#include "draw_context.h"
#include "draw_vs.h"
+#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"
boolean elts);
+struct draw_gs_llvm_iface {
+ struct lp_build_tgsi_gs_iface base;
+
+ struct draw_gs_llvm_variant *variant;
+ LLVMValueRef input;
+};
+
+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);
+
+#if HAVE_LLVM < 0x0300
+ LLVMAddTypeName(gallivm->module, struct_name, dvbuffer_type);
+
+ /* Make sure the target's struct layout cache doesn't return
+ * stale/invalid data.
+ */
+ LLVMInvalidateStructLayout(gallivm->target, dvbuffer_type);
+#endif
+
+ 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
*/
{
LLVMTargetDataRef target = gallivm->target;
LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
- LLVMTypeRef elem_types[6];
+ 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] = elem_types[0]; /* gs_constants */
- elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(float_type, 4),
+ elem_types[1] = LLVMPointerType(LLVMArrayType(LLVMArrayType(float_type, 4),
DRAW_TOTAL_CLIP_PLANES), 0);
- elem_types[3] = LLVMPointerType(float_type, 0); /* viewport */
- elem_types[4] = LLVMArrayType(texture_type,
+ elem_types[2] = LLVMPointerType(float_type, 0); /* viewport */
+ elem_types[3] = LLVMArrayType(texture_type,
PIPE_MAX_SHADER_SAMPLER_VIEWS); /* textures */
- elem_types[5] = LLVMArrayType(sampler_type,
+ elem_types[4] = LLVMArrayType(sampler_type,
PIPE_MAX_SAMPLERS); /* samplers */
context_type = LLVMStructTypeInContext(gallivm->context, elem_types,
Elements(elem_types), 0);
#endif
LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, vs_constants,
- target, context_type, 0);
- LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, gs_constants,
- target, context_type, 1);
+ target, context_type, DRAW_JIT_CTX_CONSTANTS);
LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, planes,
- target, context_type, 2);
+ target, context_type, DRAW_JIT_CTX_PLANES);
LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, viewport,
- target, context_type, 3);
+ target, context_type, DRAW_JIT_CTX_VIEWPORT);
LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures,
target, context_type,
DRAW_JIT_CTX_TEXTURES);
}
+/**
+ * Create LLVM type for struct draw_gs_jit_context
+ */
+static LLVMTypeRef
+create_gs_jit_context_type(struct gallivm_state *gallivm,
+ unsigned vector_length,
+ LLVMTypeRef texture_type, LLVMTypeRef sampler_type,
+ const char *struct_name)
+{
+ LLVMTargetDataRef target = gallivm->target;
+ LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
+ LLVMTypeRef int_type = LLVMInt32TypeInContext(gallivm->context);
+ LLVMTypeRef elem_types[DRAW_GS_JIT_CTX_NUM_FIELDS];
+ LLVMTypeRef 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),
+ DRAW_TOTAL_CLIP_PLANES), 0);
+ elem_types[2] = LLVMPointerType(float_type, 0); /* viewport */
+
+ elem_types[3] = LLVMArrayType(texture_type,
+ PIPE_MAX_SHADER_SAMPLER_VIEWS); /* textures */
+ elem_types[4] = 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[7] = 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
+
+ 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, planes,
+ target, context_type, DRAW_GS_JIT_CTX_PLANES);
+ LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, viewport,
+ target, context_type, DRAW_GS_JIT_CTX_VIEWPORT);
+ LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, textures,
+ target, context_type,
+ DRAW_GS_JIT_CTX_TEXTURES);
+ LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, samplers,
+ target, context_type,
+ DRAW_GS_JIT_CTX_SAMPLERS);
+ LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, prim_lengths,
+ target, context_type,
+ DRAW_GS_JIT_CTX_PRIM_LENGTHS);
+ LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, emitted_vertices,
