#include "util/u_debug.h"
#include "util/u_math.h"
#include "util/u_memory.h"
+#include "util/u_prim.h"
#include "tgsi/tgsi_dump.h"
#include "tgsi/tgsi_exec.h"
#include "tgsi/tgsi_info.h"
#include "lp_bld_gather.h"
#include "lp_bld_init.h"
#include "lp_bld_logic.h"
+#include "lp_bld_misc.h"
#include "lp_bld_swizzle.h"
#include "lp_bld_flow.h"
#include "lp_bld_quad.h"
{
char buf[32];
- util_snprintf(buf, sizeof buf, " %s[%u].%c = ",
- tgsi_file_name(file),
- index, "xyzw"[chan]);
+ snprintf(buf, sizeof buf, " %s[%u].%c = ",
+ tgsi_file_name(file),
+ index, "xyzw"[chan]);
lp_build_print_value(gallivm, buf, value);
}
return FALSE;
}
+/*
+ * combine the execution mask if there is one with the current mask.
+ */
+static LLVMValueRef
+mask_vec(struct lp_build_tgsi_context *bld_base)
+{
+ struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
+ LLVMBuilderRef builder = bld->bld_base.base.gallivm->builder;
+ struct lp_exec_mask *exec_mask = &bld->exec_mask;
+ LLVMValueRef bld_mask = bld->mask ? lp_build_mask_value(bld->mask) : NULL;
+ if (!exec_mask->has_mask) {
+ return bld_mask;
+ }
+ if (!bld_mask)
+ return exec_mask->exec_mask;
+ return LLVMBuildAnd(builder, lp_build_mask_value(bld->mask),
+ exec_mask->exec_mask, "");
+}
+
/*
* Returns true if we're inside a switch statement.
* It's global, meaning that it returns true even if there's
exec_mask, "break_full");
}
else {
- unsigned opcode = bld_base->instructions[bld_base->pc + 1].Instruction.Opcode;
+ enum tgsi_opcode opcode =
+ bld_base->instructions[bld_base->pc + 1].Instruction.Opcode;
boolean break_always = (opcode == TGSI_OPCODE_ENDSWITCH ||
opcode == TGSI_OPCODE_CASE);
lp_exec_mask_update(mask);
}
-static void lp_exec_break_condition(struct lp_exec_mask *mask,
- LLVMValueRef cond)
-{
- LLVMBuilderRef builder = mask->bld->gallivm->builder;
- struct function_ctx *ctx = func_ctx(mask);
- LLVMValueRef cond_mask = LLVMBuildAnd(builder,
- mask->exec_mask,
- cond, "cond_mask");
- cond_mask = LLVMBuildNot(builder, cond_mask, "break_cond");
-
- if (ctx->break_type == LP_EXEC_MASK_BREAK_TYPE_LOOP) {
- mask->break_mask = LLVMBuildAnd(builder,
- mask->break_mask,
- cond_mask, "breakc_full");
- }
- else {
- mask->switch_mask = LLVMBuildAnd(builder,
- mask->switch_mask,
- cond_mask, "breakc_switch");
- }
-
- lp_exec_mask_update(mask);
-}
-
static void lp_exec_continue(struct lp_exec_mask *mask)
{
LLVMBuilderRef builder = mask->bld->gallivm->builder;
{
unsigned pc = bld_base->pc;
struct function_ctx *ctx = func_ctx(mask);
- unsigned curr_switch_stack = ctx->switch_stack_size;
+ int curr_switch_stack = ctx->switch_stack_size;
if (ctx->switch_stack_size > LP_MAX_TGSI_NESTING) {
return false;
pc++;
}
- while (pc != -1 && pc < bld_base->num_instructions) {
- unsigned opcode = bld_base->instructions[pc].Instruction.Opcode;
+ while (pc != ~0u && pc < bld_base->num_instructions) {
+ enum tgsi_opcode opcode = bld_base->instructions[pc].Instruction.Opcode;
switch (opcode) {
case TGSI_OPCODE_CASE:
if (curr_switch_stack == ctx->switch_stack_size) {
}
curr_switch_stack--;
break;
+ default:
+ ; /* nothing */
}
pc++;
}
* which just gets rid of all case statements appearing together with
* default (or could do switch analysis at switch start time instead).
*/
- unsigned opcode = bld_base->instructions[bld_base->pc - 1].Instruction.Opcode;
+ enum tgsi_opcode opcode =
+ bld_base->instructions[bld_base->pc - 1].Instruction.Opcode;
boolean ft_into = (opcode != TGSI_OPCODE_BRK &&
opcode != TGSI_OPCODE_SWITCH);
/*
*/
static void lp_exec_mask_store(struct lp_exec_mask *mask,
struct lp_build_context *bld_store,
- LLVMValueRef pred,
LLVMValueRef val,
LLVMValueRef dst_ptr)
{
LLVMBuilderRef builder = mask->bld->gallivm->builder;
+ LLVMValueRef exec_mask = mask->has_mask ? mask->exec_mask : NULL;
assert(lp_check_value(bld_store->type, val));
assert(LLVMGetTypeKind(LLVMTypeOf(dst_ptr)) == LLVMPointerTypeKind);
- assert(LLVMGetElementType(LLVMTypeOf(dst_ptr)) == LLVMTypeOf(val));
+ assert(LLVMGetElementType(LLVMTypeOf(dst_ptr)) == LLVMTypeOf(val) ||
+ LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(dst_ptr))) == LLVMArrayTypeKind);
- /* Mix the predicate and execution mask */
- if (mask->has_mask) {
- if (pred) {
- pred = LLVMBuildAnd(builder, pred, mask->exec_mask, "");
- } else {
- pred = mask->exec_mask;
- }
- }
-
- if (pred) {
+ if (exec_mask) {
LLVMValueRef res, dst;
dst = LLVMBuildLoad(builder, dst_ptr, "");
- res = lp_build_select(bld_store, pred, val, dst);
+ res = lp_build_select(bld_store, exec_mask, val, dst);
LLVMBuildStore(builder, res, dst_ptr);
} else
LLVMBuildStore(builder, val, dst_ptr);
static LLVMValueRef
get_file_ptr(struct lp_build_tgsi_soa_context *bld,
unsigned file,
- unsigned index,
+ int index,
unsigned chan)
{
LLVMBuilderRef builder = bld->bld_base.base.gallivm->builder;
if (bld->indirect_files & (1 << file)) {
LLVMValueRef lindex = lp_build_const_int32(bld->bld_base.base.gallivm, index * 4 + chan);
- return LLVMBuildGEP(builder, var_of_array, &lindex, 1, "");
+ if (LLVMGetTypeKind(LLVMGetElementType(LLVMTypeOf(var_of_array))) == LLVMArrayTypeKind) {
+ LLVMValueRef gep[2];
+ gep[0] = lp_build_const_int32(bld->bld_base.base.gallivm, 0);
+ gep[1] = lindex;
+ return LLVMBuildGEP(builder, var_of_array, gep, 2, "");
+ } else {
+ return LLVMBuildGEP(builder, var_of_array, &lindex, 1, "");
+ }
}
else {
assert(index <= bld->bld_base.info->file_max[file]);
LLVMValueRef base_ptr,
LLVMValueRef indexes,
LLVMValueRef values,
- struct lp_exec_mask *mask,
- LLVMValueRef pred)
+ struct lp_exec_mask *mask)
{
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
unsigned i;
-
- /* Mix the predicate and execution mask */
- if (mask->has_mask) {
- if (pred) {
- pred = LLVMBuildAnd(builder, pred, mask->exec_mask, "");
- }
- else {
- pred = mask->exec_mask;
- }
- }
+ LLVMValueRef pred = mask->has_mask ? mask->exec_mask : NULL;
/*
* Loop over elements of index_vec, store scalar value.
static LLVMValueRef
get_indirect_index(struct lp_build_tgsi_soa_context *bld,
unsigned reg_file, unsigned reg_index,
- const struct tgsi_ind_register *indirect_reg)
+ const struct tgsi_ind_register *indirect_reg,
+ int index_limit)
{
LLVMBuilderRef builder = bld->bld_base.base.gallivm->builder;
struct lp_build_context *uint_bld = &bld->bld_base.uint_bld;
* larger than the declared size but smaller than the buffer size.
