp_int_type(vs_type)), clipmask_bool_ptr);
/* code generated texture sampling */
- sampler = draw_llvm_sampler_soa_create(
- draw_llvm_variant_key_samplers(&variant->key),
- context_ptr);
+ sampler = draw_llvm_sampler_soa_create(draw_llvm_variant_key_samplers(key));
if (elts) {
start = zero;
end = fetch_count;
+ count = fetch_count;
}
else {
end = lp_build_add(&bld, start, count);
fetch_max = LLVMBuildSub(builder, end, one, "fetch_max");
- lp_build_loop_begin(&lp_loop, gallivm, start);
+ lp_build_loop_begin(&lp_loop, gallivm, zero);
{
LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][TGSI_NUM_CHANNELS];
LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][LP_MAX_VECTOR_WIDTH / 32] = { { 0 } };
LLVMValueRef io;
LLVMValueRef clipmask; /* holds the clipmask value */
+ LLVMValueRef true_index_array = lp_build_zero(gallivm,
+ lp_type_uint_vec(32, 32*vector_length));
const LLVMValueRef (*ptr_aos)[TGSI_NUM_CHANNELS];
- if (elts)
- io_itr = lp_loop.counter;
- else
- io_itr = LLVMBuildSub(builder, lp_loop.counter, start, "");
+ io_itr = lp_loop.counter;
io = LLVMBuildGEP(builder, io_ptr, &io_itr, 1, "");
#if DEBUG_STORE
lp_build_printf(gallivm, " --- io %d = %p, loop counter %d\n",
io_itr, io, lp_loop.counter);
#endif
- system_values.vertex_id = lp_build_zero(gallivm, lp_type_uint_vec(32, 32*vector_length));
for (i = 0; i < vector_length; ++i) {
- LLVMValueRef true_index =
+ LLVMValueRef vert_index =
LLVMBuildAdd(builder,
lp_loop.counter,
lp_build_const_int32(gallivm, i), "");
+ LLVMValueRef true_index =
+ LLVMBuildAdd(builder, start, vert_index, "");
/* make sure we're not out of bounds which can happen
* if fetch_count % 4 != 0, because on the last iteration
if (elts) {
LLVMValueRef fetch_ptr;
- fetch_ptr = LLVMBuildGEP(builder, fetch_elts,
- &true_index, 1, "");
- true_index = LLVMBuildLoad(builder, fetch_ptr, "fetch_elt");
+ LLVMValueRef index_overflowed;
+ LLVMValueRef index_ptr =
+ lp_build_alloca(
+ gallivm,
+ lp_build_vec_type(gallivm, lp_type_int(32)), "");
+ struct lp_build_if_state if_ctx;
+ index_overflowed = LLVMBuildICmp(builder, LLVMIntUGT,
+ true_index, fetch_elt_max,
+ "index_overflowed");
+
+ lp_build_if(&if_ctx, gallivm, index_overflowed);
+ {
+ /* Generate maximum possible index so that
+ * generate_fetch can treat it just like
+ * any other overflow and return zeros.
+ * We don't have to worry about the restart
+ * primitive index because it has already been
+ * handled
+ */
+ LLVMValueRef val =
+ lp_build_const_int32(gallivm, 0xffffffff);
+ LLVMBuildStore(builder, val, index_ptr);
+ }
+ lp_build_else(&if_ctx);
+ {
+ LLVMValueRef val;
+ fetch_ptr = LLVMBuildGEP(builder, fetch_elts,
+ &true_index, 1, "");
+ val = LLVMBuildLoad(builder, fetch_ptr, "");
+ LLVMBuildStore(builder, val, index_ptr);
+ }
+ lp_build_endif(&if_ctx);
+ true_index = LLVMBuildLoad(builder, index_ptr, "true_index");
}
+ true_index_array = LLVMBuildInsertElement(
+ gallivm->builder, true_index_array, true_index,
+ lp_build_const_int32(gallivm, i), "");
- system_values.vertex_id = LLVMBuildInsertElement(gallivm->builder,
- system_values.vertex_id, true_index,
- lp_build_const_int32(gallivm, i), "");
for (j = 0; j < draw->pt.nr_vertex_elements; ++j) {
struct pipe_vertex_element *velem = &draw->pt.vertex_element[j];
LLVMValueRef vb_index =
lp_build_const_int32(gallivm, velem->vertex_buffer_index);
LLVMValueRef vb = LLVMBuildGEP(builder, vb_ptr, &vb_index, 1, "");
- generate_fetch(gallivm, vbuffers_ptr,
+ generate_fetch(gallivm, draw, vbuffers_ptr,
&aos_attribs[j][i], velem, vb, true_index,
- system_values.instance_id);
+ system_values.instance_id, start_instance);
}
}
convert_to_soa(gallivm, aos_attribs, inputs,
draw->pt.nr_vertex_elements, vs_type);
+ /* In the paths with elts vertex id has to be unaffected by the
+ * index bias and because indices inside our elements array have
+ * already had index bias applied we need to subtract it here to
+ * get back to the original index.
