int param_vertex_id;
int param_instance_id;
LLVMValueRef const_md;
- LLVMValueRef const_resource;
+ LLVMValueRef const_resource[NUM_CONST_BUFFERS];
#if HAVE_LLVM >= 0x0304
LLVMValueRef ddxy_lds;
#endif
- LLVMValueRef *constants;
+ LLVMValueRef *constants[NUM_CONST_BUFFERS];
LLVMValueRef *resources;
LLVMValueRef *samplers;
LLVMValueRef so_buffers[4];
{
struct lp_build_context * base = &si_shader_ctx->radeon_bld.soa.bld_base.base;
+ LLVMValueRef indices[2] = {
+ LLVMConstInt(LLVMInt64TypeInContext(base->gallivm->context), 0, false),
+ offset
+ };
LLVMValueRef computed_ptr = LLVMBuildGEP(
- base->gallivm->builder, base_ptr, &offset, 1, "");
+ base->gallivm->builder, base_ptr, indices, 2, "");
LLVMValueRef result = LLVMBuildLoad(base->gallivm->builder, computed_ptr, "");
LLVMSetMetadata(result, 1, si_shader_ctx->const_md);
struct si_shader_context *si_shader_ctx = si_shader_context(bld_base);
struct lp_build_context * base = &bld_base->base;
const struct tgsi_ind_register *ireg = ®->Indirect;
- unsigned idx;
+ unsigned buf, idx;
LLVMValueRef args[2];
LLVMValueRef addr;
return lp_build_gather_values(bld_base->base.gallivm, values, 4);
}
+ buf = reg->Register.Dimension ? reg->Dimension.Index : 0;
idx = reg->Register.Index * 4 + swizzle;
+
if (!reg->Register.Indirect)
- return bitcast(bld_base, type, si_shader_ctx->constants[idx]);
+ return bitcast(bld_base, type, si_shader_ctx->constants[buf][idx]);
- args[0] = si_shader_ctx->const_resource;
+ args[0] = si_shader_ctx->const_resource[buf];
args[1] = lp_build_const_int32(base->gallivm, idx * 4);
addr = si_shader_ctx->radeon_bld.soa.addr[ireg->Index][ireg->Swizzle];
addr = LLVMBuildLoad(base->gallivm->builder, addr, "load addr reg");
if (si_shader_ctx->shader->key.ps.alpha_func != PIPE_FUNC_NEVER) {
LLVMValueRef out_ptr = si_shader_ctx->radeon_bld.soa.outputs[index][3];
+ LLVMValueRef alpha_ref = LLVMGetParam(si_shader_ctx->radeon_bld.main_fn,
+ SI_PARAM_ALPHA_REF);
+
LLVMValueRef alpha_pass =
lp_build_cmp(&bld_base->base,
si_shader_ctx->shader->key.ps.alpha_func,
LLVMBuildLoad(gallivm->builder, out_ptr, ""),
- lp_build_const_float(gallivm, si_shader_ctx->shader->key.ps.alpha_ref));
+ alpha_ref);
LLVMValueRef arg =
lp_build_select(&bld_base->base,
alpha_pass,
LLVMValueRef out_elts[4];
LLVMValueRef base_elt;
LLVMValueRef ptr = LLVMGetParam(si_shader_ctx->radeon_bld.main_fn, SI_PARAM_CONST);
- LLVMValueRef const_resource = build_indexed_load(si_shader_ctx, ptr, uint->one);
+ LLVMValueRef constbuf_index = lp_build_const_int32(base->gallivm, NUM_PIPE_CONST_BUFFERS);
+ LLVMValueRef const_resource = build_indexed_load(si_shader_ctx, ptr, constbuf_index);
