};
struct nir_to_llvm_context {
+ struct ac_llvm_context ac;
const struct ac_nir_compiler_options *options;
struct ac_shader_variant_info *shader_info;
unsigned range_md_kind;
unsigned uniform_md_kind;
- unsigned fpmath_md_kind;
unsigned invariant_load_md_kind;
LLVMValueRef empty_md;
- LLVMValueRef fpmath_md_2p5_ulp;
gl_shader_stage stage;
LLVMValueRef lds;
bool has_offset;
};
-enum ac_func_attr {
- AC_FUNC_ATTR_ALWAYSINLINE = (1 << 0),
- AC_FUNC_ATTR_BYVAL = (1 << 1),
- AC_FUNC_ATTR_INREG = (1 << 2),
- AC_FUNC_ATTR_NOALIAS = (1 << 3),
- AC_FUNC_ATTR_NOUNWIND = (1 << 4),
- AC_FUNC_ATTR_READNONE = (1 << 5),
- AC_FUNC_ATTR_READONLY = (1 << 6),
- AC_FUNC_ATTR_LAST = (1 << 7)
-};
-
-#if HAVE_LLVM < 0x0400
-static LLVMAttribute ac_attr_to_llvm_attr(enum ac_func_attr attr)
-{
- switch (attr) {
- case AC_FUNC_ATTR_ALWAYSINLINE: return LLVMAlwaysInlineAttribute;
- case AC_FUNC_ATTR_BYVAL: return LLVMByValAttribute;
- case AC_FUNC_ATTR_INREG: return LLVMInRegAttribute;
- case AC_FUNC_ATTR_NOALIAS: return LLVMNoAliasAttribute;
- case AC_FUNC_ATTR_NOUNWIND: return LLVMNoUnwindAttribute;
- case AC_FUNC_ATTR_READNONE: return LLVMReadNoneAttribute;
- case AC_FUNC_ATTR_READONLY: return LLVMReadOnlyAttribute;
- default:
- fprintf(stderr, "Unhandled function attribute: %x\n", attr);
- return 0;
- }
-}
-
-#else
-
-static const char *attr_to_str(enum ac_func_attr attr)
-{
- switch (attr) {
- case AC_FUNC_ATTR_ALWAYSINLINE: return "alwaysinline";
- case AC_FUNC_ATTR_BYVAL: return "byval";
- case AC_FUNC_ATTR_INREG: return "inreg";
- case AC_FUNC_ATTR_NOALIAS: return "noalias";
- case AC_FUNC_ATTR_NOUNWIND: return "nounwind";
- case AC_FUNC_ATTR_READNONE: return "readnone";
- case AC_FUNC_ATTR_READONLY: return "readonly";
- default:
- fprintf(stderr, "Unhandled function attribute: %x\n", attr);
- return 0;
- }
-}
-
-#endif
-
-static void
-ac_add_function_attr(LLVMValueRef function,
- int attr_idx,
- enum ac_func_attr attr)
-{
-
-#if HAVE_LLVM < 0x0400
- LLVMAttribute llvm_attr = ac_attr_to_llvm_attr(attr);
- if (attr_idx == -1) {
- LLVMAddFunctionAttr(function, llvm_attr);
- } else {
- LLVMAddAttribute(LLVMGetParam(function, attr_idx - 1), llvm_attr);
- }
-#else
- LLVMContextRef context = LLVMGetModuleContext(LLVMGetGlobalParent(function));
- const char *attr_name = attr_to_str(attr);
- unsigned kind_id = LLVMGetEnumAttributeKindForName(attr_name,
- strlen(attr_name));
- LLVMAttributeRef llvm_attr = LLVMCreateEnumAttribute(context, kind_id, 0);
- LLVMAddAttributeAtIndex(function, attr_idx, llvm_attr);
-#endif
-}
-
-static LLVMValueRef
-emit_llvm_intrinsic(struct nir_to_llvm_context *ctx, const char *name,
- LLVMTypeRef return_type, LLVMValueRef *params,
- unsigned param_count, unsigned attr_mask);
static LLVMValueRef get_sampler_desc(struct nir_to_llvm_context *ctx,
nir_deref_var *deref,
enum desc_type desc_type);
create_llvm_function(LLVMContextRef ctx, LLVMModuleRef module,
LLVMBuilderRef builder, LLVMTypeRef *return_types,
unsigned num_return_elems, LLVMTypeRef *param_types,
- unsigned param_count, unsigned array_params,
+ unsigned param_count, unsigned array_params_mask,
unsigned sgpr_params, bool unsafe_math)
{
LLVMTypeRef main_function_type, ret_type;
LLVMSetFunctionCallConv(main_function, RADEON_LLVM_AMDGPU_CS);
for (unsigned i = 0; i < sgpr_params; ++i) {
- if (i < array_params) {
+ if (array_params_mask & (1 << i)) {
LLVMValueRef P = LLVMGetParam(main_function, i);
ac_add_function_attr(main_function, i + 1, AC_FUNC_ATTR_BYVAL);
ac_add_attr_dereferenceable(P, UINT64_MAX);
{
LLVMTypeRef arg_types[23];
unsigned arg_idx = 0;
- unsigned array_count = 0;
+ unsigned array_params_mask = 0;
unsigned sgpr_count = 0, user_sgpr_count;
unsigned i;
unsigned num_sets = ctx->options->layout ? ctx->options->layout->num_sets : 0;
/* 1 for each descriptor set */
for (unsigned i = 0; i < num_sets; ++i) {
if (ctx->options->layout->set[i].layout->shader_stages & (1 << ctx->stage)) {
+ array_params_mask |= (1 << arg_idx);
arg_types[arg_idx++] = const_array(ctx->i8, 1024 * 1024);
}
}
if (need_push_constants) {
/* 1 for push constants and dynamic descriptors */
+ array_params_mask |= (1 << arg_idx);
arg_types[arg_idx++] = const_array(ctx->i8, 1024 * 1024);
}
- array_count = arg_idx;
switch (ctx->stage) {
case MESA_SHADER_COMPUTE:
arg_types[arg_idx++] = LLVMVectorType(ctx->i32, 3); /* grid size */
ctx->main_function = create_llvm_function(
ctx->context, ctx->module, ctx->builder, NULL, 0, arg_types,
- arg_idx, array_count, sgpr_count, ctx->options->unsafe_math);
+ arg_idx, array_params_mask, sgpr_count, ctx->options->unsafe_math);
set_llvm_calling_convention(ctx->main_function, ctx->stage);
LLVMGetMDKindIDInContext(ctx->context, "amdgpu.uniform", 14);
ctx->empty_md = LLVMMDNodeInContext(ctx->context, NULL, 0);
- ctx->fpmath_md_kind = LLVMGetMDKindIDInContext(ctx->context, "fpmath", 6);
-
args[0] = LLVMConstReal(ctx->f32, 2.5);
- ctx->fpmath_md_2p5_ulp = LLVMMDNodeInContext(ctx->context, args, 1);
}
static int get_llvm_num_components(LLVMValueRef value)
return LLVMBuildShuffleVector(ctx->builder, value, value, swizzle, "");
}
-static LLVMValueRef