+ * in the linear paths vertex id has to be unaffected by the
+ * original start index and because we abuse the 'start' variable
+ * to either represent the actual start index or the index at which
+ * the primitive was split (we split rendering into chunks of at
+ * most 4095-vertices) we need to back out the original start
+ * index out of our vertex id here.
+ */
+ system_values.basevertex = lp_build_broadcast(gallivm, lp_build_vec_type(gallivm,
+ lp_type_uint_vec(32, 32*vector_length)),
+ vertex_id_offset);
+ system_values.vertex_id = true_index_array;
+ system_values.vertex_id_nobase = LLVMBuildSub(builder, true_index_array,
+ system_values.basevertex, "");
+
ptr_aos = (const LLVMValueRef (*)[TGSI_NUM_CHANNELS]) inputs;
generate_vs(variant,
builder,
&system_values,
context_ptr,
sampler,
- variant->key.clamp_vertex_color);
+ key->clamp_vertex_color);
if (pos != -1 && cv != -1) {
/* store original positions in clip before further manipulation */
- store_clip(gallivm, vs_type, io, outputs, 0, cv);
- store_clip(gallivm, vs_type, io, outputs, 1, pos);
+ store_clip(gallivm, vs_type, io, outputs, pos);
/* do cliptest */
if (enable_cliptest) {
gallivm,
vs_type,
outputs,
- variant->key.clip_xy,
- variant->key.clip_z,
- variant->key.clip_user,
- variant->key.clip_halfz,
- variant->key.ucp_enable,
+ key,
context_ptr, &have_clipdist);
temp = LLVMBuildOr(builder, clipmask, temp, "");
/* store temporary clipping boolean value */
*/
convert_to_aos(gallivm, io, NULL, outputs, clipmask,
vs_info->num_outputs, vs_type,
- have_clipdist);
+ enable_cliptest && key->need_edgeflags);
}
-
- lp_build_loop_end_cond(&lp_loop, end, step, LLVMIntUGE);
+ lp_build_loop_end_cond(&lp_loop, count, step, LLVMIntUGE);
sampler->destroy(sampler);
/* return clipping boolean value for function */
- ret = clipmask_booli32(gallivm, vs_type, clipmask_bool_ptr);
+ ret = clipmask_booli32(gallivm, vs_type, clipmask_bool_ptr,
+ enable_cliptest && key->need_edgeflags);
LLVMBuildRet(builder, ret);
key = (struct draw_llvm_variant_key *)store;
+ memset(key, 0, offsetof(struct draw_llvm_variant_key, vertex_element[0]));
+
key->clamp_vertex_color = llvm->draw->rasterizer->clamp_vertex_color; /**/
/* Presumably all variants of the shader should have the same
key->clip_xy = llvm->draw->clip_xy;
key->clip_z = llvm->draw->clip_z;
key->clip_user = llvm->draw->clip_user;
- key->bypass_viewport = llvm->draw->identity_viewport;
- key->clip_halfz = !llvm->draw->rasterizer->gl_rasterization_rules;
+ key->bypass_viewport = llvm->draw->bypass_viewport;
+ key->clip_halfz = llvm->draw->rasterizer->clip_halfz;
+ /* XXX assumes edgeflag output not at 0 */
key->need_edgeflags = (llvm->draw->vs.edgeflag_output ? TRUE : FALSE);
key->ucp_enable = llvm->draw->rasterizer->clip_plane_enable;
key->has_gs = llvm->draw->gs.geometry_shader != NULL;
- key->pad1 = 0;
+ key->num_outputs = draw_total_vs_outputs(llvm->draw);
/* All variants of this shader will have the same value for
* nr_samplers. Not yet trying to compact away holes in the
assert(sview_idx < Elements(draw->llvm->gs_jit_context.textures));
jit_tex = &draw->llvm->gs_jit_context.textures[sview_idx];
+ } else {
+ assert(0);
+ return;
}
jit_tex->width = width;
for (i = 0; i < draw->num_samplers[PIPE_SHADER_VERTEX]; i++) {
struct draw_jit_sampler *jit_sam = &draw->llvm->jit_context.samplers[i];
- if (draw->samplers[i]) {
+ if (draw->samplers[PIPE_SHADER_VERTEX][i]) {
const struct pipe_sampler_state *s