+ target, context_type,
+ DRAW_GS_JIT_CTX_EMITTED_VERTICES);
+ LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, emitted_prims,
+ target, context_type,
+ DRAW_GS_JIT_CTX_EMITTED_PRIMS);
+ LP_CHECK_STRUCT_SIZE(struct draw_gs_jit_context,
+ target, context_type);
+
+ return context_type;
+}
+
+
+static LLVMTypeRef
+create_gs_jit_input_type(struct gallivm_state *gallivm)
+{
+ LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
+ LLVMTypeRef input_array;
+
+ input_array = LLVMVectorType(float_type, TGSI_NUM_CHANNELS); /* num primitives */
+ input_array = LLVMArrayType(input_array, TGSI_NUM_CHANNELS); /* num channels */
+ input_array = LLVMArrayType(input_array, PIPE_MAX_SHADER_INPUTS); /* num attrs per vertex */
+ input_array = LLVMPointerType(input_array, 0); /* num vertices per prim */
+
+ return input_array;
+}
+
/**
* 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);
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);
}
llvm->nr_variants = 0;
make_empty_list(&llvm->vs_variants_list);
+ llvm->nr_gs_variants = 0;
+ make_empty_list(&llvm->gs_variants_list);
+
return llvm;
}
NULL /*struct lp_build_mask_context *mask*/,
consts_ptr,
system_values,
- NULL /*pos*/,
inputs,
outputs,
sampler,
- &llvm->draw->vs.vertex_shader->info);
+ &llvm->draw->vs.vertex_shader->info,
+ NULL);
{
LLVMValueRef out;
}
}
-
static void
generate_fetch(struct gallivm_state *gallivm,
+ struct draw_context *draw,
LLVMValueRef vbuffers_ptr,
LLVMValueRef *res,
struct pipe_vertex_element *velem,
LLVMValueRef index,
LLVMValueRef instance_id)
{
- 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;
+ index = lp_build_const_int32(gallivm, draw->start_instance);
+ current_instance = LLVMBuildUDiv(builder, instance_id,
+ lp_build_const_int32(gallivm, velem->instance_divisor),
+ "instance_divisor");
+ index = lp_build_uadd_overflow(gallivm, index, 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(),
+ map_ptr,
+ zero, zero, zero);
+ 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
indices[1] = index;
indices[2] = lp_build_const_int32(gallivm, 0);
+ data_ptr = LLVMBuildGEP(builder, data_ptr, indices, 3, "");
+ data_ptr = LLVMBuildPointerCast(builder, data_ptr, data_ptr_type, "");
+
#if DEBUG_STORE
lp_build_printf(gallivm, " ---- %p storing attribute %d (io = %p)\n", data_ptr, index, io_ptr);
#endif
- data_ptr = LLVMBuildGEP(builder, data_ptr, indices, 3, "");
- data_ptr = LLVMBuildPointerCast(builder, data_ptr, data_ptr_type, "");
-
/* Unaligned store due to the vertex header */
lp_set_store_alignment(LLVMBuildStore(builder, value, data_ptr), sizeof(float));
}
store_aos_array(struct gallivm_state *gallivm,
struct lp_type soa_type,
LLVMValueRef io_ptr,
+ LLVMValueRef *indices,
LLVMValueRef* aos,
int attrib,
int num_outputs,
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++) {
- inds[i] = lp_build_const_int32(gallivm, i);
+ linear_inds[i] = lp_build_const_int32(gallivm, i);
+ if (indices) {
+ inds[i] = indices[i];
+ } else {
+ inds[i] = linear_inds[i];
+ }
io_ptrs[i] = LLVMBuildGEP(builder, io_ptr, &inds[i], 1, "");
}
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\n, clipmask = %x\n",
+ lp_build_printf(gallivm, "io = %p, index %d, clipmask = %x\n",
io_ptrs[i], inds[i], val);
#endif
}
static void
convert_to_aos(struct gallivm_state *gallivm,
LLVMValueRef io,
+ LLVMValueRef *indices,
LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS],
LLVMValueRef clipmask,
int num_outputs,
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;
}
store_aos_array(gallivm,
soa_type,
- io,
+ io, indices,
aos,
attrib,
num_outputs,
static void
store_clip(struct gallivm_state *gallivm,
const struct lp_type vs_type,
- LLVMValueRef io_ptr,
+ LLVMValueRef io_ptr,
LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS],
boolean pre_clip_pos, int idx)
{
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef soa[4];
LLVMValueRef aos[LP_MAX_VECTOR_LENGTH];
- LLVMValueRef indices[2];
+ LLVMValueRef indices[2];
LLVMValueRef io_ptrs[LP_MAX_VECTOR_WIDTH / 32];
LLVMValueRef inds[LP_MAX_VECTOR_WIDTH / 32];
LLVMValueRef clip_ptrs[LP_MAX_VECTOR_WIDTH / 32];
indices[0] =
indices[1] = lp_build_const_int32(gallivm, 0);
-
+
for (i = 0; i < vs_type.length; i++) {
inds[i] = lp_build_const_int32(gallivm, i);
io_ptrs[i] = LLVMBuildGEP(builder, io_ptr, &inds[i], 1, "");
int i;
struct gallivm_state *gallivm = variant->gallivm;
struct lp_type f32_type = vs_type;
+ const unsigned pos = draw_current_shader_position_output(variant->llvm->draw);
LLVMTypeRef vs_type_llvm = lp_build_vec_type(gallivm, vs_type);
- LLVMValueRef out3 = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 .. wn*/
- LLVMValueRef const1 = lp_build_const_vec(gallivm, f32_type, 1.0); /*1.0 1.0 1.0 1.0*/
+ 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);
/* 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);
scale_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
out = LLVMBuildFAdd(builder, out, trans, "");
/* store transformed outputs */
- LLVMBuildStore(builder, out, outputs[0][i]);
+ LLVMBuildStore(builder, out, outputs[pos][i]);
}
-
+
}
/**
* Returns clipmask as nxi32 bitmask for the n vertices
*/
-static LLVMValueRef
+static LLVMValueRef
generate_clipmask(struct draw_llvm *llvm,
struct gallivm_state *gallivm,
struct lp_type vs_type,
{
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef mask; /* stores the <nxi32> clipmasks */
- LLVMValueRef test, temp;
+ LLVMValueRef test, temp;
LLVMValueRef zero, shift;
LLVMValueRef pos_x, pos_y, pos_z, pos_w;
LLVMValueRef cv_x, cv_y, cv_z, cv_w;
cd[0] = draw_current_shader_clipdistance_output(llvm->draw, 0);
cd[1] = draw_current_shader_clipdistance_output(llvm->draw, 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 */
/* plane 1 */
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w);
temp = shift;
- test = LLVMBuildAnd(builder, test, temp, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
mask = test;
-
+
/* plane 2 */
test = LLVMBuildFAdd(builder, pos_x, pos_w, "");
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test);
temp = LLVMBuildShl(builder, temp, shift, "");
- test = LLVMBuildAnd(builder, test, temp, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
mask = LLVMBuildOr(builder, mask, test, "");
-
+
/* plane 3 */
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_y, pos_w);
temp = LLVMBuildShl(builder, temp, shift, "");
- test = LLVMBuildAnd(builder, test, temp, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
mask = LLVMBuildOr(builder, mask, test, "");
/* plane 4 */
test = LLVMBuildFAdd(builder, pos_y, pos_w, "");
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test);
temp = LLVMBuildShl(builder, temp, shift, "");
- test = LLVMBuildAnd(builder, test, temp, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
mask = LLVMBuildOr(builder, mask, test, "");
}
if (clip_halfz) {
/* plane 5 */
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, pos_z);
- test = LLVMBuildAnd(builder, test, temp, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
mask = LLVMBuildOr(builder, mask, test, "");
- }
+ }
else {
/* plane 5 */
test = LLVMBuildFAdd(builder, pos_z, pos_w, "");
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test);
- test = LLVMBuildAnd(builder, test, temp, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
mask = LLVMBuildOr(builder, mask, test, "");
}
/* plane 6 */
test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_z, pos_w);