*/
if (reg_file != TGSI_FILE_CONSTANT) {
+ assert(index_limit >= 0);
max_index = lp_build_const_int_vec(bld->bld_base.base.gallivm,
- uint_bld->type,
- bld->bld_base.info->file_max[reg_file]);
+ uint_bld->type, index_limit);
assert(!uint_bld->type.sign);
index = lp_build_min(uint_bld, index, max_index);
case TGSI_TYPE_DOUBLE:
bld_fetch = &bld_base->dbl_bld;
break;
+ case TGSI_TYPE_UNSIGNED64:
+ bld_fetch = &bld_base->uint64_bld;
+ break;
+ case TGSI_TYPE_SIGNED64:
+ bld_fetch = &bld_base->int64_bld;
+ break;
case TGSI_TYPE_VOID:
default:
assert(0);
if (need_perelement_offset) {
LLVMValueRef pixel_offsets;
- int i;
+ unsigned i;
/* build pixel offset vector: {0, 1, 2, 3, ...} */
pixel_offsets = uint_bld->undef;
for (i = 0; i < uint_bld->type.length; i++) {
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
LLVMValueRef consts_ptr;
LLVMValueRef num_consts;
LLVMValueRef res;
+ unsigned swizzle = swizzle_in & 0xffff;
/* XXX: Handle fetching xyzw components as a vector */
- assert(swizzle != ~0);
+ assert(swizzle != ~0u);
if (reg->Register.Dimension) {
assert(!reg->Dimension.Indirect);
indirect_index = get_indirect_index(bld,
reg->Register.File,
reg->Register.Index,
- ®->Indirect);
+ ®->Indirect,
+ bld->bld_base.info->file_max[reg->Register.File]);
/* All fetches are from the same constant buffer, so
* we need to propagate the size to a vector to do a
index_vec = lp_build_shl_imm(uint_bld, indirect_index, 2);
index_vec = lp_build_add(uint_bld, index_vec, swizzle_vec);
- if (stype == TGSI_TYPE_DOUBLE) {
+ if (tgsi_type_is_64bit(stype)) {
LLVMValueRef swizzle_vec2;
- swizzle_vec2 = lp_build_const_int_vec(gallivm, uint_bld->type, swizzle + 1);
+ swizzle_vec2 = lp_build_const_int_vec(gallivm, uint_bld->type, swizzle_in >> 16);
index_vec2 = lp_build_shl_imm(uint_bld, indirect_index, 2);
index_vec2 = lp_build_add(uint_bld, index_vec2, swizzle_vec2);
}
scalar_ptr = LLVMBuildGEP(builder, consts_ptr,
&index, 1, "");
- if (stype == TGSI_TYPE_DOUBLE) {
- LLVMTypeRef dptr_type = LLVMPointerType(LLVMDoubleTypeInContext(gallivm->context), 0);
- scalar_ptr = LLVMBuildBitCast(builder, scalar_ptr, dptr_type, "");
- bld_broad = &bld_base->dbl_bld;
+
+ if (tgsi_type_is_64bit(stype) && ((swizzle_in >> 16) != swizzle + 1)) {
+
+ LLVMValueRef scalar2, scalar2_ptr;
+ LLVMValueRef shuffles[2];
+ index = lp_build_const_int32(gallivm, reg->Register.Index * 4 + (swizzle_in >> 16));
+
+ scalar2_ptr = LLVMBuildGEP(builder, consts_ptr,
+ &index, 1, "");
+
+ scalar = LLVMBuildLoad(builder, scalar_ptr, "");
+ scalar2 = LLVMBuildLoad(builder, scalar2_ptr, "");
+ shuffles[0] = lp_build_const_int32(gallivm, 0);
+ shuffles[1] = lp_build_const_int32(gallivm, 1);
+
+ res = LLVMGetUndef(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), bld_base->base.type.length * 2));
+ res = LLVMBuildInsertElement(builder, res, scalar, shuffles[0], "");
+ res = LLVMBuildInsertElement(builder, res, scalar2, shuffles[1], "");
+ } else {
+ if (stype == TGSI_TYPE_DOUBLE) {
+ LLVMTypeRef dptr_type = LLVMPointerType(LLVMDoubleTypeInContext(gallivm->context), 0);
+ scalar_ptr = LLVMBuildBitCast(builder, scalar_ptr, dptr_type, "");
+ bld_broad = &bld_base->dbl_bld;
+ } else if (stype == TGSI_TYPE_UNSIGNED64) {
+ LLVMTypeRef u64ptr_type = LLVMPointerType(LLVMInt64TypeInContext(gallivm->context), 0);
+ scalar_ptr = LLVMBuildBitCast(builder, scalar_ptr, u64ptr_type, "");
+ bld_broad = &bld_base->uint64_bld;
+ } else if (stype == TGSI_TYPE_SIGNED64) {
+ LLVMTypeRef i64ptr_type = LLVMPointerType(LLVMInt64TypeInContext(gallivm->context), 0);
+ scalar_ptr = LLVMBuildBitCast(builder, scalar_ptr, i64ptr_type, "");
+ bld_broad = &bld_base->int64_bld;
+ }
+ scalar = LLVMBuildLoad(builder, scalar_ptr, "");
+ res = lp_build_broadcast_scalar(bld_broad, scalar);
}
- scalar = LLVMBuildLoad(builder, scalar_ptr, "");
- res = lp_build_broadcast_scalar(bld_broad, scalar);
+
}
- if (stype == TGSI_TYPE_SIGNED || stype == TGSI_TYPE_UNSIGNED || stype == TGSI_TYPE_DOUBLE) {
+ if (stype == TGSI_TYPE_SIGNED || stype == TGSI_TYPE_UNSIGNED || stype == TGSI_TYPE_DOUBLE || stype == TGSI_TYPE_SIGNED64 || stype == TGSI_TYPE_UNSIGNED64) {
struct lp_build_context *bld_fetch = stype_to_fetch(bld_base, stype);
res = LLVMBuildBitCast(builder, res, bld_fetch->vec_type, "");
}
}
/**
- * Fetch double values from two separate channels.
- * Doubles are stored split across two channels, like xy and zw.
- * This function creates a set of 16 floats,
+ * Fetch 64-bit values from two separate channels.
+ * 64-bit values are stored split across two channels, like xy and zw.
+ * This function creates a set of vec_length*2 floats,
* extracts the values from the two channels,
- * puts them in the correct place, then casts to 8 doubles.
+ * puts them in the correct place, then casts to vec_length 64-bits.
*/
static LLVMValueRef
-emit_fetch_double(
+emit_fetch_64bit(
struct lp_build_tgsi_context * bld_base,
enum tgsi_opcode_type stype,
LLVMValueRef input,
LLVMValueRef res;
struct lp_build_context *bld_fetch = stype_to_fetch(bld_base, stype);
int i;
- LLVMValueRef shuffles[16];
+ LLVMValueRef shuffles[2 * (LP_MAX_VECTOR_WIDTH/32)];
int len = bld_base->base.type.length * 2;
- assert(len <= 16);
+ assert(len <= (2 * (LP_MAX_VECTOR_WIDTH/32)));
for (i = 0; i < bld_base->base.type.length * 2; i+=2) {
shuffles[i] = lp_build_const_int32(gallivm, i / 2);
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef res = NULL;
+ unsigned swizzle = swizzle_in & 0xffff;
if (bld->use_immediates_array || reg->Register.Indirect) {
LLVMValueRef imms_array;
indirect_index = get_indirect_index(bld,
reg->Register.File,
reg->Register.Index,
- ®->Indirect);
+ ®->Indirect,
+ bld->bld_base.info->file_max[reg->Register.File]);
/*
* Unlike for other reg classes, adding pixel offsets is unnecessary -
* immediates are stored as full vectors (FIXME??? - might be better
indirect_index,
swizzle,
FALSE);
- if (stype == TGSI_TYPE_DOUBLE)
+ if (tgsi_type_is_64bit(stype))
index_vec2 = get_soa_array_offsets(&bld_base->uint_bld,
indirect_index,
- swizzle + 1,
+ swizzle_in >> 16,
FALSE);
/* Gather values from the immediate register array */
res = build_gather(bld_base, imms_array, index_vec, NULL, index_vec2);
} else {
- LLVMValueRef lindex = lp_build_const_int32(gallivm,
- reg->Register.Index * 4 + swizzle);
- LLVMValueRef imms_ptr = LLVMBuildGEP(builder,
- bld->imms_array, &lindex, 1, "");
+ LLVMValueRef gep[2];
+ gep[0] = lp_build_const_int32(gallivm, 0);
+ gep[1] = lp_build_const_int32(gallivm, reg->Register.Index * 4 + swizzle);
+ LLVMValueRef imms_ptr = LLVMBuildGEP(builder,
+ bld->imms_array, gep, 2, "");
res = LLVMBuildLoad(builder, imms_ptr, "");
- if (stype == TGSI_TYPE_DOUBLE) {
- LLVMValueRef lindex1;
+ if (tgsi_type_is_64bit(stype)) {
LLVMValueRef imms_ptr2;
LLVMValueRef res2;
-
- lindex1 = lp_build_const_int32(gallivm,
- reg->Register.Index * 4 + swizzle + 1);
+ gep[1] = lp_build_const_int32(gallivm,
+ reg->Register.Index * 4 + (swizzle_in >> 16));
imms_ptr2 = LLVMBuildGEP(builder,
- bld->imms_array, &lindex1, 1, "");
+ bld->imms_array, gep, 2, "");
res2 = LLVMBuildLoad(builder, imms_ptr2, "");