= draw->samplers[PIPE_SHADER_VERTEX][i];
jit_sam->min_lod = s->min_lod;
for (i = 0; i < draw->num_samplers[PIPE_SHADER_GEOMETRY]; i++) {
struct draw_jit_sampler *jit_sam = &draw->llvm->gs_jit_context.samplers[i];
- if (draw->samplers[i]) {
+ if (draw->samplers[PIPE_SHADER_GEOMETRY][i]) {
const struct pipe_sampler_state *s
= draw->samplers[PIPE_SHADER_GEOMETRY][i];
jit_sam->min_lod = s->min_lod;
{
struct draw_llvm *llvm = variant->llvm;
- if (variant->function_elts) {
- gallivm_free_function(variant->gallivm,
- variant->function_elts, variant->jit_func_elts);
- }
-
- if (variant->function) {
- gallivm_free_function(variant->gallivm,
- variant->function, variant->jit_func);
- }
-
gallivm_destroy(variant->gallivm);
remove_from_list(&variant->list_item_local);
{
struct gallivm_state *gallivm = variant->gallivm;
LLVMBuilderRef builder = gallivm->builder;
- LLVMValueRef bits[16];
- struct lp_type mask_type = lp_int_type(gs_type);
- struct lp_type mask_elem_type = lp_elem_type(mask_type);
- LLVMValueRef mask_val = lp_build_const_vec(gallivm,
- mask_type,
- 0);
+ struct lp_type mask_type = lp_int_type(gs_type);
+ LLVMValueRef num_prims;
+ LLVMValueRef mask_val = lp_build_const_vec(gallivm, mask_type, 0);
unsigned i;
- assert(gs_type.length <= Elements(bits));
-
- for (i = gs_type.length; i >= 1; --i) {
- int idx = i - 1;
- LLVMValueRef ind = lp_build_const_int32(gallivm, i);
- bits[idx] = lp_build_compare(gallivm,
- mask_elem_type, PIPE_FUNC_GEQUAL,
- variant->num_prims, ind);
- }
- for (i = 0; i < gs_type.length; ++i) {
- LLVMValueRef ind = lp_build_const_int32(gallivm, i);
- mask_val = LLVMBuildInsertElement(builder, mask_val, bits[i], ind, "");
+ num_prims = lp_build_broadcast(gallivm, lp_build_vec_type(gallivm, mask_type),
+ variant->num_prims);
+ for (i = 0; i < gs_type.length; i++) {
+ LLVMValueRef idx = lp_build_const_int32(gallivm, i);
+ mask_val = LLVMBuildInsertElement(builder, mask_val, idx, idx, "");
}
- mask_val = lp_build_compare(gallivm,
- mask_type, PIPE_FUNC_NOTEQUAL,
- mask_val,
- lp_build_const_int_vec(gallivm, mask_type, 0));
+ mask_val = lp_build_compare(gallivm, mask_type,
+ PIPE_FUNC_GREATER, num_prims, mask_val);
return mask_val;
}
struct gallivm_state *gallivm = variant->gallivm;
LLVMContextRef context = gallivm->context;
LLVMTypeRef int32_type = LLVMInt32TypeInContext(context);
- LLVMTypeRef arg_types[6];
+ LLVMTypeRef arg_types[7];
LLVMTypeRef func_type;
LLVMValueRef variant_func;
LLVMValueRef context_ptr;
struct lp_build_sampler_soa *sampler = 0;
struct lp_build_context bld;
struct lp_bld_tgsi_system_values system_values;
+ char func_name[64];
struct lp_type gs_type;
unsigned i;
struct draw_gs_llvm_iface gs_iface;
const struct tgsi_token *tokens = variant->shader->base.state.tokens;
- LLVMValueRef consts_ptr;
+ LLVMValueRef consts_ptr, num_consts_ptr;
LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][TGSI_NUM_CHANNELS];
struct lp_build_mask_context mask;
const struct tgsi_shader_info *gs_info = &variant->shader->base.info;
memset(&system_values, 0, sizeof(system_values));
+ util_snprintf(func_name, sizeof(func_name), "draw_llvm_gs_variant%u",
+ variant->shader->variants_cached);
+
assert(variant->vertex_header_ptr_type);
arg_types[0] = get_gs_context_ptr_type(variant); /* context */
arg_types[4] = int32_type; /* instance_id */
arg_types[5] = LLVMPointerType(
LLVMVectorType(int32_type, vector_length), 0); /* prim_id_ptr */
+ arg_types[6] = int32_type;