temp = LLVMBuildShl(builder, temp, shift, "");
- test = LLVMBuildAnd(builder, test, temp, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
mask = LLVMBuildOr(builder, mask, test, "");
- }
+ }
if (clip_user) {
LLVMValueRef planes_ptr = draw_jit_context_planes(gallivm, context_ptr);
/**
* Returns boolean if any clipping has occurred
- * Used zero/non-zero i32 value to represent boolean
+ * Used zero/non-zero i32 value to represent boolean
*/
static LLVMValueRef
clipmask_booli32(struct gallivm_state *gallivm,
return ret;
}
+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_indirect,
+ LLVMValueRef vertex_index,
+ LLVMValueRef attrib_index,
+ LLVMValueRef swizzle_index)
+{
+ const struct draw_gs_llvm_iface *gs = draw_gs_llvm_iface(gs_iface);
+ struct gallivm_state *gallivm = bld_base->base.gallivm;
+ LLVMBuilderRef builder = gallivm->builder;
+ LLVMValueRef indices[3];
+ LLVMValueRef res;
+ struct lp_type type = bld_base->base.type;
+
+ if (is_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 = LLVMBuildExtractElement(builder,
+ vertex_index, idx, "");
+ LLVMValueRef channel_vec, value;
+ indices[0] = vert_chan_index;
+ indices[1] = attrib_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;
+}
+
+static void
+draw_gs_llvm_emit_vertex(const struct lp_build_tgsi_gs_iface *gs_base,
+ struct lp_build_tgsi_context * bld_base,
+ LLVMValueRef (*outputs)[4],
+ LLVMValueRef emitted_vertices_vec)
+{
+ const struct draw_gs_llvm_iface *gs_iface = draw_gs_llvm_iface(gs_base);
+ struct draw_gs_llvm_variant *variant = gs_iface->variant;
+ struct gallivm_state *gallivm = variant->gallivm;
+ LLVMBuilderRef builder = gallivm->builder;
+ struct lp_type gs_type = bld_base->base.type;
+ LLVMValueRef clipmask = lp_build_const_int_vec(gallivm,
+ lp_int_type(gs_type), 0);
+ LLVMValueRef indices[LP_MAX_VECTOR_LENGTH];
+ 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;
+
+ for (i = 0; i < gs_type.length; ++i) {
+ LLVMValueRef ind = lp_build_const_int32(gallivm, i);
+ LLVMValueRef currently_emitted =
+ LLVMBuildExtractElement(builder, emitted_vertices_vec, ind, "");
+ indices[i] = LLVMBuildMul(builder, ind, next_prim_offset, "");
+ indices[i] = LLVMBuildAdd(builder, indices[i], currently_emitted, "");
+ }
+
+ convert_to_aos(gallivm, io, indices,
+ outputs, clipmask,
+ gs_info->num_outputs, gs_type,
+ FALSE);
+}
+
+static void
+draw_gs_llvm_end_primitive(const struct lp_build_tgsi_gs_iface *gs_base,
+ struct lp_build_tgsi_context * bld_base,
+ LLVMValueRef verts_per_prim_vec,
+ LLVMValueRef emitted_prims_vec)
+{
+ const struct draw_gs_llvm_iface *gs_iface = draw_gs_llvm_iface(gs_base);
+ struct draw_gs_llvm_variant *variant = gs_iface->variant;
+ struct gallivm_state *gallivm = variant->gallivm;
+ LLVMBuilderRef builder = gallivm->builder;
+ LLVMValueRef prim_lengts_ptr =
+ draw_gs_jit_prim_lengths(variant->gallivm, variant->context_ptr);
+ unsigned i;
+
+ for (i = 0; i < bld_base->base.type.length; ++i) {
+ LLVMValueRef ind = lp_build_const_int32(gallivm, i);
+ LLVMValueRef prims_emitted =
+ LLVMBuildExtractElement(builder, emitted_prims_vec, ind, "");
+ LLVMValueRef store_ptr;
+ LLVMValueRef num_vertices =
+ LLVMBuildExtractElement(builder, verts_per_prim_vec, ind, "");
+
+ store_ptr = LLVMBuildGEP(builder, prim_lengts_ptr, &prims_emitted, 1, "");
+ store_ptr = LLVMBuildLoad(builder, store_ptr, "");
+ store_ptr = LLVMBuildGEP(builder, store_ptr, &ind, 1, "");
+ LLVMBuildStore(builder, num_vertices, store_ptr);
+ }
+}
+
+static void
+draw_gs_llvm_epilogue(const struct lp_build_tgsi_gs_iface *gs_base,
+ struct lp_build_tgsi_context * bld_base,
+ LLVMValueRef total_emitted_vertices_vec,
+ LLVMValueRef emitted_prims_vec)
+{
+ const struct draw_gs_llvm_iface *gs_iface = draw_gs_llvm_iface(gs_base);