- res = emit_fetch_double(bld_base, stype, res, res2);
+ res = emit_fetch_64bit(bld_base, stype, res, res2);
}
}
}
else {
res = bld->immediates[reg->Register.Index][swizzle];
- if (stype == TGSI_TYPE_DOUBLE)
- res = emit_fetch_double(bld_base, stype, res, bld->immediates[reg->Register.Index][swizzle + 1]);
+ if (tgsi_type_is_64bit(stype))
+ res = emit_fetch_64bit(bld_base, stype, res, bld->immediates[reg->Register.Index][swizzle_in >> 16]);
}
- if (stype == TGSI_TYPE_UNSIGNED) {
- res = LLVMBuildBitCast(builder, res, bld_base->uint_bld.vec_type, "");
- } else if (stype == TGSI_TYPE_SIGNED) {
- res = LLVMBuildBitCast(builder, res, bld_base->int_bld.vec_type, "");
- } else if (stype == TGSI_TYPE_DOUBLE) {
- res = LLVMBuildBitCast(builder, res, bld_base->dbl_bld.vec_type, "");
+ if (stype == TGSI_TYPE_SIGNED || stype == TGSI_TYPE_UNSIGNED || tgsi_type_is_64bit(stype)) {
+ struct lp_build_context *bld_fetch = stype_to_fetch(bld_base, stype);
+ res = LLVMBuildBitCast(builder, res, bld_fetch->vec_type, "");
}
return res;
}
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef res;
+ unsigned swizzle = swizzle_in & 0xffff;
if (reg->Register.Indirect) {
LLVMValueRef indirect_index;
indirect_index = get_indirect_index(bld,
reg->Register.File,
reg->Register.Index,
- ®->Indirect);
+ ®->Indirect,
+ bld->bld_base.info->file_max[reg->Register.File]);
index_vec = get_soa_array_offsets(&bld_base->uint_bld,
indirect_index,
swizzle,
TRUE);
- if (stype == TGSI_TYPE_DOUBLE) {
+ if (tgsi_type_is_64bit(stype)) {
index_vec2 = get_soa_array_offsets(&bld_base->uint_bld,
indirect_index,
- swizzle + 1,
+ swizzle_in >> 16,
TRUE);
}
/* cast inputs_array pointer to float* */
bld->inputs_array, &lindex, 1, "");
res = LLVMBuildLoad(builder, input_ptr, "");
- if (stype == TGSI_TYPE_DOUBLE) {
+ if (tgsi_type_is_64bit(stype)) {
LLVMValueRef lindex1;
LLVMValueRef input_ptr2;
LLVMValueRef res2;
lindex1 = lp_build_const_int32(gallivm,
- reg->Register.Index * 4 + swizzle + 1);
+ reg->Register.Index * 4 + (swizzle_in >> 16));
input_ptr2 = LLVMBuildGEP(builder,
bld->inputs_array, &lindex1, 1, "");
res2 = LLVMBuildLoad(builder, input_ptr2, "");
- res = emit_fetch_double(bld_base, stype, res, res2);
+ res = emit_fetch_64bit(bld_base, stype, res, res2);
}
}
else {
res = bld->inputs[reg->Register.Index][swizzle];
- if (stype == TGSI_TYPE_DOUBLE)
- res = emit_fetch_double(bld_base, stype, res, bld->inputs[reg->Register.Index][swizzle + 1]);
+ if (tgsi_type_is_64bit(stype))
+ res = emit_fetch_64bit(bld_base, stype, res, bld->inputs[reg->Register.Index][swizzle_in >> 16]);
}
}
assert(res);
- if (stype == TGSI_TYPE_UNSIGNED) {
- res = LLVMBuildBitCast(builder, res, bld_base->uint_bld.vec_type, "");
- } else if (stype == TGSI_TYPE_SIGNED) {
- res = LLVMBuildBitCast(builder, res, bld_base->int_bld.vec_type, "");
- } else if (stype == TGSI_TYPE_DOUBLE) {
- res = LLVMBuildBitCast(builder, res, bld_base->dbl_bld.vec_type, "");
+ if (stype == TGSI_TYPE_SIGNED || stype == TGSI_TYPE_UNSIGNED || tgsi_type_is_64bit(stype)) {
+ struct lp_build_context *bld_fetch = stype_to_fetch(bld_base, stype);
+ res = LLVMBuildBitCast(builder, res, bld_fetch->vec_type, "");
}
return res;
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef attrib_index = NULL;
LLVMValueRef vertex_index = NULL;
+ unsigned swizzle = swizzle_in & 0xffff;
LLVMValueRef swizzle_index = lp_build_const_int32(gallivm, swizzle);
LLVMValueRef res;
}
if (reg->Register.Indirect) {
+ /*
+ * XXX: this is possibly not quite the right value, since file_max may be
+ * larger than the max attrib index, due to it being the max of declared
+ * inputs AND the max vertices per prim (which is 6 for tri adj).
+ * It should however be safe to use (since we always allocate
+ * PIPE_MAX_SHADER_INPUTS (80) for it, which is overallocated quite a bit).
+ */
+ int index_limit = info->file_max[reg->Register.File];
attrib_index = get_indirect_index(bld,
reg->Register.File,
reg->Register.Index,
- ®->Indirect);
+ ®->Indirect,
+ index_limit);
} else {
attrib_index = lp_build_const_int32(gallivm, reg->Register.Index);
}
if (reg->Dimension.Indirect) {
+ /*
+ * A fixed 6 should do as well (which is what we allocate).
+ */
+ int index_limit = u_vertices_per_prim(info->properties[TGSI_PROPERTY_GS_INPUT_PRIM]);
vertex_index = get_indirect_index(bld,
reg->Register.File,
reg->Dimension.Index,
- ®->DimIndirect);
+ ®->DimIndirect,
+ index_limit);
} else {
vertex_index = lp_build_const_int32(gallivm, reg->Dimension.Index);
}
swizzle_index);
assert(res);
-
- if (stype == TGSI_TYPE_UNSIGNED) {
+ if (tgsi_type_is_64bit(stype)) {
+ LLVMValueRef swizzle_index = lp_build_const_int32(gallivm, swizzle_in >> 16);
+ LLVMValueRef res2;
+ res2 = bld->gs_iface->fetch_input(bld->gs_iface, bld_base,
+ reg->Dimension.Indirect,
+ vertex_index,
+ reg->Register.Indirect,
+ attrib_index,
+ swizzle_index);
+ assert(res2);
+ res = emit_fetch_64bit(bld_base, stype, res, res2);
+ } else if (stype == TGSI_TYPE_UNSIGNED) {
res = LLVMBuildBitCast(builder, res, bld_base->uint_bld.vec_type, "");
} else if (stype == TGSI_TYPE_SIGNED) {
res = LLVMBuildBitCast(builder, res, bld_base->int_bld.vec_type, "");
- } else if (stype == TGSI_TYPE_DOUBLE) {
- res = LLVMBuildBitCast(builder, res, bld_base->dbl_bld.vec_type, "");
}
return res;
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef res;
+ unsigned swizzle = swizzle_in & 0xffff;
if (reg->Register.Indirect) {
LLVMValueRef indirect_index;
indirect_index = get_indirect_index(bld,
reg->Register.File,
reg->Register.Index,
- ®->Indirect);
+ ®->Indirect,
+ bld->bld_base.info->file_max[reg->Register.File]);
index_vec = get_soa_array_offsets(&bld_base->uint_bld,
indirect_index,
swizzle,
TRUE);
- if (stype == TGSI_TYPE_DOUBLE) {
+ if (tgsi_type_is_64bit(stype)) {
index_vec2 = get_soa_array_offsets(&bld_base->uint_bld,
indirect_index,
- swizzle + 1,
+ swizzle_in >> 16,
TRUE);
}
temp_ptr = lp_get_temp_ptr_soa(bld, reg->Register.Index, swizzle);
res = LLVMBuildLoad(builder, temp_ptr, "");
- if (stype == TGSI_TYPE_DOUBLE) {
+ if (tgsi_type_is_64bit(stype)) {
LLVMValueRef temp_ptr2, res2;
- temp_ptr2 = lp_get_temp_ptr_soa(bld, reg->Register.Index, swizzle + 1);
+ temp_ptr2 = lp_get_temp_ptr_soa(bld, reg->Register.Index, swizzle_in >> 16);
res2 = LLVMBuildLoad(builder, temp_ptr2, "");
- res = emit_fetch_double(bld_base, stype, res, res2);
+ res = emit_fetch_64bit(bld_base, stype, res, res2);
}
}
- if (stype == TGSI_TYPE_SIGNED || stype == TGSI_TYPE_UNSIGNED || stype == TGSI_TYPE_DOUBLE) {
+ if (stype == TGSI_TYPE_SIGNED ||
+ stype == TGSI_TYPE_UNSIGNED ||
+ stype == TGSI_TYPE_DOUBLE ||
+ stype == TGSI_TYPE_SIGNED64 ||
+ stype == TGSI_TYPE_UNSIGNED64) {
struct lp_build_context *bld_fetch = stype_to_fetch(bld_base, stype);
res = LLVMBuildBitCast(builder, res, bld_fetch->vec_type, "");
}
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMValueRef *ddx,
LLVMValueRef *ddy)
{
- if(res)
+ if (res)
*res = src;
/* TODO: use interpolation coeffs for inputs */
- if(ddx)
+ if (ddx)
*ddx = lp_build_ddx(&bld->bld_base.base, src);
- if(ddy)
+ if (ddy)
*ddy = lp_build_ddy(&bld->bld_base.base, src);
}
-
/**
- * Predicate.