func_type = LLVMFunctionType(int32_type, arg_types, Elements(arg_types), 0);
- variant_func = LLVMAddFunction(gallivm->module, "draw_geometry_shader",
- func_type);
+ variant_func = LLVMAddFunction(gallivm->module, func_name, func_type);
+
variant->function = variant_func;
LLVMSetFunctionCallConv(variant_func, LLVMCCallConv);
num_prims = LLVMGetParam(variant_func, 3);
system_values.instance_id = LLVMGetParam(variant_func, 4);
prim_id_ptr = LLVMGetParam(variant_func, 5);
+ system_values.invocation_id = LLVMGetParam(variant_func, 6);
lp_build_name(context_ptr, "context");
lp_build_name(input_array, "input");
lp_build_name(io_ptr, "io");
- lp_build_name(io_ptr, "num_prims");
+ lp_build_name(num_prims, "num_prims");
lp_build_name(system_values.instance_id, "instance_id");
lp_build_name(prim_id_ptr, "prim_id_ptr");
+ lp_build_name(system_values.invocation_id, "invocation_id");
variant->context_ptr = context_ptr;
variant->io_ptr = io_ptr;
gs_type.length = vector_length;
consts_ptr = draw_gs_jit_context_constants(variant->gallivm, context_ptr);
+ num_consts_ptr =
+ draw_gs_jit_context_num_constants(variant->gallivm, context_ptr);
/* code generated texture sampling */
- sampler = draw_llvm_sampler_soa_create(variant->key.samplers,
- context_ptr);
+ sampler = draw_llvm_sampler_soa_create(variant->key.samplers);
mask_val = generate_mask_value(variant, gs_type);
lp_build_mask_begin(&mask, gallivm, gs_type, mask_val);
if (gs_info->uses_primid) {
- system_values.prim_id = LLVMBuildLoad(builder, prim_id_ptr, "prim_id");;
+ system_values.prim_id = LLVMBuildLoad(builder, prim_id_ptr, "prim_id");
+ }
+
+ if (gallivm_debug & (GALLIVM_DEBUG_TGSI | GALLIVM_DEBUG_IR)) {
+ tgsi_dump(tokens, 0);
+ draw_gs_llvm_dump_variant_key(&variant->key);
}
lp_build_tgsi_soa(variant->gallivm,
gs_type,
&mask,
consts_ptr,
+ num_consts_ptr,
&system_values,
- NULL /*pos*/,
NULL,
outputs,
+ context_ptr,
+ NULL,
sampler,
&llvm->draw->gs.geometry_shader->info,
(const struct lp_build_tgsi_gs_iface *)&gs_iface);
struct llvm_geometry_shader *shader =
llvm_geometry_shader(llvm->draw->gs.geometry_shader);
LLVMTypeRef vertex_header;
+ char module_name[64];
variant = MALLOC(sizeof *variant +
shader->variant_key_size -
sizeof variant->key);
- if (variant == NULL)
+ if (!variant)
return NULL;
variant->llvm = llvm;
variant->shader = shader;
- variant->gallivm = gallivm_create();
+ util_snprintf(module_name, sizeof(module_name), "draw_llvm_gs_variant%u",
+ variant->shader->variants_cached);
+
+ variant->gallivm = gallivm_create(module_name, llvm->context);
create_gs_jit_types(variant);
variant->jit_func = (draw_gs_jit_func)
gallivm_jit_function(variant->gallivm, variant->function);
+ gallivm_free_ir(variant->gallivm);
+
variant->list_item_global.base = variant;
variant->list_item_local.base = variant;
/*variant->no = */shader->variants_created++;
{
struct draw_llvm *llvm = variant->llvm;
- if (variant->function) {
- gallivm_free_function(variant->gallivm,
- variant->function, variant->jit_func);
- }
-
gallivm_destroy(variant->gallivm);
remove_from_list(&variant->list_item_local);
key = (struct draw_gs_llvm_variant_key *)store;
+ memset(key, 0, offsetof(struct draw_gs_llvm_variant_key, samplers[0]));
+
+ key->num_outputs = draw_total_gs_outputs(llvm->draw);
+
/* All variants of this shader will have the same value for
* nr_samplers. Not yet trying to compact away holes in the
* sampler array.
projects / mesa.git / blobdiff