+ struct draw_gs_llvm_variant *variant = gs_iface->variant;
+ struct gallivm_state *gallivm = variant->gallivm;
+ LLVMBuilderRef builder = gallivm->builder;
+ LLVMValueRef emitted_verts_ptr =
+ draw_gs_jit_emitted_vertices(gallivm, variant->context_ptr);
+ LLVMValueRef emitted_prims_ptr =
+ draw_gs_jit_emitted_prims(gallivm, variant->context_ptr);
+ LLVMValueRef zero = lp_build_const_int32(gallivm, 0);
+
+ emitted_verts_ptr = LLVMBuildGEP(builder, emitted_verts_ptr, &zero, 0, "");
+ emitted_prims_ptr = LLVMBuildGEP(builder, emitted_prims_ptr, &zero, 0, "");
+
+ LLVMBuildStore(builder, total_emitted_vertices_vec, emitted_verts_ptr);
+ LLVMBuildStore(builder, emitted_prims_vec, emitted_prims_ptr);
+}
static void
draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
struct gallivm_state *gallivm = variant->gallivm;
LLVMContextRef context = gallivm->context;
LLVMTypeRef int32_type = LLVMInt32TypeInContext(context);
- LLVMTypeRef arg_types[8];
+ LLVMTypeRef arg_types[10];
+ unsigned num_arg_types =
+ elts ? Elements(arg_types) : Elements(arg_types) - 1;
LLVMTypeRef func_type;
LLVMValueRef context_ptr;
LLVMBasicBlockRef block;
LLVMBuilderRef builder;
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;
LLVMValueRef stride, step, io_itr;
LLVMValueRef io_ptr, vbuffers_ptr, vb_ptr;
LLVMValueRef zero = lp_build_const_int32(gallivm, 0);
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.
*/
- 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 = gs || key->bypass_viewport;
+ const boolean enable_cliptest = !gs && (key->clip_xy ||
+ key->clip_z ||
+ key->clip_user);
LLVMValueRef variant_func;
const unsigned pos = draw_current_shader_position_output(llvm->draw);
const unsigned cv = draw_current_shader_clipvertex_output(llvm->draw);
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 */
-
- 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 */
+
+ func_type = LLVMFunctionType(int32_type, arg_types, num_arg_types, 0);
variant_func = LLVMAddFunction(gallivm->module,
elts ? "draw_llvm_shader_elts" : "draw_llvm_shader",
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);
+ 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));
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");
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;
}
/* code generated texture sampling */
sampler = draw_llvm_sampler_soa_create(
- draw_llvm_variant_key_samplers(&variant->key),
+ draw_llvm_variant_key_samplers(key),
context_ptr);
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 clipmask; /* holds the clipmask value */
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
#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, "");
+ LLVMValueRef vertex_id;
/* 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");
}
-
- system_values.vertex_id = LLVMBuildInsertElement(gallivm->builder,
- system_values.vertex_id, true_index,
- lp_build_const_int32(gallivm, i), "");
+ /* 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.
+ */
+ vertex_id = LLVMBuildSub(builder, true_index, vertex_id_offset, "");
+
+ system_values.vertex_id = LLVMBuildInsertElement(
+ gallivm->builder,
+ system_values.vertex_id, vertex_id,
+ 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,
context_ptr,
sampler,
- variant->key.clamp_vertex_color);
+ key->clamp_vertex_color);
- if (pos != -1) {
+ 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);
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->clip_xy,
+ key->clip_z,
+ key->clip_user,
+ key->clip_halfz,
+ key->ucp_enable,
context_ptr, &have_clipdist);
temp = LLVMBuildOr(builder, clipmask, temp, "");
/* store temporary clipping boolean value */
clipmask = lp_build_const_int_vec(gallivm, lp_int_type(vs_type), 0);