-build_gather_values_extended(struct nir_to_llvm_context *ctx,
- LLVMValueRef *values,
- unsigned value_count,
- unsigned value_stride,
- bool load)
-{
- LLVMBuilderRef builder = ctx->builder;
- LLVMValueRef vec;
- unsigned i;
-
-
- if (value_count == 1) {
- if (load)
- return LLVMBuildLoad(builder, values[0], "");
- return values[0];
- }
-
- for (i = 0; i < value_count; i++) {
- LLVMValueRef value = values[i * value_stride];
- if (load)
- value = LLVMBuildLoad(builder, value, "");
-
- if (!i)
- vec = LLVMGetUndef( LLVMVectorType(LLVMTypeOf(value), value_count));
- LLVMValueRef index = LLVMConstInt(ctx->i32, i, false);
- vec = LLVMBuildInsertElement(builder, vec, value, index, "");
- }
- return vec;
-}
-
-
static void
build_store_values_extended(struct nir_to_llvm_context *ctx,
LLVMValueRef *values,
}
}
-static LLVMValueRef
-build_gather_values(struct nir_to_llvm_context *ctx,
- LLVMValueRef *values,
- unsigned value_count)
-{
- return build_gather_values_extended(ctx, values, value_count, 1, false);
-}
-
static LLVMTypeRef get_def_type(struct nir_to_llvm_context *ctx,
nir_ssa_def *def)
{
masks[0], "");
} else if (src_components == 1 && num_components > 1) {
LLVMValueRef values[] = {value, value, value, value};
- value = build_gather_values(ctx, values, num_components);
+ value = ac_build_gather_values(&ctx->ac, values, num_components);
} else {
LLVMValueRef swizzle = LLVMConstVector(masks, num_components);
value = LLVMBuildShuffleVector(ctx->builder, value, value,
LLVMValueRef params[] = {
to_float(ctx, src0),
};
- return emit_llvm_intrinsic(ctx, intrin, ctx->f32, params, 1, AC_FUNC_ATTR_READNONE);
+ return ac_emit_llvm_intrinsic(&ctx->ac, intrin, ctx->f32, params, 1, AC_FUNC_ATTR_READNONE);
}
static LLVMValueRef emit_intrin_2f_param(struct nir_to_llvm_context *ctx,
to_float(ctx, src0),
to_float(ctx, src1),
};
- return emit_llvm_intrinsic(ctx, intrin, ctx->f32, params, 2, AC_FUNC_ATTR_READNONE);
+ return ac_emit_llvm_intrinsic(&ctx->ac, intrin, ctx->f32, params, 2, AC_FUNC_ATTR_READNONE);
}
static LLVMValueRef emit_intrin_3f_param(struct nir_to_llvm_context *ctx,
to_float(ctx, src1),
to_float(ctx, src2),
};
- return emit_llvm_intrinsic(ctx, intrin, ctx->f32, params, 3, AC_FUNC_ATTR_READNONE);
+ return ac_emit_llvm_intrinsic(&ctx->ac, intrin, ctx->f32, params, 3, AC_FUNC_ATTR_READNONE);
}
static LLVMValueRef emit_bcsel(struct nir_to_llvm_context *ctx,
*/
LLVMConstInt(ctx->i32, 1, false),
};
- return emit_llvm_intrinsic(ctx, "llvm.cttz.i32", ctx->i32, params, 2, AC_FUNC_ATTR_READNONE);
+ return ac_emit_llvm_intrinsic(&ctx->ac, "llvm.cttz.i32", ctx->i32, params, 2, AC_FUNC_ATTR_READNONE);
}
static LLVMValueRef emit_ifind_msb(struct nir_to_llvm_context *ctx,
LLVMValueRef src0)
{
- LLVMValueRef msb = emit_llvm_intrinsic(ctx, "llvm.AMDGPU.flbit.i32",
+ LLVMValueRef msb = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.AMDGPU.flbit.i32",
ctx->i32, &src0, 1,
AC_FUNC_ATTR_READNONE);
src0,
ctx->i32one,
};
- LLVMValueRef msb = emit_llvm_intrinsic(ctx, "llvm.ctlz.i32",
+ LLVMValueRef msb = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.ctlz.i32",
ctx->i32, args, ARRAY_SIZE(args),
AC_FUNC_ATTR_READNONE);
LLVMValueRef params[] = {
fsrc0,
};
- LLVMValueRef floor = emit_llvm_intrinsic(ctx, intr,
+ LLVMValueRef floor = ac_emit_llvm_intrinsic(&ctx->ac, intr,
ctx->f32, params, 1,
AC_FUNC_ATTR_READNONE);
return LLVMBuildFSub(ctx->builder, fsrc0, floor, "");
ret_type = LLVMStructTypeInContext(ctx->context, types,
2, true);
- res = emit_llvm_intrinsic(ctx, intrin, ret_type,
+ res = ac_emit_llvm_intrinsic(&ctx->ac, intrin, ret_type,
params, 2, AC_FUNC_ATTR_READNONE);
res = LLVMBuildExtractValue(ctx->builder, res, 1, "");
{
LLVMValueRef result;
LLVMValueRef icond = LLVMBuildICmp(ctx->builder, LLVMIntEQ, srcs[2], LLVMConstInt(ctx->i32, 32, false), "");
- result = emit_llvm_intrinsic(ctx, intrin, ctx->i32, srcs, 3, AC_FUNC_ATTR_READNONE);
+ result = ac_emit_llvm_intrinsic(&ctx->ac, intrin, ctx->i32, srcs, 3, AC_FUNC_ATTR_READNONE);
result = LLVMBuildSelect(ctx->builder, icond, srcs[0], result, "");
return result;
LLVMValueRef tid_args[2];
tid_args[0] = LLVMConstInt(ctx->i32, 0xffffffff, false);
tid_args[1] = ctx->i32zero;
- tid_args[1] = emit_llvm_intrinsic(ctx,
+ tid_args[1] = ac_emit_llvm_intrinsic(&ctx->ac,
"llvm.amdgcn.mbcnt.lo", ctx->i32,
tid_args, 2, AC_FUNC_ATTR_READNONE);
- tid = emit_llvm_intrinsic(ctx,
+ tid = ac_emit_llvm_intrinsic(&ctx->ac,
"llvm.amdgcn.mbcnt.hi", ctx->i32,
tid_args, 2, AC_FUNC_ATTR_READNONE);
set_range_metadata(ctx, tid, 0, 64);
args[0] = LLVMBuildMul(ctx->builder, tl_tid,
LLVMConstInt(ctx->i32, 4, false), "");
args[1] = src0;
- tl = emit_llvm_intrinsic(ctx, "llvm.amdgcn.ds.bpermute",
+ tl = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.ds.bpermute",
ctx->i32, args, 2,
AC_FUNC_ATTR_READNONE);
args[0] = LLVMBuildMul(ctx->builder, trbl_tid,
LLVMConstInt(ctx->i32, 4, false), "");
- trbl = emit_llvm_intrinsic(ctx, "llvm.amdgcn.ds.bpermute",
+ trbl = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.ds.bpermute",
ctx->i32, args, 2,
AC_FUNC_ATTR_READNONE);
} else {
result[i] = emit_ddxy(ctx, nir_op_fddx, a);