- */
-static void
-emit_fetch_predicate(
- struct lp_build_tgsi_soa_context *bld,
- const struct tgsi_full_instruction *inst,
- LLVMValueRef *pred)
-{
- LLVMBuilderRef builder = bld->bld_base.base.gallivm->builder;
- unsigned index;
- unsigned char swizzles[4];
- LLVMValueRef unswizzled[4] = {NULL, NULL, NULL, NULL};
- LLVMValueRef value;
- unsigned chan;
-
- if (!inst->Instruction.Predicate) {
- TGSI_FOR_EACH_CHANNEL( chan ) {
- pred[chan] = NULL;
- }
- return;
- }
-
- swizzles[0] = inst->Predicate.SwizzleX;
- swizzles[1] = inst->Predicate.SwizzleY;
- swizzles[2] = inst->Predicate.SwizzleZ;
- swizzles[3] = inst->Predicate.SwizzleW;
-
- index = inst->Predicate.Index;
- assert(index < LP_MAX_TGSI_PREDS);
-
- TGSI_FOR_EACH_CHANNEL( chan ) {
- unsigned swizzle = swizzles[chan];
-
- /*
- * Only fetch the predicate register channels that are actually listed
- * in the swizzles
- */
- if (!unswizzled[swizzle]) {
- value = LLVMBuildLoad(builder,
- bld->preds[index][swizzle], "");
-
- /*
- * Convert the value to an integer mask.
- *
- * TODO: Short-circuit this comparison -- a D3D setp_xx instructions
- * is needlessly causing two comparisons due to storing the intermediate
- * result as float vector instead of an integer mask vector.
- */
- value = lp_build_compare(bld->bld_base.base.gallivm,
- bld->bld_base.base.type,
- PIPE_FUNC_NOTEQUAL,
- value,
- bld->bld_base.base.zero);
- if (inst->Predicate.Negate) {
- value = LLVMBuildNot(builder, value, "");
- }
-
- unswizzled[swizzle] = value;
- } else {
- value = unswizzled[swizzle];
- }
-
- pred[chan] = value;
- }
-}
-
-/**
- * store an array of 8 doubles into two arrays of 8 floats
+ * store an array of vec-length 64-bit into two arrays of vec_length floats
* i.e.
* value is d0, d1, d2, d3 etc.
- * each double has high and low pieces x, y
+ * each 64-bit has high and low pieces x, y
* so gets stored into the separate channels as:
* chan_ptr = d0.x, d1.x, d2.x, d3.x
* chan_ptr2 = d0.y, d1.y, d2.y, d3.y
*/
static void
-emit_store_double_chan(struct lp_build_tgsi_context *bld_base,
- int dtype,
- LLVMValueRef chan_ptr, LLVMValueRef chan_ptr2,
- LLVMValueRef pred,
- LLVMValueRef value)
+emit_store_64bit_chan(struct lp_build_tgsi_context *bld_base,
+ LLVMValueRef chan_ptr, LLVMValueRef chan_ptr2,
+ LLVMValueRef value)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
struct lp_build_context *float_bld = &bld_base->base;
- int i;
+ unsigned i;
LLVMValueRef temp, temp2;
- LLVMValueRef shuffles[8];
- LLVMValueRef shuffles2[8];
+ LLVMValueRef shuffles[LP_MAX_VECTOR_WIDTH/32];
+ LLVMValueRef shuffles2[LP_MAX_VECTOR_WIDTH/32];
for (i = 0; i < bld_base->base.type.length; i++) {
shuffles[i] = lp_build_const_int32(gallivm, i * 2);
bld_base->base.type.length),
"");
- lp_exec_mask_store(&bld->exec_mask, float_bld, pred, temp, chan_ptr);
- lp_exec_mask_store(&bld->exec_mask, float_bld, pred, temp2, chan_ptr2);
+ lp_exec_mask_store(&bld->exec_mask, float_bld, temp, chan_ptr);
+ lp_exec_mask_store(&bld->exec_mask, float_bld, temp2, chan_ptr2);
}
/**
const struct tgsi_full_instruction *inst,
unsigned index,
unsigned chan_index,
- LLVMValueRef pred,
LLVMValueRef value)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct lp_build_context *float_bld = &bld_base->base;
struct lp_build_context *int_bld = &bld_base->int_bld;
LLVMValueRef indirect_index = NULL;
- enum tgsi_opcode_type dtype = tgsi_opcode_infer_dst_type(inst->Instruction.Opcode);
+ enum tgsi_opcode_type dtype = tgsi_opcode_infer_dst_type(inst->Instruction.Opcode, index);
/*
* Apply saturation.
if (reg->Register.Indirect) {
/*
* Currently the mesa/st doesn't generate indirect stores
- * to doubles, it normally uses MOV to do indirect stores.
+ * to 64-bit values, it normally uses MOV to do indirect stores.
*/
- assert(dtype != TGSI_TYPE_DOUBLE);
+ assert(!tgsi_type_is_64bit(dtype));
indirect_index = get_indirect_index(bld,
reg->Register.File,
reg->Register.Index,
- ®->Indirect);
+ ®->Indirect,
+ bld->bld_base.info->file_max[reg->Register.File]);
} else {
assert(reg->Register.Index <=
bld_base->info->file_max[reg->Register.File]);
/* Scatter store values into output registers */
emit_mask_scatter(bld, outputs_array, index_vec, value,
- &bld->exec_mask, pred);
+ &bld->exec_mask);
}
else {
LLVMValueRef out_ptr = lp_get_output_ptr(bld, reg->Register.Index,
chan_index);
- if (dtype == TGSI_TYPE_DOUBLE) {
+ if (tgsi_type_is_64bit(dtype)) {
LLVMValueRef out_ptr2 = lp_get_output_ptr(bld, reg->Register.Index,
chan_index + 1);
- emit_store_double_chan(bld_base, dtype, out_ptr, out_ptr2,
- pred, value);
+ emit_store_64bit_chan(bld_base, out_ptr, out_ptr2,
+ value);
} else
- lp_exec_mask_store(&bld->exec_mask, float_bld, pred, value, out_ptr);
+ lp_exec_mask_store(&bld->exec_mask, float_bld, value, out_ptr);
}
break;
case TGSI_FILE_TEMPORARY:
/* Temporaries are always stored as floats */
- if (dtype != TGSI_TYPE_DOUBLE)
+ if (!tgsi_type_is_64bit(dtype))
value = LLVMBuildBitCast(builder, value, float_bld->vec_type, "");
else
value = LLVMBuildBitCast(builder, value, LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), bld_base->base.type.length * 2), "");
/* Scatter store values into temp registers */
emit_mask_scatter(bld, temps_array, index_vec, value,
- &bld->exec_mask, pred);
+ &bld->exec_mask);
}
else {
LLVMValueRef temp_ptr;
temp_ptr = lp_get_temp_ptr_soa(bld, reg->Register.Index, chan_index);
- if (dtype == TGSI_TYPE_DOUBLE) {
+ if (tgsi_type_is_64bit(dtype)) {
LLVMValueRef temp_ptr2 = lp_get_temp_ptr_soa(bld,
reg->Register.Index,
chan_index + 1);
- emit_store_double_chan(bld_base, dtype, temp_ptr, temp_ptr2,
- pred, value);
+ emit_store_64bit_chan(bld_base, temp_ptr, temp_ptr2,
+ value);
}
else
- lp_exec_mask_store(&bld->exec_mask, float_bld, pred, value, temp_ptr);
+ lp_exec_mask_store(&bld->exec_mask, float_bld, value, temp_ptr);
}
break;
assert(dtype == TGSI_TYPE_SIGNED);
assert(LLVMTypeOf(value) == int_bld->vec_type);
value = LLVMBuildBitCast(builder, value, int_bld->vec_type, "");
- lp_exec_mask_store(&bld->exec_mask, int_bld, pred, value,
+ lp_exec_mask_store(&bld->exec_mask, int_bld, value,
bld->addr[reg->Register.Index][chan_index]);
break;
- case TGSI_FILE_PREDICATE:
- assert(LLVMTypeOf(value) == float_bld->vec_type);
- value = LLVMBuildBitCast(builder, value, float_bld->vec_type, "");
- lp_exec_mask_store(&bld->exec_mask, float_bld, pred, value,
- bld->preds[reg->Register.Index][chan_index]);
- break;
-
default:
assert( 0 );
}
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_instruction * inst,
const struct tgsi_opcode_info * info,
+ unsigned index,
LLVMValueRef dst[4])
{
- unsigned chan_index;
- struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
- enum tgsi_opcode_type dtype = tgsi_opcode_infer_dst_type(inst->Instruction.Opcode);
- if(info->num_dst) {
- LLVMValueRef pred[TGSI_NUM_CHANNELS];
-
- emit_fetch_predicate( bld, inst, pred );
+ enum tgsi_opcode_type dtype = tgsi_opcode_infer_dst_type(inst->Instruction.Opcode, index);
- TGSI_FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-
- if (dtype == TGSI_TYPE_DOUBLE && (chan_index == 1 || chan_index == 3))
- continue;
- emit_store_chan(bld_base, inst, 0, chan_index, pred[chan_index], dst[chan_index]);
- }
+ unsigned writemask = inst->Dst[index].Register.WriteMask;
+ while (writemask) {
+ unsigned chan_index = u_bit_scan(&writemask);
+ if (tgsi_type_is_64bit(dtype) && (chan_index == 1 || chan_index == 3))
+ continue;
+ emit_store_chan(bld_base, inst, index, chan_index, dst[chan_index]);
}
}
reg->Register.File == TGSI_FILE_IMMEDIATE) {
lod_property = LP_SAMPLER_LOD_SCALAR;
}
- else if (bld_base->info->processor == TGSI_PROCESSOR_FRAGMENT) {
- if (gallivm_debug & GALLIVM_DEBUG_NO_QUAD_LOD) {
+ else if (bld_base->info->processor == PIPE_SHADER_FRAGMENT) {
+ if (gallivm_perf & GALLIVM_PERF_NO_QUAD_LOD) {
lod_property = LP_SAMPLER_LOD_PER_ELEMENT;
}
else {
* could also check all src regs if constant but I doubt such
* cases exist in practice.