}
- /* store clipmask in vertex header,
- * original positions in clip
- * and transformed positions in data
- */
- convert_to_aos(gallivm, io, outputs, clipmask,
+ /* store clipmask in vertex header,
+ * original positions in clip
+ * and transformed positions in data
+ */
+ convert_to_aos(gallivm, io, NULL, outputs, clipmask,
vs_info->num_outputs, vs_type,
have_clipdist);
}
-
- lp_build_loop_end_cond(&lp_loop, end, step, LLVMIntUGE);
+ lp_build_loop_end_cond(&lp_loop, count, step, LLVMIntUGE);
sampler->destroy(sampler);
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->clip_halfz = llvm->draw->rasterizer->clip_halfz;
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->num_outputs = draw_total_vs_outputs(llvm->draw);
key->pad1 = 0;
- key->pad2 = 0;
/* All variants of this shader will have the same value for
* nr_samplers. Not yet trying to compact away holes in the
debug_printf("bypass_viewport = %u\n", key->bypass_viewport);
debug_printf("clip_halfz = %u\n", key->clip_halfz);
debug_printf("need_edgeflags = %u\n", key->need_edgeflags);
+ debug_printf("has_gs = %u\n", key->has_gs);
debug_printf("ucp_enable = %u\n", key->ucp_enable);
for (i = 0 ; i < key->nr_vertex_elements; i++) {
void
draw_llvm_set_mapped_texture(struct draw_context *draw,
+ unsigned shader_stage,
unsigned sview_idx,
uint32_t width, uint32_t height, uint32_t depth,
uint32_t first_level, uint32_t last_level,
unsigned j;
struct draw_jit_texture *jit_tex;
- assert(sview_idx < Elements(draw->llvm->jit_context.textures));
+ assert(shader_stage == PIPE_SHADER_VERTEX ||
+ shader_stage == PIPE_SHADER_GEOMETRY);
+
+ if (shader_stage == PIPE_SHADER_VERTEX) {
+ assert(sview_idx < Elements(draw->llvm->jit_context.textures));
- jit_tex = &draw->llvm->jit_context.textures[sview_idx];
+ jit_tex = &draw->llvm->jit_context.textures[sview_idx];
+ } else if (shader_stage == PIPE_SHADER_GEOMETRY) {
+ 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;
jit_tex->height = height;
void
-draw_llvm_set_sampler_state(struct draw_context *draw)
+draw_llvm_set_sampler_state(struct draw_context *draw,
+ unsigned shader_type)
{
unsigned i;
- 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]) {
- const struct pipe_sampler_state *s
- = draw->samplers[PIPE_SHADER_VERTEX][i];
- jit_sam->min_lod = s->min_lod;
- jit_sam->max_lod = s->max_lod;
- jit_sam->lod_bias = s->lod_bias;
- COPY_4V(jit_sam->border_color, s->border_color.f);
+ if (shader_type == PIPE_SHADER_VERTEX) {
+ 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]) {
+ const struct pipe_sampler_state *s
+ = draw->samplers[PIPE_SHADER_VERTEX][i];
+ jit_sam->min_lod = s->min_lod;
+ jit_sam->max_lod = s->max_lod;
+ jit_sam->lod_bias = s->lod_bias;
+ COPY_4V(jit_sam->border_color, s->border_color.f);
+ }
+ }
+ } else if (shader_type == PIPE_SHADER_GEOMETRY) {
+ 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]) {
+ const struct pipe_sampler_state *s
+ = draw->samplers[PIPE_SHADER_GEOMETRY][i];
+ jit_sam->min_lod = s->min_lod;
+ jit_sam->max_lod = s->max_lod;
+ jit_sam->lod_bias = s->lod_bias;
+ COPY_4V(jit_sam->border_color, s->border_color.f);
+ }
}
}
}
llvm->nr_variants--;
FREE(variant);
}
+
+
+/**
+ * Create LLVM types for various structures.
+ */
+static void
+create_gs_jit_types(struct draw_gs_llvm_variant *var)
+{
+ struct gallivm_state *gallivm = var->gallivm;
+ LLVMTypeRef texture_type, sampler_type, context_type;
+
+ texture_type = create_jit_texture_type(gallivm, "texture");
+ sampler_type = create_jit_sampler_type(gallivm, "sampler");
+
+ context_type = create_gs_jit_context_type(gallivm,
+ var->shader->base.vector_length,
+ texture_type, sampler_type,
+ "draw_gs_jit_context");
+ var->context_ptr_type = LLVMPointerType(context_type, 0);