result[2+i] = emit_ddxy(ctx, nir_op_fddy, a);
}
- return build_gather_values(ctx, result, 4);
-}
-
-static LLVMValueRef emit_fdiv(struct nir_to_llvm_context *ctx,
- LLVMValueRef num,
- LLVMValueRef den)
-{
- LLVMValueRef ret = LLVMBuildFDiv(ctx->builder, num, den, "");
-
- if (!LLVMIsConstant(ret))
- LLVMSetMetadata(ret, ctx->fpmath_md_kind, ctx->fpmath_md_2p5_ulp);
- return ret;
+ return ac_build_gather_values(&ctx->ac, result, 4);
}
static void visit_alu(struct nir_to_llvm_context *ctx, nir_alu_instr *instr)
case nir_op_fmod:
src[0] = to_float(ctx, src[0]);
src[1] = to_float(ctx, src[1]);
- result = emit_fdiv(ctx, src[0], src[1]);
+ result = ac_emit_fdiv(&ctx->ac, src[0], src[1]);
result = emit_intrin_1f_param(ctx, "llvm.floor.f32", result);
result = LLVMBuildFMul(ctx->builder, src[1] , result, "");
result = LLVMBuildFSub(ctx->builder, src[0], result, "");
case nir_op_fdiv:
src[0] = to_float(ctx, src[0]);
src[1] = to_float(ctx, src[1]);
- result = emit_fdiv(ctx, src[0], src[1]);
+ result = ac_emit_fdiv(&ctx->ac, src[0], src[1]);
break;
case nir_op_frcp:
src[0] = to_float(ctx, src[0]);
- result = emit_fdiv(ctx, ctx->f32one, src[0]);
+ result = ac_emit_fdiv(&ctx->ac, ctx->f32one, src[0]);
break;
case nir_op_iand:
result = LLVMBuildAnd(ctx->builder, src[0], src[1], "");
break;
case nir_op_frsq:
result = emit_intrin_1f_param(ctx, "llvm.sqrt.f32", src[0]);
- result = emit_fdiv(ctx, ctx->f32one, result);
+ result = ac_emit_fdiv(&ctx->ac, ctx->f32one, result);
break;
case nir_op_fpow:
result = emit_intrin_2f_param(ctx, "llvm.pow.f32", src[0], src[1]);
result = emit_bitfield_insert(ctx, src[0], src[1], src[2], src[3]);
break;
case nir_op_bitfield_reverse:
- result = emit_llvm_intrinsic(ctx, "llvm.bitreverse.i32", ctx->i32, src, 1, AC_FUNC_ATTR_READNONE);
+ result = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.bitreverse.i32", ctx->i32, src, 1, AC_FUNC_ATTR_READNONE);
break;
case nir_op_bit_count:
- result = emit_llvm_intrinsic(ctx, "llvm.ctpop.i32", ctx->i32, src, 1, AC_FUNC_ATTR_READNONE);
+ result = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.ctpop.i32", ctx->i32, src, 1, AC_FUNC_ATTR_READNONE);
break;
case nir_op_vec2:
case nir_op_vec3:
case nir_op_vec4:
for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++)
src[i] = to_integer(ctx, src[i]);
- result = build_gather_values(ctx, src, num_components);
+ result = ac_build_gather_values(&ctx->ac, src, num_components);
break;
case nir_op_f2i:
src[0] = to_float(ctx, src[0]);
LLVMPointerType(type, addr_space), "");
}
-static LLVMValueRef
-emit_llvm_intrinsic(struct nir_to_llvm_context *ctx, const char *name,
- LLVMTypeRef return_type, LLVMValueRef *params,
- unsigned param_count, unsigned attrib_mask)
-{
- LLVMValueRef function;
-
- function = LLVMGetNamedFunction(ctx->module, name);
- if (!function) {
- LLVMTypeRef param_types[32], function_type;
- unsigned i;
-
- assert(param_count <= 32);
-
- for (i = 0; i < param_count; ++i) {
- assert(params[i]);
- param_types[i] = LLVMTypeOf(params[i]);
- }
- function_type =
- LLVMFunctionType(return_type, param_types, param_count, 0);
- function = LLVMAddFunction(ctx->module, name, function_type);
-
- LLVMSetFunctionCallConv(function, LLVMCCallConv);
- LLVMSetLinkage(function, LLVMExternalLinkage);
-
- attrib_mask |= AC_FUNC_ATTR_NOUNWIND;
- while (attrib_mask) {
- enum ac_func_attr attr = 1u << u_bit_scan(&attrib_mask);
- ac_add_function_attr(function, -1, attr);
- }
- }
- return LLVMBuildCall(ctx->builder, function, params, param_count, "");
-}
-
static LLVMValueRef
get_buffer_size(struct nir_to_llvm_context *ctx, LLVMValueRef descriptor, bool in_elements)
{
txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0, 0); /* slc */
txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0, 0); /* tfe */
txq_args[txq_arg_count++] = LLVMConstInt(ctx->i32, 0, 0); /* lwe */
- size = emit_llvm_intrinsic(ctx, "llvm.SI.getresinfo.i32", ctx->v4i32,
+ size = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.getresinfo.i32", ctx->v4i32,
txq_args, txq_arg_count,
AC_FUNC_ATTR_READNONE);
half_texel[c] = LLVMBuildExtractElement(ctx->builder, size,
LLVMConstInt(ctx->i32, c, false), "");
half_texel[c] = LLVMBuildUIToFP(ctx->builder, half_texel[c], ctx->f32, "");
- half_texel[c] = emit_fdiv(ctx, ctx->f32one, half_texel[c]);
+ half_texel[c] = ac_emit_fdiv(&ctx->ac, ctx->f32one, half_texel[c]);
half_texel[c] = LLVMBuildFMul(ctx->builder, half_texel[c],
LLVMConstReal(ctx->f32, -0.5), "");
}
}
tinfo->args[0] = coord;
- return emit_llvm_intrinsic(ctx, intr_name, tinfo->dst_type, tinfo->args, tinfo->arg_count,
+ return ac_emit_llvm_intrinsic(&ctx->ac, intr_name, tinfo->dst_type, tinfo->args, tinfo->arg_count,
AC_FUNC_ATTR_READNONE | AC_FUNC_ATTR_NOUNWIND);
}
(int)has_offset + (int)is_shadow);
}
}
- return emit_llvm_intrinsic(ctx, intr_name, tinfo->dst_type, tinfo->args, tinfo->arg_count,
+ return ac_emit_llvm_intrinsic(&ctx->ac, intr_name, tinfo->dst_type, tinfo->args, tinfo->arg_count,
AC_FUNC_ATTR_READNONE | AC_FUNC_ATTR_NOUNWIND);
}
static LLVMValueRef visit_load_push_constant(struct nir_to_llvm_context *ctx,
nir_intrinsic_instr *instr)