*/
- if (bld->bld_base.info->processor == TGSI_PROCESSOR_FRAGMENT) {
- if (gallivm_debug & GALLIVM_DEBUG_NO_QUAD_LOD) {
+ if (bld->bld_base.info->processor == PIPE_SHADER_FRAGMENT) {
+ if (gallivm_perf & GALLIVM_PERF_NO_QUAD_LOD) {
lod_property = LP_SAMPLER_LOD_PER_ELEMENT;
}
else {
const struct tgsi_full_instruction *inst,
enum lp_build_tex_modifier modifier,
boolean compare,
+ enum lp_sampler_op_type sample_type,
LLVMValueRef *texel)
{
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
unsigned num_offsets, num_derivs, i;
unsigned layer_coord = 0;
- unsigned sample_key = LP_SAMPLER_OP_TEXTURE << LP_SAMPLER_OP_TYPE_SHIFT;
+ unsigned sample_key = sample_type << LP_SAMPLER_OP_TYPE_SHIFT;
memset(¶ms, 0, sizeof(params));
* could also check all src regs if constant but I doubt such
* cases exist in practice.
*/
- if (bld->bld_base.info->processor == TGSI_PROCESSOR_FRAGMENT) {
- if (gallivm_debug & GALLIVM_DEBUG_NO_QUAD_LOD) {
+ if (bld->bld_base.info->processor == PIPE_SHADER_FRAGMENT) {
+ if (gallivm_perf & GALLIVM_PERF_NO_QUAD_LOD) {
lod_property = LP_SAMPLER_LOD_PER_ELEMENT;
}
else {
bld->bld_base.base.gallivm,
¶ms);
- if (inst->Src[1].Register.SwizzleX != PIPE_SWIZZLE_RED ||
- inst->Src[1].Register.SwizzleY != PIPE_SWIZZLE_GREEN ||
- inst->Src[1].Register.SwizzleZ != PIPE_SWIZZLE_BLUE ||
- inst->Src[1].Register.SwizzleW != PIPE_SWIZZLE_ALPHA) {
+ if (inst->Src[1].Register.SwizzleX != PIPE_SWIZZLE_X ||
+ inst->Src[1].Register.SwizzleY != PIPE_SWIZZLE_Y ||
+ inst->Src[1].Register.SwizzleZ != PIPE_SWIZZLE_Z ||
+ inst->Src[1].Register.SwizzleW != PIPE_SWIZZLE_W) {
unsigned char swizzles[4];
swizzles[0] = inst->Src[1].Register.SwizzleX;
swizzles[1] = inst->Src[1].Register.SwizzleY;
explicit_lod = lp_build_emit_fetch(&bld->bld_base, inst, 0, 3);
lod_property = lp_build_lod_property(&bld->bld_base, inst, 0);
}
- /* XXX: for real msaa support, the w component would be the sample index. */
+ /*
+ * XXX: for real msaa support, the w component (or src2.x for sample_i_ms)
+ * would be the sample index.
+ */
for (i = 0; i < dims; i++) {
coords[i] = lp_build_emit_fetch(&bld->bld_base, inst, 0, i);
¶ms);
if (is_samplei &&
- (inst->Src[1].Register.SwizzleX != PIPE_SWIZZLE_RED ||
- inst->Src[1].Register.SwizzleY != PIPE_SWIZZLE_GREEN ||
- inst->Src[1].Register.SwizzleZ != PIPE_SWIZZLE_BLUE ||
- inst->Src[1].Register.SwizzleW != PIPE_SWIZZLE_ALPHA)) {
+ (inst->Src[1].Register.SwizzleX != PIPE_SWIZZLE_X ||
+ inst->Src[1].Register.SwizzleY != PIPE_SWIZZLE_Y ||
+ inst->Src[1].Register.SwizzleZ != PIPE_SWIZZLE_Z ||
+ inst->Src[1].Register.SwizzleW != PIPE_SWIZZLE_W)) {
unsigned char swizzles[4];
swizzles[0] = inst->Src[1].Register.SwizzleX;
swizzles[1] = inst->Src[1].Register.SwizzleY;
unsigned i;
unsigned unit = inst->Src[1].Register.Index;
unsigned target, pipe_target;
+ struct lp_sampler_size_query_params params;
if (is_sviewinfo) {
target = bld->sv[unit].Resource;
pipe_target = tgsi_to_pipe_tex_target(target);
+ params.int_type = bld->bld_base.int_bld.type;
+ params.texture_unit = unit;
+ params.target = pipe_target;
+ params.context_ptr = bld->context_ptr;
+ params.is_sviewinfo = TRUE;
+ params.lod_property = lod_property;
+ params.explicit_lod = explicit_lod;
+ params.sizes_out = sizes_out;
+
bld->sampler->emit_size_query(bld->sampler,
bld->bld_base.base.gallivm,
- bld->bld_base.int_bld.type,
- unit, pipe_target,
- bld->context_ptr,
- TRUE,
- lod_property,
- explicit_lod,
- sizes_out);
+ ¶ms);
}
static boolean
near_end_of_shader(struct lp_build_tgsi_soa_context *bld,
int pc)
{
- int i;
+ unsigned i;
for (i = 0; i < 5; i++) {
- unsigned opcode;
+ enum tgsi_opcode opcode;
if (pc + i >= bld->bld_base.info->num_instructions)
return TRUE;
opcode == TGSI_OPCODE_SAMPLE_C_LZ ||
opcode == TGSI_OPCODE_SAMPLE_D ||
opcode == TGSI_OPCODE_SAMPLE_I ||
+ opcode == TGSI_OPCODE_SAMPLE_I_MS ||
opcode == TGSI_OPCODE_SAMPLE_L ||
opcode == TGSI_OPCODE_SVIEWINFO ||
opcode == TGSI_OPCODE_CAL ||
- opcode == TGSI_OPCODE_CALLNZ ||
opcode == TGSI_OPCODE_IF ||
opcode == TGSI_OPCODE_UIF ||
opcode == TGSI_OPCODE_BGNLOOP ||
int chan;
if (index < 8 * sizeof(unsigned) &&
- (info->file_mask[file] & (1 << index)) == 0) {
+ (info->file_mask[file] & (1u << index)) == 0) {
/* This was not declared.*/
continue;
}
}
break;
- case TGSI_FILE_PREDICATE:
- assert(last < LP_MAX_TGSI_PREDS);
- for (idx = first; idx <= last; ++idx) {
- for (i = 0; i < TGSI_NUM_CHANNELS; i++)
- bld->preds[idx][i] = lp_build_alloca(gallivm, vec_type,
- "predicate");
- }
- break;
-
case TGSI_FILE_SAMPLER_VIEW:
/*
* The target stored here MUST match whatever there actually
bld->consts_sizes[idx2D] =
lp_build_array_get(gallivm, bld->const_sizes_ptr, index2D);
}
- break;
+ break;
+ case TGSI_FILE_BUFFER:
+ {
+ unsigned idx = decl->Range.First;
+ LLVMValueRef index = lp_build_const_int32(gallivm, idx);
+ assert(idx < LP_MAX_TGSI_SHADER_BUFFERS);
+ bld->ssbos[idx] =
+ lp_build_array_get(gallivm, bld->ssbo_ptr, index);
+ bld->ssbo_sizes[idx] =
+ lp_build_array_get(gallivm, bld->ssbo_sizes_ptr, index);
+ }
+ break;
default:
/* don't need to declare other vars */
break;
break;
case TGSI_IMM_FLOAT64:
+ case TGSI_IMM_UINT64:
+ case TGSI_IMM_INT64:
case TGSI_IMM_UINT32:
for( i = 0; i < size; ++i ) {
LLVMValueRef tmp = lp_build_const_vec(gallivm, bld_base->uint_bld.type, imm->u[i].Uint);
unsigned index = bld->num_immediates;
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
+ LLVMValueRef gep[2];
+ gep[0] = lp_build_const_int32(gallivm, 0);
assert(bld->indirect_files & (1 << TGSI_FILE_IMMEDIATE));
for (i = 0; i < 4; ++i ) {
- LLVMValueRef lindex = lp_build_const_int32(
- bld->bld_base.base.gallivm, index * 4 + i);
+ gep[1] = lp_build_const_int32(gallivm, index * 4 + i);
LLVMValueRef imm_ptr = LLVMBuildGEP(builder,
- bld->imms_array, &lindex, 1, "");
+ bld->imms_array, gep, 2, "");
LLVMBuildStore(builder, imms[i], imm_ptr);
}
} else {
unsigned index = bld->num_immediates;
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
+ LLVMValueRef gep[2];
+ gep[0] = lp_build_const_int32(gallivm, 0);
for (i = 0; i < 4; ++i ) {
- LLVMValueRef lindex = lp_build_const_int32(
- bld->bld_base.base.gallivm, index * 4 + i);
+ gep[1] = lp_build_const_int32(gallivm, index * 4 + i);
LLVMValueRef imm_ptr = LLVMBuildGEP(builder,
- bld->imms_array, &lindex, 1, "");
+ bld->imms_array, gep, 2, "");
LLVMBuildStore(builder,
bld->immediates[index][i],
imm_ptr);
emit_data->output, 2, LP_SAMPLER_OP_GATHER);
}
+static void
+lodq_emit(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
+
+ emit_tex(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_NONE,
+ emit_data->output, 1, LP_SAMPLER_OP_LODQ);
+}
+
static void