+
+ var->input_array_type = create_gs_jit_input_type(gallivm);
+}
+
+static LLVMTypeRef
+get_gs_context_ptr_type(struct draw_gs_llvm_variant *variant)
+{
+ if (!variant->context_ptr_type)
+ create_gs_jit_types(variant);
+ return variant->context_ptr_type;
+}
+
+static LLVMValueRef
+generate_mask_value(struct draw_gs_llvm_variant *variant,
+ struct lp_type gs_type)
+{
+ struct gallivm_state *gallivm = variant->gallivm;
+ LLVMBuilderRef builder = gallivm->builder;
+ 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;
+
+ 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_GREATER, num_prims, mask_val);
+
+ return mask_val;
+}
+
+static void
+draw_gs_llvm_generate(struct draw_llvm *llvm,
+ struct draw_gs_llvm_variant *variant)
+{
+ struct gallivm_state *gallivm = variant->gallivm;
+ LLVMContextRef context = gallivm->context;
+ LLVMTypeRef int32_type = LLVMInt32TypeInContext(context);
+ LLVMTypeRef arg_types[6];
+ LLVMTypeRef func_type;
+ LLVMValueRef variant_func;
+ LLVMValueRef context_ptr;
+ LLVMValueRef prim_id_ptr;
+ LLVMBasicBlockRef block;
+ LLVMBuilderRef builder;
+ LLVMValueRef io_ptr, input_array, num_prims, mask_val;
+ struct lp_build_sampler_soa *sampler = 0;
+ struct lp_build_context bld;
+ struct lp_bld_tgsi_system_values system_values;
+ 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 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;
+ unsigned vector_length = variant->shader->base.vector_length;
+
+ memset(&system_values, 0, sizeof(system_values));
+
+ assert(variant->vertex_header_ptr_type);
+
+ arg_types[0] = get_gs_context_ptr_type(variant); /* context */
+ arg_types[1] = variant->input_array_type; /* input */
+ arg_types[2] = variant->vertex_header_ptr_type; /* vertex_header */
+ arg_types[3] = int32_type; /* num_prims */
+ arg_types[4] = int32_type; /* instance_id */
+ arg_types[5] = LLVMPointerType(
+ LLVMVectorType(int32_type, vector_length), 0); /* prim_id_ptr */
+
+ func_type = LLVMFunctionType(int32_type, arg_types, Elements(arg_types), 0);
+
+ variant_func = LLVMAddFunction(gallivm->module, "draw_geometry_shader",
+ func_type);
+ variant->function = variant_func;
+
+ LLVMSetFunctionCallConv(variant_func, LLVMCCallConv);
+
+ for (i = 0; i < Elements(arg_types); ++i)
+ if (LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
+ LLVMAddAttribute(LLVMGetParam(variant_func, i),
+ LLVMNoAliasAttribute);
+
+ context_ptr = LLVMGetParam(variant_func, 0);
+ input_array = LLVMGetParam(variant_func, 1);
+ io_ptr = LLVMGetParam(variant_func, 2);
+ num_prims = LLVMGetParam(variant_func, 3);
+ system_values.instance_id = LLVMGetParam(variant_func, 4);
+ prim_id_ptr = LLVMGetParam(variant_func, 5);
+
+ lp_build_name(context_ptr, "context");
+ lp_build_name(input_array, "input");
+ lp_build_name(io_ptr, "io");
+ 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");
+
+ variant->context_ptr = context_ptr;
+ variant->io_ptr = io_ptr;
+ variant->num_prims = num_prims;
+
+ gs_iface.base.fetch_input = draw_gs_llvm_fetch_input;
+ gs_iface.base.emit_vertex = draw_gs_llvm_emit_vertex;
+ gs_iface.base.end_primitive = draw_gs_llvm_end_primitive;
+ gs_iface.base.gs_epilogue = draw_gs_llvm_epilogue;
+ gs_iface.input = input_array;
+ gs_iface.variant = variant;
+
+ /*
+ * Function body
+ */
+
+ block = LLVMAppendBasicBlockInContext(gallivm->context, variant_func, "entry");
+ builder = gallivm->builder;
+ LLVMPositionBuilderAtEnd(builder, block);
+
+ lp_build_context_init(&bld, gallivm, lp_type_int(32));
+
+ memset(&gs_type, 0, sizeof gs_type);
+ gs_type.floating = TRUE; /* floating point values */
+ gs_type.sign = TRUE; /* values are signed */
+ gs_type.norm = FALSE; /* values are not limited to [0,1] or [-1,1] */