{
- LLVMValueRef ptr;
+ LLVMValueRef ptr, addr;
+
+ addr = LLVMConstInt(ctx->i32, nir_intrinsic_base(instr), 0);
+ addr = LLVMBuildAdd(ctx->builder, addr, get_src(ctx, instr->src[0]), "");
- ptr = build_gep0(ctx, ctx->push_constants, get_src(ctx, instr->src[0]));
+ ptr = build_gep0(ctx, ctx->push_constants, addr);
ptr = cast_ptr(ctx, ptr, get_def_type(ctx, &instr->dest.ssa));
return LLVMBuildLoad(ctx->builder, ptr, "");
}
params[0] = data;
params[3] = offset;
- emit_llvm_intrinsic(ctx, store_name,
- LLVMVoidTypeInContext(ctx->context), params, 6, 0);
+ ac_emit_llvm_intrinsic(&ctx->ac, store_name,
+ ctx->voidt, params, 6, 0);
}
}
abort();
}
- return emit_llvm_intrinsic(ctx, name, ctx->i32, params, arg_count, 0);
+ return ac_emit_llvm_intrinsic(&ctx->ac, name, ctx->i32, params, arg_count, 0);
}
static LLVMValueRef visit_load_buffer(struct nir_to_llvm_context *ctx,
};
LLVMValueRef ret =
- emit_llvm_intrinsic(ctx, load_name, data_type, params, 5, 0);
+ ac_emit_llvm_intrinsic(&ctx->ac, load_name, data_type, params, 5, 0);
if (instr->num_components == 3)
ret = trim_vector(ctx, ret, 3);
static LLVMValueRef visit_load_ubo_buffer(struct nir_to_llvm_context *ctx,
nir_intrinsic_instr *instr)
{
- const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic];
- const char *load_name;
- LLVMTypeRef data_type = ctx->f32;
LLVMValueRef results[4], ret;
LLVMValueRef rsrc = get_src(ctx, instr->src[0]);
LLVMValueRef offset = get_src(ctx, instr->src[1]);
LLVMBuildAdd(ctx->builder, LLVMConstInt(ctx->i32, 4 * i, 0),
offset, "")
};
- results[i] = emit_llvm_intrinsic(ctx, "llvm.SI.load.const", ctx->f32,
+ results[i] = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.load.const", ctx->f32,
params, 2, AC_FUNC_ATTR_READNONE);
}
- ret = build_gather_values(ctx, results, instr->num_components);
+ ret = ac_build_gather_values(&ctx->ac, results, instr->num_components);
return LLVMBuildBitCast(ctx->builder, ret,
get_def_type(ctx, &instr->dest.ssa), "");
}
unsigned count = glsl_count_attribute_slots(
instr->variables[0]->var->type,
ctx->stage == MESA_SHADER_VERTEX);
- LLVMValueRef tmp_vec = build_gather_values_extended(
- ctx, ctx->inputs + idx + chan, count,
+ LLVMValueRef tmp_vec = ac_build_gather_values_extended(
+ &ctx->ac, ctx->inputs + idx + chan, count,
4, false);
values[chan] = LLVMBuildExtractElement(ctx->builder,
} else
values[chan] = ctx->inputs[idx + chan + const_index * 4];
}
- return to_integer(ctx, build_gather_values(ctx, values, ve));
+ return to_integer(ctx, ac_build_gather_values(&ctx->ac, values, ve));
break;
case nir_var_local:
radv_get_deref_offset(ctx, &instr->variables[0]->deref, false,
if (indir_index) {
unsigned count = glsl_count_attribute_slots(
instr->variables[0]->var->type, false);
- LLVMValueRef tmp_vec = build_gather_values_extended(
- ctx, ctx->locals + idx + chan, count,
+ LLVMValueRef tmp_vec = ac_build_gather_values_extended(
+ &ctx->ac, ctx->locals + idx + chan, count,
4, true);
values[chan] = LLVMBuildExtractElement(ctx->builder,
values[chan] = LLVMBuildLoad(ctx->builder, ctx->locals[idx + chan + const_index * 4], "");
}
}
- return to_integer(ctx, build_gather_values(ctx, values, ve));
+ return to_integer(ctx, ac_build_gather_values(&ctx->ac, values, ve));
case nir_var_shader_out:
radv_get_deref_offset(ctx, &instr->variables[0]->deref, false,
&const_index, &indir_index);
if (indir_index) {
unsigned count = glsl_count_attribute_slots(
instr->variables[0]->var->type, false);
- LLVMValueRef tmp_vec = build_gather_values_extended(
- ctx, ctx->outputs + idx + chan, count,
+ LLVMValueRef tmp_vec = ac_build_gather_values_extended(
+ &ctx->ac, ctx->outputs + idx + chan, count,
4, true);
values[chan] = LLVMBuildExtractElement(ctx->builder,
"");
}
}
- return to_integer(ctx, build_gather_values(ctx, values, ve));
+ return to_integer(ctx, ac_build_gather_values(&ctx->ac, values, ve));
case nir_var_shared: {
radv_get_deref_offset(ctx, &instr->variables[0]->deref, false,
&const_index, &indir_index);
LLVMValueRef ptr = get_shared_memory_ptr(ctx, idx, ctx->i32);
LLVMValueRef derived_ptr;
- LLVMValueRef index = ctx->i32zero;
- if (indir_index)
- index = LLVMBuildAdd(ctx->builder, index, indir_index, "");
- derived_ptr = LLVMBuildGEP(ctx->builder, ptr, &index, 1, "");
- return to_integer(ctx, LLVMBuildLoad(ctx->builder, derived_ptr, ""));
- break;
+ for (unsigned chan = 0; chan < ve; chan++) {
+ LLVMValueRef index = LLVMConstInt(ctx->i32, chan, false);
+ if (indir_index)
+ index = LLVMBuildAdd(ctx->builder, index, indir_index, "");
+ derived_ptr = LLVMBuildGEP(ctx->builder, ptr, &index, 1, "");
+ values[chan] = LLVMBuildLoad(ctx->builder, derived_ptr, "");
+ }
+ return to_integer(ctx, ac_build_gather_values(&ctx->ac, values, ve));
}
default:
break;
if (indir_index) {
unsigned count = glsl_count_attribute_slots(
instr->variables[0]->var->type, false);
- LLVMValueRef tmp_vec = build_gather_values_extended(
- ctx, ctx->outputs + idx + chan, count,
+ LLVMValueRef tmp_vec = ac_build_gather_values_extended(
+ &ctx->ac, ctx->outputs + idx + chan, count,
stride, true);
if (get_llvm_num_components(tmp_vec) > 1) {