txq_emit(
const struct lp_build_tgsi_action * action,
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
emit_sample(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_NONE,
- FALSE, emit_data->output);
+ FALSE, LP_SAMPLER_OP_TEXTURE, emit_data->output);
}
static void
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
emit_sample(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_LOD_BIAS,
- FALSE, emit_data->output);
+ FALSE, LP_SAMPLER_OP_TEXTURE, emit_data->output);
}
static void
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
emit_sample(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_NONE,
- TRUE, emit_data->output);
+ TRUE, LP_SAMPLER_OP_TEXTURE, emit_data->output);
}
static void
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
emit_sample(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_LOD_ZERO,
- TRUE, emit_data->output);
+ TRUE, LP_SAMPLER_OP_TEXTURE, emit_data->output);
}
static void
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
emit_sample(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV,
- FALSE, emit_data->output);
+ FALSE, LP_SAMPLER_OP_TEXTURE, emit_data->output);
}
static void
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
emit_sample(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_EXPLICIT_LOD,
- FALSE, emit_data->output);
+ FALSE, LP_SAMPLER_OP_TEXTURE, emit_data->output);
+}
+
+static void
+gather4_emit(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
+
+ emit_sample(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_NONE,
+ FALSE, LP_SAMPLER_OP_GATHER, emit_data->output);
}
static void
emit_size_query(bld, emit_data->inst, emit_data->output, TRUE);
}
-static LLVMValueRef
-mask_vec(struct lp_build_tgsi_context *bld_base)
+static void
+lod_emit(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
+
+ emit_sample(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_NONE,
+ FALSE, LP_SAMPLER_OP_LODQ, emit_data->output);
+}
+
+static void
+load_emit(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct lp_build_tgsi_soa_context *bld = lp_soa_context(bld_base);
+ struct gallivm_state * gallivm = bld_base->base.gallivm;
LLVMBuilderRef builder = bld->bld_base.base.gallivm->builder;
- struct lp_exec_mask *exec_mask = &bld->exec_mask;
+ const struct tgsi_full_src_register *bufreg = &emit_data->inst->Src[0];
+ unsigned buf = bufreg->Register.Index;
+ assert(bufreg->Register.File == TGSI_FILE_BUFFER);
+ struct lp_build_context *uint_bld = &bld_base->uint_bld;
- if (!exec_mask->has_mask) {
- return lp_build_mask_value(bld->mask);
+ if (0) {
+ /* for indirect support with ARB_gpu_shader5 */
+ } else {
+ LLVMValueRef index;
+ LLVMValueRef scalar, scalar_ptr;
+ unsigned chan_index;
+
+ index = lp_build_emit_fetch(&bld->bld_base, emit_data->inst, 1, 0);
+ index = lp_build_shr_imm(uint_bld, index, 2);
+
+ scalar_ptr = bld->ssbos[buf];
+
+ LLVMValueRef ssbo_limit;
+
+ ssbo_limit = LLVMBuildAShr(gallivm->builder, bld->ssbo_sizes[buf], lp_build_const_int32(gallivm, 2), "");
+ ssbo_limit = lp_build_broadcast_scalar(uint_bld, ssbo_limit);
+
+ TGSI_FOR_EACH_DST0_ENABLED_CHANNEL(emit_data->inst, chan_index) {
+ LLVMValueRef loop_index = lp_build_add(uint_bld, index, lp_build_const_int_vec(gallivm, uint_bld->type, chan_index));
+
+ LLVMValueRef exec_mask = mask_vec(bld_base);
+ LLVMValueRef ssbo_oob_cmp = lp_build_cmp(uint_bld, PIPE_FUNC_LESS, loop_index, ssbo_limit);
+ exec_mask = LLVMBuildAnd(builder, exec_mask, ssbo_oob_cmp, "");
+
+ LLVMValueRef result = lp_build_alloca(gallivm, uint_bld->vec_type, "");
+ struct lp_build_loop_state loop_state;
+ lp_build_loop_begin(&loop_state, gallivm, lp_build_const_int32(gallivm, 0));
+
+ struct lp_build_if_state ifthen;
+ LLVMValueRef cond, temp_res;
+
+ loop_index = LLVMBuildExtractElement(gallivm->builder, loop_index,
+ loop_state.counter, "");
+
+ cond = LLVMBuildICmp(gallivm->builder, LLVMIntNE, exec_mask, uint_bld->zero, "");
+ cond = LLVMBuildExtractElement(gallivm->builder, cond, loop_state.counter, "");
+
+ lp_build_if(&ifthen, gallivm, cond);
+ scalar = lp_build_pointer_get(builder, scalar_ptr, loop_index);
+
+ temp_res = LLVMBuildLoad(builder, result, "");
+ temp_res = LLVMBuildInsertElement(builder, temp_res, scalar, loop_state.counter, "");
+ LLVMBuildStore(builder, temp_res, result);
+ lp_build_else(&ifthen);
+ temp_res = LLVMBuildLoad(builder, result, "");
+ temp_res = LLVMBuildInsertElement(builder, temp_res, lp_build_const_int32(gallivm, 0), loop_state.counter, "");
+ LLVMBuildStore(builder, temp_res, result);
+ lp_build_endif(&ifthen);
+ lp_build_loop_end_cond(&loop_state, lp_build_const_int32(gallivm, uint_bld->type.length),
+ NULL, LLVMIntUGE);
+ emit_data->output[chan_index] = LLVMBuildLoad(gallivm->builder, result, "");
+ }
+ }
+}
+
+static void
+store_emit(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct lp_build_tgsi_soa_context *bld = lp_soa_context(bld_base);
+ struct gallivm_state * gallivm = bld_base->base.gallivm;
+ LLVMBuilderRef builder = bld->bld_base.base.gallivm->builder;
+ struct lp_build_context *uint_bld = &bld_base->uint_bld;
+ const struct tgsi_full_dst_register *bufreg = &emit_data->inst->Dst[0];
+ unsigned buf = bufreg->Register.Index;
+ assert(bufreg->Register.File == TGSI_FILE_BUFFER);
+
+ if (0) {
+
+ } else {
+ LLVMValueRef index; /* index into the const buffer */
+ LLVMValueRef scalar_ptr;
+ LLVMValueRef value;
+ unsigned chan_index;
+
+ index = lp_build_emit_fetch(&bld->bld_base, emit_data->inst, 0, 0);
+ index = lp_build_shr_imm(uint_bld, index, 2);
+
+ scalar_ptr = bld->ssbos[buf];
+
+ LLVMValueRef ssbo_limit;
+
+ ssbo_limit = LLVMBuildAShr(gallivm->builder, bld->ssbo_sizes[buf], lp_build_const_int32(gallivm, 2), "");
+ ssbo_limit = lp_build_broadcast_scalar(uint_bld, ssbo_limit);
+
+ TGSI_FOR_EACH_DST0_ENABLED_CHANNEL(emit_data->inst, chan_index) {
+ LLVMValueRef loop_index = lp_build_add(uint_bld, index, lp_build_const_int_vec(gallivm, uint_bld->type, chan_index));
+
+ value = lp_build_emit_fetch(&bld->bld_base, emit_data->inst, 1, chan_index);
+
+ LLVMValueRef exec_mask = mask_vec(bld_base);
+ LLVMValueRef ssbo_oob_cmp = lp_build_cmp(uint_bld, PIPE_FUNC_LESS, loop_index, ssbo_limit);
+ exec_mask = LLVMBuildAnd(builder, exec_mask, ssbo_oob_cmp, "");
+
+ struct lp_build_loop_state loop_state;
+ lp_build_loop_begin(&loop_state, gallivm, lp_build_const_int32(gallivm, 0));
+
+ LLVMValueRef value_ptr = LLVMBuildExtractElement(gallivm->builder, value,
+ loop_state.counter, "");
+ value_ptr = LLVMBuildBitCast(gallivm->builder, value_ptr, uint_bld->elem_type, "");
+
+ struct lp_build_if_state ifthen;
+ LLVMValueRef cond;
+