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
LLVMValueRef out_ptr = si_shader_ctx->radeon_bld.soa.outputs[index][chan];
LLVMValueRef last_args[9] = { 0 };
LLVMValueRef pos_args[4][9] = { { 0 } };
unsigned semantic_name;
- unsigned color_count = 0;
unsigned param_count = 0;
- int depth_index = -1, stencil_index = -1;
+ int depth_index = -1, stencil_index = -1, psize_index = -1, edgeflag_index = -1;
int i;
if (si_shader_ctx->shader->selector->so.num_outputs) {
/* Select the correct target */
switch(semantic_name) {
case TGSI_SEMANTIC_PSIZE:
- shader->vs_out_misc_write = 1;
- shader->vs_out_point_size = 1;
- target = V_008DFC_SQ_EXP_POS + 1;
- break;
+ shader->vs_out_misc_write = true;
+ shader->vs_out_point_size = true;
+ psize_index = index;
+ continue;
+ case TGSI_SEMANTIC_EDGEFLAG:
+ shader->vs_out_misc_write = true;
+ shader->vs_out_edgeflag = true;
+ edgeflag_index = index;
+ continue;
case TGSI_SEMANTIC_POSITION:
if (si_shader_ctx->type == TGSI_PROCESSOR_VERTEX) {
target = V_008DFC_SQ_EXP_POS;
shader->output[i].param_offset = param_count;
param_count++;
} else {
- target = V_008DFC_SQ_EXP_MRT + color_count;
+ target = V_008DFC_SQ_EXP_MRT + shader->output[i].sid;
if (si_shader_ctx->shader->key.ps.alpha_to_one) {
si_alpha_to_one(bld_base, index);
}
- if (color_count == 0 &&
+ if (shader->output[i].sid == 0 &&
si_shader_ctx->shader->key.ps.alpha_func != PIPE_FUNC_ALWAYS)
si_alpha_test(bld_base, index);
-
- color_count++;
}
break;
case TGSI_SEMANTIC_CLIPDIST:
LLVMVoidTypeInContext(base->gallivm->context),
args, 9);
}
-
}
if (semantic_name == TGSI_SEMANTIC_CLIPDIST) {
if (si_shader_ctx->type == TGSI_PROCESSOR_VERTEX) {
unsigned pos_idx = 0;
+ /* We need to add the position output manually if it's missing. */
+ if (!pos_args[0][0]) {
+ pos_args[0][0] = lp_build_const_int32(base->gallivm, 0xf); /* writemask */
+ pos_args[0][1] = uint->zero; /* EXEC mask */
+ pos_args[0][2] = uint->zero; /* last export? */
+ pos_args[0][3] = lp_build_const_int32(base->gallivm, V_008DFC_SQ_EXP_POS);
+ pos_args[0][4] = uint->zero; /* COMPR flag */
+ pos_args[0][5] = base->zero; /* X */
+ pos_args[0][6] = base->zero; /* Y */
+ pos_args[0][7] = base->zero; /* Z */
+ pos_args[0][8] = base->one; /* W */
+ }
+
+ /* Write the misc vector (point size, edgeflag, layer, viewport). */
+ if (shader->vs_out_misc_write) {
+ pos_args[1][0] = lp_build_const_int32(base->gallivm, /* writemask */
+ shader->vs_out_point_size |
+ (shader->vs_out_edgeflag << 1));
+ pos_args[1][1] = uint->zero; /* EXEC mask */
+ pos_args[1][2] = uint->zero; /* last export? */
+ pos_args[1][3] = lp_build_const_int32(base->gallivm, V_008DFC_SQ_EXP_POS + 1);