if (indir_index) {
unsigned count = glsl_count_attribute_slots(
instr->variables[0]->var->type, false);
- LLVMValueRef tmp_vec = build_gather_values_extended(
- ctx, ctx->locals + idx + chan, count,
+ LLVMValueRef tmp_vec = ac_build_gather_values_extended(
+ &ctx->ac, ctx->locals + idx + chan, count,
4, true);
tmp_vec = LLVMBuildInsertElement(ctx->builder, tmp_vec,
&const_index, &indir_index);
ptr = get_shared_memory_ptr(ctx, idx, ctx->i32);
- LLVMValueRef index = ctx->i32zero;
LLVMValueRef derived_ptr;
- if (indir_index)
- index = LLVMBuildAdd(ctx->builder, index, indir_index, "");
- derived_ptr = LLVMBuildGEP(ctx->builder, ptr, &index, 1, "");
- LLVMBuildStore(ctx->builder,
- to_integer(ctx, src), derived_ptr);
+ for (unsigned chan = 0; chan < 4; chan++) {
+ if (!(writemask & (1 << chan)))
+ continue;
+
+ LLVMValueRef index = LLVMConstInt(ctx->i32, chan, false);
+
+ if (get_llvm_num_components(src) == 1)
+ value = src;
+ else
+ value = LLVMBuildExtractElement(ctx->builder, src,
+ LLVMConstInt(ctx->i32,
+ chan, false),
+ "");
+
+ if (indir_index)
+ index = LLVMBuildAdd(ctx->builder, index, indir_index, "");
+
+ derived_ptr = LLVMBuildGEP(ctx->builder, ptr, &index, 1, "");
+ LLVMBuildStore(ctx->builder,
+ to_integer(ctx, value), derived_ptr);
+ }
break;
}
default:
return array ? 2 : 1;
case GLSL_SAMPLER_DIM_2D:
return array ? 3 : 2;
+ case GLSL_SAMPLER_DIM_MS:
+ return array ? 4 : 3;
case GLSL_SAMPLER_DIM_3D:
case GLSL_SAMPLER_DIM_CUBE:
return 3;
case GLSL_SAMPLER_DIM_RECT:
case GLSL_SAMPLER_DIM_SUBPASS:
return 2;
+ case GLSL_SAMPLER_DIM_SUBPASS_MS:
+ return 3;
default:
break;
}
};
LLVMValueRef res;
int count;
- count = image_type_to_components_count(glsl_get_sampler_dim(type),
+ enum glsl_sampler_dim dim = glsl_get_sampler_dim(type);
+ bool is_ms = (dim == GLSL_SAMPLER_DIM_MS ||
+ dim == GLSL_SAMPLER_DIM_SUBPASS_MS);
+
+ count = image_type_to_components_count(dim,
glsl_sampler_type_is_array(type));
if (count == 1) {
res = src0;
} else {
int chan;
+ if (is_ms)
+ count--;
for (chan = 0; chan < count; ++chan) {
coords[chan] = LLVMBuildExtractElement(ctx->builder, src0, masks[chan], "");
}
for (chan = 0; chan < count; ++chan)
coords[chan] = LLVMBuildAdd(ctx->builder, coords[chan], LLVMBuildFPToUI(ctx->builder, ctx->frag_pos[chan], ctx->i32, ""), "");
}
+ if (is_ms) {
+ coords[count] = llvm_extract_elem(ctx, get_src(ctx, instr->src[1]), 0);
+ count++;
+ }
+
if (count == 3) {
coords[3] = LLVMGetUndef(ctx->i32);
count = 4;
}
- res = build_gather_values(ctx, coords, count);
+ res = ac_build_gather_values(&ctx->ac, coords, count);
}
return res;
}
params[2] = LLVMConstInt(ctx->i32, 0, false); /* voffset */
params[3] = LLVMConstInt(ctx->i1, 0, false); /* glc */
params[4] = LLVMConstInt(ctx->i1, 0, false); /* slc */
- res = emit_llvm_intrinsic(ctx, "llvm.amdgcn.buffer.load.format.v4f32", ctx->v4f32,
+ res = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.buffer.load.format.v4f32", ctx->v4f32,
params, 5, 0);
res = trim_vector(ctx, res, instr->dest.ssa.num_components);
LLVMTypeOf(params[1]), /* rsrc */
intrinsic_name, sizeof(intrinsic_name));
- res = emit_llvm_intrinsic(ctx, intrinsic_name, ctx->v4f32,
+ res = ac_emit_llvm_intrinsic(&ctx->ac, intrinsic_name, ctx->v4f32,
params, 7, AC_FUNC_ATTR_READONLY);
}
return to_integer(ctx, res);
params[3] = LLVMConstInt(ctx->i32, 0, false); /* voffset */
params[4] = i1false; /* glc */
params[5] = i1false; /* slc */
- emit_llvm_intrinsic(ctx, "llvm.amdgcn.buffer.store.format.v4f32", ctx->voidt,
+ ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.buffer.store.format.v4f32", ctx->voidt,
params, 6, 0);
} else {
bool is_da = glsl_sampler_type_is_array(type) ||
LLVMTypeOf(params[2]), /* rsrc */
intrinsic_name, sizeof(intrinsic_name));
- emit_llvm_intrinsic(ctx, intrinsic_name, ctx->voidt,
+ ac_emit_llvm_intrinsic(&ctx->ac, intrinsic_name, ctx->voidt,
params, 8, 0);
}
snprintf(intrinsic_name, sizeof(intrinsic_name),
"%s.%s.%s", base_name, atomic_name, coords_type);
- return emit_llvm_intrinsic(ctx, intrinsic_name, ctx->i32, params, param_count, 0);
+ return ac_emit_llvm_intrinsic(&ctx->ac, intrinsic_name, ctx->i32, params, param_count, 0);
}
static LLVMValueRef visit_image_size(struct nir_to_llvm_context *ctx,
params[8] = ctx->i32zero;
params[9] = ctx->i32zero;
- res = emit_llvm_intrinsic(ctx, "llvm.SI.getresinfo.i32", ctx->v4i32,
+ res = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.getresinfo.i32", ctx->v4i32,
params, 10, AC_FUNC_ATTR_READNONE);
if (glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_CUBE &&
LLVMValueRef args[1] = {
LLVMConstInt(ctx->i32, 0xf70, false),
};
- emit_llvm_intrinsic(ctx, "llvm.amdgcn.s.waitcnt",
+ ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.s.waitcnt",
ctx->voidt, args, 1, 0);
}
static void emit_barrier(struct nir_to_llvm_context *ctx)
{
// TODO tess
- emit_llvm_intrinsic(ctx, "llvm.amdgcn.s.barrier",
+ ac_emit_llvm_intrinsic(&ctx->ac, "llvm.amdgcn.s.barrier",
ctx->voidt, NULL, 0, 0);
}
cond = LLVMBuildSelect(ctx->builder, cond,
LLVMConstReal(ctx->f32, -1.0f),
ctx->f32zero, "");
- emit_llvm_intrinsic(ctx, "llvm.AMDGPU.kill",