+ loop_index = LLVMBuildExtractElement(gallivm->builder, loop_index,
+ loop_state.counter, "");
+
+ cond = LLVMBuildICmp(gallivm->builder, LLVMIntNE, exec_mask, uint_bld->zero, "");
+ cond = LLVMBuildExtractElement(gallivm->builder, cond, loop_state.counter, "");
+ lp_build_if(&ifthen, gallivm, cond);
+
+ lp_build_pointer_set(builder, scalar_ptr, loop_index, value_ptr);
+
+ lp_build_endif(&ifthen);
+ lp_build_loop_end_cond(&loop_state, lp_build_const_int32(gallivm, uint_bld->type.length),
+ NULL, LLVMIntUGE);
+ }
+ }
+}
+
+static void
+resq_emit(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct lp_build_tgsi_soa_context *bld = lp_soa_context(bld_base);
+ struct lp_build_context *uint_bld = &bld_base->uint_bld;
+ const struct tgsi_full_src_register *bufreg = &emit_data->inst->Src[0];
+
+ unsigned buf = bufreg->Register.Index;
+ assert(bufreg->Register.File == TGSI_FILE_BUFFER);
+
+ LLVMValueRef num_ssbo = bld->ssbo_sizes[buf];
+
+ emit_data->output[emit_data->chan] = lp_build_broadcast_scalar(uint_bld, num_ssbo);
+}
+
+static void
+atomic_emit(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct lp_build_tgsi_soa_context *bld = lp_soa_context(bld_base);
+ struct gallivm_state * gallivm = bld_base->base.gallivm;
+ LLVMBuilderRef builder = gallivm->builder;
+ struct lp_build_context *uint_bld = &bld_base->uint_bld;
+ const struct tgsi_full_src_register *bufreg = &emit_data->inst->Src[0];
+
+ assert(bufreg->Register.File == TGSI_FILE_BUFFER);
+ unsigned buf = bufreg->Register.Index;
+
+ LLVMAtomicRMWBinOp op;
+ switch (emit_data->inst->Instruction.Opcode) {
+ case TGSI_OPCODE_ATOMUADD:
+ op = LLVMAtomicRMWBinOpAdd;
+ break;
+ case TGSI_OPCODE_ATOMXCHG:
+ op = LLVMAtomicRMWBinOpXchg;
+ break;
+ case TGSI_OPCODE_ATOMAND:
+ op = LLVMAtomicRMWBinOpAnd;
+ break;
+ case TGSI_OPCODE_ATOMOR:
+ op = LLVMAtomicRMWBinOpOr;
+ break;
+ case TGSI_OPCODE_ATOMXOR:
+ op = LLVMAtomicRMWBinOpXor;
+ break;
+ case TGSI_OPCODE_ATOMUMIN:
+ op = LLVMAtomicRMWBinOpUMin;
+ break;
+ case TGSI_OPCODE_ATOMUMAX:
+ op = LLVMAtomicRMWBinOpUMax;
+ break;
+ case TGSI_OPCODE_ATOMIMIN:
+ op = LLVMAtomicRMWBinOpMin;
+ break;
+ case TGSI_OPCODE_ATOMIMAX:
+ op = LLVMAtomicRMWBinOpMax;
+ break;
+ case TGSI_OPCODE_ATOMCAS:
+ break;
+ default:
+ assert(0);
+ return;
+ }
+
+ if (0) {
+ } else {
+ LLVMValueRef index; /* index into the const buffer */
+ LLVMValueRef scalar, scalar_ptr;
+ LLVMValueRef value;
+
+ index = lp_build_emit_fetch(&bld->bld_base, emit_data->inst, 1, 0);
+ value = lp_build_emit_fetch(&bld->bld_base, emit_data->inst, 2, 0);
+
+ index = lp_build_shr_imm(uint_bld, index, 2);
+ index = lp_build_add(uint_bld, index, lp_build_const_int_vec(gallivm, uint_bld->type, emit_data->chan));
+
+ scalar_ptr = bld->ssbos[buf];
+
+ LLVMValueRef atom_res = lp_build_alloca(gallivm,
+ uint_bld->vec_type, "");
+
+ LLVMValueRef ssbo_limit;
+ ssbo_limit = LLVMBuildAShr(gallivm->builder, bld->ssbo_sizes[buf], lp_build_const_int32(gallivm, 2), "");
+ ssbo_limit = lp_build_broadcast_scalar(uint_bld, ssbo_limit);
+
+ LLVMValueRef exec_mask = mask_vec(bld_base);
+ LLVMValueRef ssbo_oob_cmp = lp_build_cmp(uint_bld, PIPE_FUNC_LESS, index, ssbo_limit);
+ exec_mask = LLVMBuildAnd(builder, exec_mask, ssbo_oob_cmp, "");
+
+ struct lp_build_loop_state loop_state;
+ lp_build_loop_begin(&loop_state, gallivm, lp_build_const_int32(gallivm, 0));
+
+ LLVMValueRef value_ptr = LLVMBuildExtractElement(gallivm->builder, value,
+ loop_state.counter, "");
+ value_ptr = LLVMBuildBitCast(gallivm->builder, value_ptr, uint_bld->elem_type, "");
+
+ index = LLVMBuildExtractElement(gallivm->builder, index,
+ loop_state.counter, "");
+
+ scalar_ptr = LLVMBuildGEP(builder, scalar_ptr,
+ &index, 1, "");
+
+ struct lp_build_if_state ifthen;
+ LLVMValueRef cond, temp_res;
+
+ cond = LLVMBuildICmp(gallivm->builder, LLVMIntNE, exec_mask, uint_bld->zero, "");
+ cond = LLVMBuildExtractElement(gallivm->builder, cond, loop_state.counter, "");
+ lp_build_if(&ifthen, gallivm, cond);
+
+ if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_ATOMCAS) {
+ LLVMValueRef cas_src = lp_build_emit_fetch(&bld->bld_base, emit_data->inst, 3, 0);
+ LLVMValueRef cas_src_ptr = LLVMBuildExtractElement(gallivm->builder, cas_src,
+ loop_state.counter, "");
+ cas_src_ptr = LLVMBuildBitCast(gallivm->builder, cas_src_ptr, uint_bld->elem_type, "");
+ scalar = LLVMBuildAtomicCmpXchg(builder, scalar_ptr, value_ptr,
+ cas_src_ptr,
+ LLVMAtomicOrderingSequentiallyConsistent,
+ LLVMAtomicOrderingSequentiallyConsistent,
+ false);
+ scalar = LLVMBuildExtractValue(gallivm->builder, scalar, 0, "");
+ } else {
+ scalar = LLVMBuildAtomicRMW(builder, op,
+ scalar_ptr, value_ptr,
+ LLVMAtomicOrderingSequentiallyConsistent,
+ false);
+ }
+ temp_res = LLVMBuildLoad(builder, atom_res, "");
+ temp_res = LLVMBuildInsertElement(builder, temp_res, scalar, loop_state.counter, "");
+ LLVMBuildStore(builder, temp_res, atom_res);
+ lp_build_else(&ifthen);
+ temp_res = LLVMBuildLoad(builder, atom_res, "");
+ temp_res = LLVMBuildInsertElement(builder, temp_res, lp_build_const_int32(gallivm, 0), loop_state.counter, "");
+ LLVMBuildStore(builder, temp_res, atom_res);
+ lp_build_endif(&ifthen);
+
+ lp_build_loop_end_cond(&loop_state, lp_build_const_int32(gallivm, uint_bld->type.length),
+ NULL, LLVMIntUGE);
+ emit_data->output[emit_data->chan] = LLVMBuildLoad(gallivm->builder, atom_res, "");
}
- return LLVMBuildAnd(builder, lp_build_mask_value(bld->mask),
- exec_mask->exec_mask, "");
}
static void
lp_exec_break(&bld->exec_mask, bld_base);
}
-static void
-breakc_emit(
- const struct lp_build_tgsi_action * action,
- struct lp_build_tgsi_context * bld_base,
- struct lp_build_emit_data * emit_data)
-{
- struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
- LLVMBuilderRef builder = bld_base->base.gallivm->builder;
- struct lp_build_context *uint_bld = &bld_base->uint_bld;
- LLVMValueRef unsigned_cond =
- LLVMBuildBitCast(builder, emit_data->args[0], uint_bld->vec_type, "");
- LLVMValueRef cond = lp_build_cmp(uint_bld, PIPE_FUNC_NOTEQUAL,
- unsigned_cond,
- uint_bld->zero);
-
- lp_exec_break_condition(&bld->exec_mask, cond);
-}
-
static void
if_emit(
const struct lp_build_tgsi_action * action,
struct gallivm_state * gallivm = bld_base->base.gallivm;
if (bld->indirect_files & (1 << TGSI_FILE_TEMPORARY)) {
- LLVMValueRef array_size =
- lp_build_const_int32(gallivm,
- bld_base->info->file_max[TGSI_FILE_TEMPORARY] * 4 + 4);
- bld->temps_array = lp_build_array_alloca(gallivm,
- bld_base->base.vec_type, array_size,
- "temp_array");
+ unsigned array_size = bld_base->info->file_max[TGSI_FILE_TEMPORARY] * 4 + 4;
+ bld->temps_array = lp_build_alloca_undef(gallivm,