+ gs_type.width = 32; /* 32-bit float */
+ gs_type.length = vector_length;
+
+ consts_ptr = draw_gs_jit_context_constants(variant->gallivm, context_ptr);
+
+ /* code generated texture sampling */
+ sampler = draw_llvm_sampler_soa_create(variant->key.samplers,
+ context_ptr);
+
+ 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");;
+ }
+
+ 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,
+ tokens,
+ gs_type,
+ &mask,
+ consts_ptr,
+ &system_values,
+ NULL,
+ outputs,
+ sampler,
+ &llvm->draw->gs.geometry_shader->info,
+ (const struct lp_build_tgsi_gs_iface *)&gs_iface);
+
+ sampler->destroy(sampler);
+
+ lp_build_mask_end(&mask);
+
+ LLVMBuildRet(builder, lp_build_zero(gallivm, lp_type_uint(32)));
+
+ gallivm_verify_function(gallivm, variant_func);
+}
+
+
+struct draw_gs_llvm_variant *
+draw_gs_llvm_create_variant(struct draw_llvm *llvm,
+ unsigned num_outputs,
+ const struct draw_gs_llvm_variant_key *key)
+{
+ struct draw_gs_llvm_variant *variant;
+ struct llvm_geometry_shader *shader =
+ llvm_geometry_shader(llvm->draw->gs.geometry_shader);
+ LLVMTypeRef vertex_header;
+
+ variant = MALLOC(sizeof *variant +
+ shader->variant_key_size -
+ sizeof variant->key);
+ if (variant == NULL)
+ return NULL;
+
+ variant->llvm = llvm;
+ variant->shader = shader;
+
+ variant->gallivm = gallivm_create();
+
+ create_gs_jit_types(variant);
+
+ memcpy(&variant->key, key, shader->variant_key_size);
+
+ vertex_header = create_jit_vertex_header(variant->gallivm, num_outputs);
+
+ variant->vertex_header_ptr_type = LLVMPointerType(vertex_header, 0);
+
+ draw_gs_llvm_generate(llvm, variant);
+
+ gallivm_compile_module(variant->gallivm);
+
+ variant->jit_func = (draw_gs_jit_func)
+ gallivm_jit_function(variant->gallivm, variant->function);
+
+ variant->list_item_global.base = variant;
+ variant->list_item_local.base = variant;
+ /*variant->no = */shader->variants_created++;
+ variant->list_item_global.base = variant;
+
+ return variant;
+}
+
+void
+draw_gs_llvm_destroy_variant(struct draw_gs_llvm_variant *variant)
+{
+ 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);
+ variant->shader->variants_cached--;
+ remove_from_list(&variant->list_item_global);
+ llvm->nr_gs_variants--;
+ FREE(variant);
+}
+
+struct draw_gs_llvm_variant_key *
+draw_gs_llvm_make_variant_key(struct draw_llvm *llvm, char *store)
+{
+ unsigned i;
+ struct draw_gs_llvm_variant_key *key;
+ struct draw_sampler_static_state *draw_sampler;
+
+ key = (struct draw_gs_llvm_variant_key *)store;
+
+ 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.
+ */
+ key->nr_samplers = llvm->draw->gs.geometry_shader->info.file_max[TGSI_FILE_SAMPLER] + 1;
+ if (llvm->draw->gs.geometry_shader->info.file_max[TGSI_FILE_SAMPLER_VIEW] != -1) {
+ key->nr_sampler_views =
+ llvm->draw->gs.geometry_shader->info.file_max[TGSI_FILE_SAMPLER_VIEW] + 1;
+ }
+ else {
+ key->nr_sampler_views = key->nr_samplers;
+ }
+
+ draw_sampler = key->samplers;
+
+ memset(draw_sampler, 0, MAX2(key->nr_samplers, key->nr_sampler_views) * sizeof *draw_sampler);
+
+ for (i = 0 ; i < key->nr_samplers; i++) {
+ lp_sampler_static_sampler_state(&draw_sampler[i].sampler_state,
+ llvm->draw->samplers[PIPE_SHADER_GEOMETRY][i]);
+ }
+ for (i = 0 ; i < key->nr_sampler_views; i++) {
+ lp_sampler_static_texture_state(&draw_sampler[i].texture_state,
+ llvm->draw->sampler_views[PIPE_SHADER_GEOMETRY][i]);
+ }
+
+ return key;
+}
+
+void
+draw_gs_llvm_dump_variant_key(struct draw_gs_llvm_variant_key *key)
+{
+ unsigned i;
+ struct draw_sampler_static_state *sampler = key->samplers;
+
+ for (i = 0 ; i < key->nr_sampler_views; i++) {
+ debug_printf("sampler[%i].src_format = %s\n", i,
+ util_format_name(sampler[i].texture_state.format));
+ }
+}