- LLVMVoidTypeInContext(ctx->context),
- &cond, 1, 0);
+ ac_emit_llvm_intrinsic(&ctx->ac, "llvm.AMDGPU.kill",
+ ctx->voidt,
+ &cond, 1, 0);
}
static LLVMValueRef
result[0] = build_indexed_load_const(ctx, ctx->sample_positions, offset0);
result[1] = build_indexed_load_const(ctx, ctx->sample_positions, offset1);
- return build_gather_values(ctx, result, 2);
+ return ac_build_gather_values(&ctx->ac, result, 2);
}
static LLVMValueRef load_sample_pos(struct nir_to_llvm_context *ctx)
values[0] = emit_ffract(ctx, ctx->frag_pos[0]);
values[1] = emit_ffract(ctx, ctx->frag_pos[1]);
- return build_gather_values(ctx, values, 2);
+ return ac_build_gather_values(&ctx->ac, values, 2);
}
static LLVMValueRef visit_interp(struct nir_to_llvm_context *ctx,
ij_out[i] = LLVMBuildBitCast(ctx->builder,
temp2, ctx->i32, "");
}
- interp_param = build_gather_values(ctx, ij_out, 2);
+ interp_param = ac_build_gather_values(&ctx->ac, ij_out, 2);
}
intr_name = interp_param ? "llvm.SI.fs.interp" : "llvm.SI.fs.constant";
args[1] = attr_number;
args[2] = ctx->prim_mask;
args[3] = interp_param;
- result[chan] = emit_llvm_intrinsic(ctx, intr_name,
+ result[chan] = ac_emit_llvm_intrinsic(&ctx->ac, intr_name,
ctx->f32, args, args[3] ? 4 : 3,
AC_FUNC_ATTR_READNONE);
}
- return build_gather_values(ctx, result, 2);
+ return ac_build_gather_values(&ctx->ac, result, 2);
}
static void visit_intrinsic(struct nir_to_llvm_context *ctx,
break;
case nir_intrinsic_discard:
ctx->shader_info->fs.can_discard = true;
- emit_llvm_intrinsic(ctx, "llvm.AMDGPU.kilp",
- LLVMVoidTypeInContext(ctx->context),
- NULL, 0, 0);
+ ac_emit_llvm_intrinsic(&ctx->ac, "llvm.AMDGPU.kilp",
+ ctx->voidt,
+ NULL, 0, 0);
break;
case nir_intrinsic_discard_if:
emit_discard_if(ctx, instr);
type = ctx->v4i32;
type_size = 16;
break;
+ default:
+ unreachable("invalid desc_type\n");
}
if (deref->deref.child) {
param[count++] = LLVMGetUndef(ctx->i32);
if (count > 1)
- tinfo->args[0] = build_gather_values(ctx, param, count);
+ tinfo->args[0] = ac_build_gather_values(&ctx->ac, param, count);
else
tinfo->args[0] = param[0];
*fmask_ptr = get_sampler_desc(ctx, instr->texture, DESC_FMASK);
}
-static LLVMValueRef build_cube_intrinsic(struct nir_to_llvm_context *ctx,
- LLVMValueRef *in)
-{
-
- LLVMValueRef v, cube_vec;
-
- if (1) {
- LLVMTypeRef f32 = LLVMTypeOf(in[0]);
- LLVMValueRef out[4];
-
- out[0] = emit_llvm_intrinsic(ctx, "llvm.amdgcn.cubetc",
- f32, in, 3, AC_FUNC_ATTR_READNONE);
- out[1] = emit_llvm_intrinsic(ctx, "llvm.amdgcn.cubesc",
- f32, in, 3, AC_FUNC_ATTR_READNONE);
- out[2] = emit_llvm_intrinsic(ctx, "llvm.amdgcn.cubema",
- f32, in, 3, AC_FUNC_ATTR_READNONE);
- out[3] = emit_llvm_intrinsic(ctx, "llvm.amdgcn.cubeid",
- f32, in, 3, AC_FUNC_ATTR_READNONE);
-
- return build_gather_values(ctx, out, 4);
- } else {
- LLVMValueRef c[4];
- c[0] = in[0];
- c[1] = in[1];
- c[2] = in[2];
- c[3] = LLVMGetUndef(LLVMTypeOf(in[0]));
- cube_vec = build_gather_values(ctx, c, 4);
- v = emit_llvm_intrinsic(ctx, "llvm.AMDGPU.cube", LLVMTypeOf(cube_vec),
- &cube_vec, 1, AC_FUNC_ATTR_READNONE);
- }
- return v;
-}
-
-static void cube_to_2d_coords(struct nir_to_llvm_context *ctx,
- LLVMValueRef *in, LLVMValueRef *out)
-{
- LLVMValueRef coords[4];
- LLVMValueRef mad_args[3];
- LLVMValueRef v;
- LLVMValueRef tmp;
- int i;
-
- v = build_cube_intrinsic(ctx, in);
- for (i = 0; i < 4; i++)
- coords[i] = LLVMBuildExtractElement(ctx->builder, v,
- LLVMConstInt(ctx->i32, i, false), "");
-
- coords[2] = emit_llvm_intrinsic(ctx, "llvm.fabs.f32", ctx->f32,
- &coords[2], 1, AC_FUNC_ATTR_READNONE);
- coords[2] = emit_fdiv(ctx, ctx->f32one, coords[2]);
-
- mad_args[1] = coords[2];
- mad_args[2] = LLVMConstReal(ctx->f32, 1.5);
- mad_args[0] = coords[0];
-
- /* emit MAD */
- tmp = LLVMBuildFMul(ctx->builder, mad_args[0], mad_args[1], "");
- coords[0] = LLVMBuildFAdd(ctx->builder, tmp, mad_args[2], "");
-
- mad_args[0] = coords[1];
-
- /* emit MAD */
- tmp = LLVMBuildFMul(ctx->builder, mad_args[0], mad_args[1], "");
- coords[1] = LLVMBuildFAdd(ctx->builder, tmp, mad_args[2], "");
-
- /* apply xyz = yxw swizzle to cooords */
- out[0] = coords[1];
- out[1] = coords[0];
- out[2] = coords[3];
-}
-
-static void emit_prepare_cube_coords(struct nir_to_llvm_context *ctx,
- LLVMValueRef *coords_arg, int num_coords,
- bool is_deriv,
- bool is_array, LLVMValueRef *derivs_arg)
-{
- LLVMValueRef coords[4];
- int i;
- cube_to_2d_coords(ctx, coords_arg, coords);
-
- if (is_deriv && derivs_arg) {
- LLVMValueRef derivs[4];
- int axis;
-
- /* Convert cube derivatives to 2D derivatives. */
- for (axis = 0; axis < 2; axis++) {
- LLVMValueRef shifted_cube_coords[4], shifted_coords[4];
-
- /* Shift the cube coordinates by the derivatives to get
- * the cube coordinates of the "neighboring pixel".
- */
- for (i = 0; i < 3; i++)
- shifted_cube_coords[i] =
- LLVMBuildFAdd(ctx->builder, coords_arg[i],
- derivs_arg[axis*3+i], "");
- shifted_cube_coords[3] = LLVMGetUndef(ctx->f32);
-
- /* Project the shifted cube coordinates onto the face. */
- cube_to_2d_coords(ctx, shifted_cube_coords,
- shifted_coords);
-
- /* Subtract both sets of 2D coordinates to get 2D derivatives.