+ pos_args[1][4] = uint->zero; /* COMPR flag */
+ pos_args[1][5] = base->zero; /* X */
+ pos_args[1][6] = base->zero; /* Y */
+ pos_args[1][7] = base->zero; /* Z */
+ pos_args[1][8] = base->zero; /* W */
+
+ if (shader->vs_out_point_size) {
+ pos_args[1][5] = LLVMBuildLoad(base->gallivm->builder,
+ si_shader_ctx->radeon_bld.soa.outputs[psize_index][0], "");
+ }
+
+ if (shader->vs_out_edgeflag) {
+ LLVMValueRef output = LLVMBuildLoad(base->gallivm->builder,
+ si_shader_ctx->radeon_bld.soa.outputs[edgeflag_index][0], "");
+
+ /* The output is a float, but the hw expects an integer
+ * with the first bit containing the edge flag. */
+ output = LLVMBuildFPToUI(base->gallivm->builder, output,
+ bld_base->uint_bld.elem_type, "");
+
+ output = lp_build_min(&bld_base->int_bld, output, bld_base->int_bld.one);
+
+ /* The LLVM intrinsic expects a float. */
+ pos_args[1][6] = LLVMBuildBitCast(base->gallivm->builder, output,
+ base->elem_type, "");
+ }
+ }
+
for (i = 0; i < 4; i++)
if (pos_args[i][0])
shader->nr_pos_exports++;
const struct tgsi_full_instruction * inst = emit_data->inst;
unsigned opcode = inst->Instruction.Opcode;
unsigned target = inst->Texture.Texture;
- unsigned sampler_src, sampler_index;
LLVMValueRef coords[4];
LLVMValueRef address[16];
int ref_pos;
unsigned num_coords = tgsi_util_get_texture_coord_dim(target, &ref_pos);
unsigned count = 0;
unsigned chan;
+ unsigned sampler_src = emit_data->inst->Instruction.NumSrcRegs - 1;
+ unsigned sampler_index = emit_data->inst->Src[sampler_src].Register.Index;
+
+ if (target == TGSI_TEXTURE_BUFFER) {
+ LLVMTypeRef i128 = LLVMIntTypeInContext(gallivm->context, 128);
+ LLVMTypeRef v2i128 = LLVMVectorType(i128, 2);
+ LLVMTypeRef i8 = LLVMInt8TypeInContext(gallivm->context);
+ LLVMTypeRef v16i8 = LLVMVectorType(i8, 16);
+
+ /* Truncate v32i8 to v16i8. */
+ LLVMValueRef res = si_shader_ctx->resources[sampler_index];
+ res = LLVMBuildBitCast(gallivm->builder, res, v2i128, "");
+ res = LLVMBuildExtractElement(gallivm->builder, res, bld_base->uint_bld.zero, "");
+ res = LLVMBuildBitCast(gallivm->builder, res, v16i8, "");
+
+ emit_data->dst_type = LLVMVectorType(bld_base->base.elem_type, 4);
+ emit_data->args[0] = res;
+ emit_data->args[1] = bld_base->uint_bld.zero;
+ emit_data->args[2] = lp_build_emit_fetch(bld_base, emit_data->inst, 0, 0);
+ emit_data->arg_count = 3;
+ return;
+ }
/* Fetch and project texture coordinates */
coords[3] = lp_build_emit_fetch(bld_base, emit_data->inst, 0, TGSI_CHAN_W);
"");
}
- sampler_src = emit_data->inst->Instruction.NumSrcRegs - 1;
- sampler_index = emit_data->inst->Src[sampler_src].Register.Index;
-
/* Adjust the sample index according to FMASK.