+ LLVMArrayType(bld_base->base.vec_type, array_size),
+ "temp_array");
}
if (bld->indirect_files & (1 << TGSI_FILE_OUTPUT)) {
}
if (bld->indirect_files & (1 << TGSI_FILE_IMMEDIATE)) {
- LLVMValueRef array_size =
- lp_build_const_int32(gallivm,
- bld_base->info->file_max[TGSI_FILE_IMMEDIATE] * 4 + 4);
- bld->imms_array = lp_build_array_alloca(gallivm,
- bld_base->base.vec_type, array_size,
+ unsigned array_size = bld_base->info->file_max[TGSI_FILE_IMMEDIATE] * 4 + 4;
+ bld->imms_array = lp_build_alloca_undef(gallivm,
+ LLVMArrayType(bld_base->base.vec_type, array_size),
"imms_array");
}
void
lp_build_tgsi_soa(struct gallivm_state *gallivm,
const struct tgsi_token *tokens,
- struct lp_type type,
- struct lp_build_mask_context *mask,
- LLVMValueRef consts_ptr,
- LLVMValueRef const_sizes_ptr,
- const struct lp_bld_tgsi_system_values *system_values,
- const LLVMValueRef (*inputs)[TGSI_NUM_CHANNELS],
- LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS],
- LLVMValueRef context_ptr,
- LLVMValueRef thread_data_ptr,
- struct lp_build_sampler_soa *sampler,
- const struct tgsi_shader_info *info,
- const struct lp_build_tgsi_gs_iface *gs_iface)
+ const struct lp_build_tgsi_params *params,
+ LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS])
{
struct lp_build_tgsi_soa_context bld;
-
+ struct lp_type type = params->type;
struct lp_type res_type;
assert(type.length <= LP_MAX_VECTOR_LENGTH);
dbl_type.width *= 2;
lp_build_context_init(&bld.bld_base.dbl_bld, gallivm, dbl_type);
}
- bld.mask = mask;
- bld.inputs = inputs;
+ {
+ struct lp_type uint64_type;
+ uint64_type = lp_uint_type(type);
+ uint64_type.width *= 2;
+ lp_build_context_init(&bld.bld_base.uint64_bld, gallivm, uint64_type);
+ }
+ {
+ struct lp_type int64_type;
+ int64_type = lp_int_type(type);
+ int64_type.width *= 2;
+ lp_build_context_init(&bld.bld_base.int64_bld, gallivm, int64_type);
+ }
+ bld.mask = params->mask;
+ bld.inputs = params->inputs;
bld.outputs = outputs;
- bld.consts_ptr = consts_ptr;
- bld.const_sizes_ptr = const_sizes_ptr;
- bld.sampler = sampler;
- bld.bld_base.info = info;
- bld.indirect_files = info->indirect_files;
- bld.context_ptr = context_ptr;
- bld.thread_data_ptr = thread_data_ptr;
+ bld.consts_ptr = params->consts_ptr;
+ bld.const_sizes_ptr = params->const_sizes_ptr;
+ bld.ssbo_ptr = params->ssbo_ptr;
+ bld.ssbo_sizes_ptr = params->ssbo_sizes_ptr;
+ bld.sampler = params->sampler;
+ bld.bld_base.info = params->info;
+ bld.indirect_files = params->info->indirect_files;
+ bld.context_ptr = params->context_ptr;
+ bld.thread_data_ptr = params->thread_data_ptr;
/*
* If the number of temporaries is rather large then we just
* allocate them as an array right from the start and treat
* like indirect temporaries.
*/
- if (info->file_max[TGSI_FILE_TEMPORARY] >= LP_MAX_INLINED_TEMPS) {
+ if (params->info->file_max[TGSI_FILE_TEMPORARY] >= LP_MAX_INLINED_TEMPS) {
bld.indirect_files |= (1 << TGSI_FILE_TEMPORARY);
}
/*
* a dynamically allocated array.
*/
bld.use_immediates_array =
- (info->file_max[TGSI_FILE_IMMEDIATE] >= LP_MAX_INLINED_IMMEDIATES);
+ (params->info->file_max[TGSI_FILE_IMMEDIATE] >= LP_MAX_INLINED_IMMEDIATES);
if (bld.use_immediates_array) {
bld.indirect_files |= (1 << TGSI_FILE_IMMEDIATE);
}
bld.bld_base.op_actions[TGSI_OPCODE_BGNLOOP].emit = bgnloop_emit;
bld.bld_base.op_actions[TGSI_OPCODE_BGNSUB].emit = bgnsub_emit;
bld.bld_base.op_actions[TGSI_OPCODE_BRK].emit = brk_emit;
- bld.bld_base.op_actions[TGSI_OPCODE_BREAKC].emit = breakc_emit;
bld.bld_base.op_actions[TGSI_OPCODE_CAL].emit = cal_emit;
bld.bld_base.op_actions[TGSI_OPCODE_CASE].emit = case_emit;
bld.bld_base.op_actions[TGSI_OPCODE_CONT].emit = cont_emit;
bld.bld_base.op_actions[TGSI_OPCODE_TXB2].emit = txb2_emit;
bld.bld_base.op_actions[TGSI_OPCODE_TXL2].emit = txl2_emit;
bld.bld_base.op_actions[TGSI_OPCODE_TG4].emit = tg4_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_LODQ].emit = lodq_emit;
/* DX10 sampling ops */
bld.bld_base.op_actions[TGSI_OPCODE_SAMPLE].emit = sample_emit;
bld.bld_base.op_actions[TGSI_OPCODE_SAMPLE_B].emit = sample_b_emit;
bld.bld_base.op_actions[TGSI_OPCODE_SAMPLE_C_LZ].emit = sample_c_lz_emit;
bld.bld_base.op_actions[TGSI_OPCODE_SAMPLE_D].emit = sample_d_emit;
bld.bld_base.op_actions[TGSI_OPCODE_SAMPLE_I].emit = sample_i_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_SAMPLE_I_MS].emit = sample_i_emit;
bld.bld_base.op_actions[TGSI_OPCODE_SAMPLE_L].emit = sample_l_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_GATHER4].emit = gather4_emit;
bld.bld_base.op_actions[TGSI_OPCODE_SVIEWINFO].emit = sviewinfo_emit;
-
- if (gs_iface) {
+ bld.bld_base.op_actions[TGSI_OPCODE_LOD].emit = lod_emit;
+
+ bld.bld_base.op_actions[TGSI_OPCODE_LOAD].emit = load_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_STORE].emit = store_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_RESQ].emit = resq_emit;
+
+ bld.bld_base.op_actions[TGSI_OPCODE_ATOMUADD].emit = atomic_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_ATOMXCHG].emit = atomic_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_ATOMCAS].emit = atomic_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_ATOMAND].emit = atomic_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_ATOMOR].emit = atomic_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_ATOMXOR].emit = atomic_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_ATOMUMIN].emit = atomic_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_ATOMUMAX].emit = atomic_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_ATOMIMIN].emit = atomic_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_ATOMIMAX].emit = atomic_emit;
+
+ if (params->gs_iface) {
/* There's no specific value for this because it should always
* be set, but apps using ext_geometry_shader4 quite often
* were forgetting so we're using MAX_VERTEX_VARYING from
/* inputs are always indirect with gs */
bld.indirect_files |= (1 << TGSI_FILE_INPUT);
- bld.gs_iface = gs_iface;
+ bld.gs_iface = params->gs_iface;
bld.bld_base.emit_fetch_funcs[TGSI_FILE_INPUT] = emit_fetch_gs_input;
bld.bld_base.op_actions[TGSI_OPCODE_EMIT].emit = emit_vertex;
bld.bld_base.op_actions[TGSI_OPCODE_ENDPRIM].emit = end_primitive;
max_output_vertices =
- info->properties[TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES];
+ params->info->properties[TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES];
if (!max_output_vertices)
max_output_vertices = 32;
lp_exec_mask_init(&bld.exec_mask, &bld.bld_base.int_bld);
- bld.system_values = *system_values;
+ bld.system_values = *params->system_values;
lp_build_tgsi_llvm(&bld.bld_base, tokens);