- * This won't work if the shifted coordinates ended up
- * in a different face.
- */
- for (i = 0; i < 2; i++)
- derivs[axis * 2 + i] =
- LLVMBuildFSub(ctx->builder, shifted_coords[i],
- coords[i], "");
- }
-
- memcpy(derivs_arg, derivs, sizeof(derivs));
- }
-
- if (is_array) {
- /* for cube arrays coord.z = coord.w(array_index) * 8 + face */
- /* coords_arg.w component - array_index for cube arrays */
- LLVMValueRef tmp = LLVMBuildFMul(ctx->builder, coords_arg[3], LLVMConstReal(ctx->f32, 8.0), "");
- coords[2] = LLVMBuildFAdd(ctx->builder, tmp, coords[2], "");
- }
-
- memcpy(coords_arg, coords, sizeof(coords));
-}
-
static void visit_tex(struct nir_to_llvm_context *ctx, nir_tex_instr *instr)
{
LLVMValueRef result = NULL;
unsigned dmask = 0xf;
LLVMValueRef address[16];
LLVMValueRef coords[5];
- LLVMValueRef coord = NULL, lod = NULL, comparitor = NULL;
+ LLVMValueRef coord = NULL, lod = NULL, comparator = NULL;
LLVMValueRef bias = NULL, offsets = NULL;
LLVMValueRef res_ptr, samp_ptr, fmask_ptr = NULL, sample_index = NULL;
LLVMValueRef ddx = NULL, ddy = NULL;
break;
case nir_tex_src_projector:
break;
- case nir_tex_src_comparitor:
- comparitor = get_src(ctx, instr->src[i].src);
+ case nir_tex_src_comparator:
+ comparator = get_src(ctx, instr->src[i].src);
break;
case nir_tex_src_offset:
offsets = get_src(ctx, instr->src[i].src);
}
/* Pack depth comparison value */
- if (instr->is_shadow && comparitor) {
- address[count++] = llvm_extract_elem(ctx, comparitor, 0);
+ if (instr->is_shadow && comparator) {
+ address[count++] = llvm_extract_elem(ctx, comparator, 0);
}
/* pack derivatives */
coords[chan] = to_float(ctx, coords[chan]);
if (instr->coord_components == 3)
coords[3] = LLVMGetUndef(ctx->f32);
- emit_prepare_cube_coords(ctx, coords, instr->coord_components, instr->op == nir_texop_txd, instr->is_array, derivs);
+ ac_prepare_cube_coords(&ctx->ac,
+ instr->op == nir_texop_txd, instr->is_array,
+ coords, derivs);
if (num_deriv_comp)
num_deriv_comp--;
}
/* This seems like a bit of a hack - but it passes Vulkan CTS with it */
if (instr->sampler_dim != GLSL_SAMPLER_DIM_3D && instr->op != nir_texop_txf) {
coords[2] = to_float(ctx, coords[2]);
- coords[2] = emit_llvm_intrinsic(ctx, "llvm.rint.f32", ctx->f32, &coords[2],
+ coords[2] = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.rint.f32", ctx->f32, &coords[2],
1, 0);
coords[2] = to_integer(ctx, coords[2]);
}
args[0] = t_list;
args[1] = LLVMConstInt(ctx->i32, 0, false);
args[2] = buffer_index;
- input = emit_llvm_intrinsic(ctx,
+ input = ac_emit_llvm_intrinsic(&ctx->ac,
"llvm.SI.vs.load.input", ctx->v4f32, args, 3,
AC_FUNC_ATTR_READNONE | AC_FUNC_ATTR_NOUNWIND);
args[1] = attr_number;
args[2] = prim_mask;
args[3] = interp_param;
- result[chan] = emit_llvm_intrinsic(ctx, intr_name,
+ result[chan] = ac_emit_llvm_intrinsic(&ctx->ac, intr_name,
ctx->f32, args, args[3] ? 4 : 3,
AC_FUNC_ATTR_READNONE | AC_FUNC_ATTR_NOUNWIND);
}
for(int i = 0; i < 3; ++i)
inputs[i] = ctx->frag_pos[i];
- inputs[3] = emit_fdiv(ctx, ctx->f32one, ctx->frag_pos[3]);
+ inputs[3] = ac_emit_fdiv(&ctx->ac, ctx->f32one, ctx->frag_pos[3]);
}
}
ctx->shader_info->fs.num_interp = index;
handle_shader_output_decl(struct nir_to_llvm_context *ctx,
struct nir_variable *variable)
{
- int idx = variable->data.location;
+ int idx = variable->data.location + variable->data.index;
unsigned attrib_count = glsl_count_attribute_slots(variable->type, false);
variable->data.driver_location = idx * 4;
si_build_alloca_undef(ctx, ctx->f32, "");
}
}
- ctx->output_mask |= ((1ull << attrib_count) - 1) << variable->data.location;
+ ctx->output_mask |= ((1ull << attrib_count) - 1) << idx;
}
static void
};
LLVMValueRef packed;
- packed = emit_llvm_intrinsic(ctx, "llvm.SI.packf16",
+ packed = ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.packf16",
ctx->i32, pack_args, 2,
AC_FUNC_ATTR_READNONE);
args[chan + 5] = packed;
}
static void
-handle_vs_outputs_post(struct nir_to_llvm_context *ctx,
- struct nir_shader *nir)
+handle_vs_outputs_post(struct nir_to_llvm_context *ctx)
{
uint32_t param_count = 0;
unsigned target;
unsigned pos_idx, num_pos_exports = 0;
LLVMValueRef args[9];
LLVMValueRef pos_args[4][9] = { { 0 } };
- LLVMValueRef psize_value = 0;
+ LLVMValueRef psize_value = NULL, layer_value = NULL, viewport_index_value = NULL;
int i;
const uint64_t clip_mask = ctx->output_mask & ((1ull << VARYING_SLOT_CLIP_DIST0) |
(1ull << VARYING_SLOT_CLIP_DIST1) |
ctx->shader_info->vs.writes_pointsize = true;
psize_value = values[0];
continue;
+ } else if (i == VARYING_SLOT_LAYER) {
+ ctx->shader_info->vs.writes_layer = true;
+ layer_value = values[0];
+ continue;
+ } else if (i == VARYING_SLOT_VIEWPORT) {
+ ctx->shader_info->vs.writes_viewport_index = true;
+ viewport_index_value = values[0];
+ continue;
} else if (i >= VARYING_SLOT_VAR0) {
ctx->shader_info->vs.export_mask |= 1u << (i - VARYING_SLOT_VAR0);
target = V_008DFC_SQ_EXP_PARAM + param_count;
memcpy(pos_args[target - V_008DFC_SQ_EXP_POS],
args, sizeof(args));
} else {
- emit_llvm_intrinsic(ctx,
- "llvm.SI.export",
- LLVMVoidTypeInContext(ctx->context),