*
* For uncompressed MSAA surfaces, FMASK should return 0x76543210,
struct lp_build_emit_data * emit_data)
{
struct lp_build_context * base = &bld_base->base;
- char intr_name[23];
+ char intr_name[127];
+
+ if (emit_data->inst->Texture.Texture == TGSI_TEXTURE_BUFFER) {
+ emit_data->output[emit_data->chan] = build_intrinsic(
+ base->gallivm->builder,
+ "llvm.SI.vs.load.input", emit_data->dst_type,
+ emit_data->args, emit_data->arg_count,
+ LLVMReadNoneAttribute | LLVMNoUnwindAttribute);
+ return;
+ }
sprintf(intr_name, "%sv%ui32", action->intr_name,
LLVMGetVectorSize(LLVMTypeOf(emit_data->args[0])));
{
struct si_shader_context *si_shader_ctx = si_shader_context(bld_base);
const struct tgsi_full_instruction *inst = emit_data->inst;
+ struct gallivm_state *gallivm = bld_base->base.gallivm;
+
+ if (inst->Texture.Texture == TGSI_TEXTURE_BUFFER) {
+ LLVMTypeRef i32 = LLVMInt32TypeInContext(gallivm->context);
+ LLVMTypeRef v8i32 = LLVMVectorType(i32, 8);
+
+ /* Read the size from the buffer descriptor directly. */
+ LLVMValueRef size = si_shader_ctx->resources[inst->Src[1].Register.Index];
+ size = LLVMBuildBitCast(gallivm->builder, size, v8i32, "");
+ size = LLVMBuildExtractElement(gallivm->builder, size,
+ lp_build_const_int32(gallivm, 2), "");
+ emit_data->args[0] = size;
+ return;
+ }
/* Mip level */
emit_data->args[0] = lp_build_emit_fetch(bld_base, inst, 0, TGSI_CHAN_X);
4);
}
+static void build_txq_intrinsic(const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ if (emit_data->inst->Texture.Texture == TGSI_TEXTURE_BUFFER) {
+ /* Just return the buffer size. */
+ emit_data->output[emit_data->chan] = emit_data->args[0];
+ return;
+ }
+
+ build_tgsi_intrinsic_nomem(action, bld_base, emit_data);
+}
+
#if HAVE_LLVM >= 0x0304
static void si_llvm_emit_ddxy(
static const struct lp_build_tgsi_action txq_action = {
.fetch_args = txq_fetch_args,
- .emit = build_tgsi_intrinsic_nomem,
+ .emit = build_txq_intrinsic,
.intr_name = "llvm.SI.resinfo"
};
{
struct lp_build_tgsi_context *bld_base = &si_shader_ctx->radeon_bld.soa.bld_base;
struct gallivm_state *gallivm = bld_base->base.gallivm;
- LLVMTypeRef params[20], f32, i8, i32, v2i32, v3i32;
+ LLVMTypeRef params[21], f32, i8, i32, v2i32, v3i32;
unsigned i, last_sgpr, num_params;
i8 = LLVMInt8TypeInContext(gallivm->context);
v2i32 = LLVMVectorType(i32, 2);
v3i32 = LLVMVectorType(i32, 3);
- params[SI_PARAM_CONST] = LLVMPointerType(LLVMVectorType(i8, 16), CONST_ADDR_SPACE);
- params[SI_PARAM_SAMPLER] = params[SI_PARAM_CONST];
- params[SI_PARAM_RESOURCE] = LLVMPointerType(LLVMVectorType(i8, 32), CONST_ADDR_SPACE);
+ params[SI_PARAM_CONST] = LLVMPointerType(
+ LLVMArrayType(LLVMVectorType(i8, 16), NUM_CONST_BUFFERS), CONST_ADDR_SPACE);
+ /* We assume at most 16 textures per program at the moment.
+ * This need probably need to be changed to support bindless textures */
+ params[SI_PARAM_SAMPLER] = LLVMPointerType(
+ LLVMArrayType(LLVMVectorType(i8, 16), NUM_SAMPLER_VIEWS), CONST_ADDR_SPACE);
+ params[SI_PARAM_RESOURCE] = LLVMPointerType(
+ LLVMArrayType(LLVMVectorType(i8, 32), NUM_SAMPLER_STATES), CONST_ADDR_SPACE);
switch (si_shader_ctx->type) {
case TGSI_PROCESSOR_VERTEX:
break;
case TGSI_PROCESSOR_FRAGMENT:
+ params[SI_PARAM_ALPHA_REF] = f32;
params[SI_PARAM_PRIM_MASK] = i32;
last_sgpr = SI_PARAM_PRIM_MASK;
params[SI_PARAM_PERSP_SAMPLE] = v2i32;
for (i = 0; i <= last_sgpr; ++i) {
LLVMValueRef P = LLVMGetParam(si_shader_ctx->radeon_bld.main_fn, i);
- LLVMAddAttribute(P, LLVMInRegAttribute);
+ switch (i) {
+ default:
+ LLVMAddAttribute(P, LLVMInRegAttribute);
+ break;
+#if HAVE_LLVM >= 0x0304
+ /* We tell llvm that array inputs are passed by value to allow Sinking pass
+ * to move load. Inputs are constant so this is fine. */
+ case SI_PARAM_CONST:
+ case SI_PARAM_SAMPLER:
+ case SI_PARAM_RESOURCE:
+ LLVMAddAttribute(P, LLVMByValAttribute);
+ break;
+#endif
+ }
}
#if HAVE_LLVM >= 0x0304
struct lp_build_tgsi_context * bld_base = &si_shader_ctx->radeon_bld.soa.bld_base;
struct gallivm_state * gallivm = bld_base->base.gallivm;
const struct tgsi_shader_info * info = bld_base->info;
+ unsigned buf;
+ LLVMValueRef ptr = LLVMGetParam(si_shader_ctx->radeon_bld.main_fn, SI_PARAM_CONST);
- unsigned i, num_const = info->file_max[TGSI_FILE_CONSTANT] + 1;
-
- LLVMValueRef ptr;
+ for (buf = 0; buf < NUM_CONST_BUFFERS; buf++) {
+ unsigned i, num_const = info->const_file_max[buf] + 1;
- if (num_const == 0)
- return;
+ if (num_const == 0)
+ continue;
- /* Allocate space for the constant values */
- si_shader_ctx->constants = CALLOC(num_const * 4, sizeof(LLVMValueRef));
-
- /* Load the resource descriptor */
- ptr = LLVMGetParam(si_shader_ctx->radeon_bld.main_fn, SI_PARAM_CONST);
- si_shader_ctx->const_resource = build_indexed_load(si_shader_ctx, ptr, bld_base->uint_bld.zero);
-
- /* Load the constants, we rely on the code sinking to do the rest */
- for (i = 0; i < num_const * 4; ++i) {
- LLVMValueRef args[2] = {
- si_shader_ctx->const_resource,
- lp_build_const_int32(gallivm, i * 4)
- };
- si_shader_ctx->constants[i] = build_intrinsic(gallivm->builder, "llvm.SI.load.const",
- bld_base->base.elem_type, args, 2, LLVMReadNoneAttribute | LLVMNoUnwindAttribute);
+ /* Allocate space for the constant values */
+ si_shader_ctx->constants[buf] = CALLOC(num_const * 4, sizeof(LLVMValueRef));
+
+ /* Load the resource descriptor */
+ si_shader_ctx->const_resource[buf] =
+ build_indexed_load(si_shader_ctx, ptr, lp_build_const_int32(gallivm, buf));
+
+ /* Load the constants, we rely on the code sinking to do the rest */
+ for (i = 0; i < num_const * 4; ++i) {
+ LLVMValueRef args[2] = {
+ si_shader_ctx->const_resource[buf],
+ lp_build_const_int32(gallivm, i * 4)
+ };
+ si_shader_ctx->constants[buf][i] =
+ build_intrinsic(gallivm->builder, "llvm.SI.load.const",
+ bld_base->base.elem_type, args, 2,
+ LLVMReadNoneAttribute | LLVMNoUnwindAttribute);
+ }
}
}
if (!lp_build_tgsi_llvm(bld_base, sel->tokens)) {
fprintf(stderr, "Failed to translate shader from TGSI to LLVM\n");
- FREE(si_shader_ctx.constants);
+ for (int i = 0; i < NUM_CONST_BUFFERS; i++)
+ FREE(si_shader_ctx.constants[i]);
FREE(si_shader_ctx.resources);
FREE(si_shader_ctx.samplers);
return -EINVAL;
radeon_llvm_dispose(&si_shader_ctx.radeon_bld);
tgsi_parse_free(&si_shader_ctx.parse);
- FREE(si_shader_ctx.constants);
+ for (int i = 0; i < NUM_CONST_BUFFERS; i++)
+ FREE(si_shader_ctx.constants[i]);
FREE(si_shader_ctx.resources);
FREE(si_shader_ctx.samplers);