- args, 9, 0);
+ ac_emit_llvm_intrinsic(&ctx->ac,
+ "llvm.SI.export",
+ ctx->voidt,
+ args, 9, 0);
}
}
pos_args[0][8] = ctx->f32one; /* W */
}
- if (ctx->shader_info->vs.writes_pointsize == true) {
- pos_args[1][0] = LLVMConstInt(ctx->i32, (ctx->shader_info->vs.writes_pointsize == true), false); /* writemask */
+ uint32_t mask = ((ctx->shader_info->vs.writes_pointsize == true ? 1 : 0) |
+ (ctx->shader_info->vs.writes_layer == true ? 4 : 0) |
+ (ctx->shader_info->vs.writes_viewport_index == true ? 8 : 0));
+ if (mask) {
+ pos_args[1][0] = LLVMConstInt(ctx->i32, mask, false); /* writemask */
pos_args[1][1] = ctx->i32zero; /* EXEC mask */
pos_args[1][2] = ctx->i32zero; /* last export? */
pos_args[1][3] = LLVMConstInt(ctx->i32, V_008DFC_SQ_EXP_POS + 1, false);
if (ctx->shader_info->vs.writes_pointsize == true)
pos_args[1][5] = psize_value;
+ if (ctx->shader_info->vs.writes_layer == true)
+ pos_args[1][7] = layer_value;
+ if (ctx->shader_info->vs.writes_viewport_index == true)
+ pos_args[1][8] = viewport_index_value;
}
for (i = 0; i < 4; i++) {
if (pos_args[i][0])
pos_args[i][3] = LLVMConstInt(ctx->i32, V_008DFC_SQ_EXP_POS + pos_idx++, false);
if (pos_idx == num_pos_exports)
pos_args[i][2] = ctx->i32one;
- emit_llvm_intrinsic(ctx,
- "llvm.SI.export",
- LLVMVoidTypeInContext(ctx->context),
- pos_args[i], 9, 0);
+ ac_emit_llvm_intrinsic(&ctx->ac,
+ "llvm.SI.export",
+ ctx->voidt,
+ pos_args[i], 9, 0);
}
ctx->shader_info->vs.pos_exports = num_pos_exports;
} else if (args[0] == ctx->i32zero)
return; /* unnecessary NULL export */
- emit_llvm_intrinsic(ctx, "llvm.SI.export",
+ ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.export",
ctx->voidt, args, 9, 0);
}
mask |= 0x01;
args[0] = LLVMConstInt(ctx->i32, mask, false);
- emit_llvm_intrinsic(ctx, "llvm.SI.export",
+ ac_emit_llvm_intrinsic(&ctx->ac, "llvm.SI.export",
ctx->voidt, args, 9, 0);
}
static void
-handle_fs_outputs_post(struct nir_to_llvm_context *ctx,
- struct nir_shader *nir)
+handle_fs_outputs_post(struct nir_to_llvm_context *ctx)
{
unsigned index = 0;
LLVMValueRef depth = NULL, stencil = NULL, samplemask = NULL;
}
static void
-handle_shader_outputs_post(struct nir_to_llvm_context *ctx,
- struct nir_shader *nir)
+handle_shader_outputs_post(struct nir_to_llvm_context *ctx)
{
switch (ctx->stage) {
case MESA_SHADER_VERTEX:
- handle_vs_outputs_post(ctx, nir);
+ handle_vs_outputs_post(ctx);
break;
case MESA_SHADER_FRAGMENT:
- handle_fs_outputs_post(ctx, nir);
+ handle_fs_outputs_post(ctx);
break;
default:
break;
ctx.context = LLVMContextCreate();
ctx.module = LLVMModuleCreateWithNameInContext("shader", ctx.context);
+ ac_llvm_context_init(&ctx.ac, ctx.context);
+ ctx.ac.module = ctx.module;
+
ctx.has_ds_bpermute = ctx.options->chip_class >= VI;
memset(shader_info, 0, sizeof(*shader_info));
setup_types(&ctx);
ctx.builder = LLVMCreateBuilderInContext(ctx.context);
+ ctx.ac.builder = ctx.builder;
ctx.stage = nir->stage;
for (i = 0; i < AC_UD_MAX_SETS; i++)
visit_cf_list(&ctx, &func->impl->body);
phi_post_pass(&ctx);
- handle_shader_outputs_post(&ctx, nir);
+ handle_shader_outputs_post(&ctx);
LLVMBuildRetVoid(ctx.builder);
ac_llvm_finalize_module(&ctx);
return retval;
}
-void ac_compile_nir_shader(LLVMTargetMachineRef tm,
- struct ac_shader_binary *binary,
- struct ac_shader_config *config,
- struct ac_shader_variant_info *shader_info,
- struct nir_shader *nir,
- const struct ac_nir_compiler_options *options,
- bool dump_shader)
+static void ac_compile_llvm_module(LLVMTargetMachineRef tm,
+ LLVMModuleRef llvm_module,
+ struct ac_shader_binary *binary,
+ struct ac_shader_config *config,
+ struct ac_shader_variant_info *shader_info,
+ gl_shader_stage stage,
+ bool dump_shader)
{
-
- LLVMModuleRef llvm_module = ac_translate_nir_to_llvm(tm, nir, shader_info,
- options);
if (dump_shader)
LLVMDumpModule(llvm_module);
LLVMDisposeModule(llvm_module);
LLVMContextDispose(ctx);
- if (nir->stage == MESA_SHADER_FRAGMENT) {
+ if (stage == MESA_SHADER_FRAGMENT) {
shader_info->num_input_vgprs = 0;
if (G_0286CC_PERSP_SAMPLE_ENA(config->spi_ps_input_addr))
shader_info->num_input_vgprs += 2;
/* +3 for scratch wave offset and VCC */
config->num_sgprs = MAX2(config->num_sgprs,
shader_info->num_input_sgprs + 3);
- if (nir->stage == MESA_SHADER_COMPUTE) {
+}
+
+void ac_compile_nir_shader(LLVMTargetMachineRef tm,
+ struct ac_shader_binary *binary,
+ struct ac_shader_config *config,
+ struct ac_shader_variant_info *shader_info,
+ struct nir_shader *nir,
+ const struct ac_nir_compiler_options *options,
+ bool dump_shader)
+{
+
+ LLVMModuleRef llvm_module = ac_translate_nir_to_llvm(tm, nir, shader_info,
+ options);
+
+ ac_compile_llvm_module(tm, llvm_module, binary, config, shader_info, nir->stage, dump_shader);
+ switch (nir->stage) {
+ case MESA_SHADER_COMPUTE:
for (int i = 0; i < 3; ++i)
shader_info->cs.block_size[i] = nir->info->cs.local_size[i];
- }
-
- if (nir->stage == MESA_SHADER_FRAGMENT)
+ break;
+ case MESA_SHADER_FRAGMENT:
shader_info->fs.early_fragment_test = nir->info->fs.early_fragment_tests;
+ break;
+ default:
+ break;
+ }
}
projects / mesa.git / blobdiff