[Mesa-dev] [PATCH 5/6] r600g: Add TGSI->LLVM implementation
Tom Stellard
tstellar at gmail.com
Tue Apr 17 09:11:05 PDT 2012
---
src/gallium/drivers/r600/r600_llvm.c | 300 ++++++++++++++++++++++++++++++++++
src/gallium/drivers/r600/r600_llvm.h | 29 ++++
2 files changed, 329 insertions(+), 0 deletions(-)
create mode 100644 src/gallium/drivers/r600/r600_llvm.c
create mode 100644 src/gallium/drivers/r600/r600_llvm.h
diff --git a/src/gallium/drivers/r600/r600_llvm.c b/src/gallium/drivers/r600/r600_llvm.c
new file mode 100644
index 0000000..267fb19
--- /dev/null
+++ b/src/gallium/drivers/r600/r600_llvm.c
@@ -0,0 +1,300 @@
+#include "r600_llvm.h"
+
+#include "gallivm/lp_bld_const.h"
+#include "gallivm/lp_bld_intr.h"
+#include "gallivm/lp_bld_gather.h"
+#include "tgsi/tgsi_parse.h"
+#include "util/u_double_list.h"
+
+#include "r600.h"
+#include "r600_asm.h"
+#include "r600_opcodes.h"
+#include "r600_shader.h"
+#include "radeon_llvm.h"
+#include "radeon_llvm_emit.h"
+
+#include <stdio.h>
+
+static LLVMValueRef llvm_fetch_const(
+ struct lp_build_tgsi_context * bld_base,
+ const struct tgsi_full_src_register *reg,
+ enum tgsi_opcode_type type,
+ unsigned swizzle)
+{
+ return lp_build_intrinsic_unary(bld_base->base.gallivm->builder,
+ "llvm.AMDGPU.load.const", bld_base->base.elem_type,
+ lp_build_const_int32(bld_base->base.gallivm,
+ radeon_llvm_reg_index_soa(reg->Register.Index, swizzle)));
+}
+
+static void llvm_load_input(
+ struct radeon_llvm_context * ctx,
+ unsigned input_index,
+ const struct tgsi_full_declaration *decl)
+{
+ unsigned chan;
+
+ for (chan = 0; chan < 4; chan++) {
+ unsigned soa_index = radeon_llvm_reg_index_soa(input_index,
+ chan);
+
+ /* The * 4 is assuming that we are in soa mode. */
+ LLVMValueRef reg = lp_build_const_int32(
+ ctx->soa.bld_base.base.gallivm,
+ soa_index + (ctx->reserved_reg_count * 4));
+ ctx->inputs[soa_index] = lp_build_intrinsic_unary(
+ ctx->soa.bld_base.base.gallivm->builder,
+ "llvm.R600.load.input",
+ ctx->soa.bld_base.base.elem_type, reg);
+ }
+}
+
+static void llvm_emit_prologue(struct lp_build_tgsi_context * bld_base)
+{
+ struct radeon_llvm_context * ctx = radeon_llvm_context(bld_base);
+ struct lp_build_context * base = &bld_base->base;
+ unsigned i;
+
+ /* Reserve special input registers */
+ for (i = 0; i < ctx->reserved_reg_count; i++) {
+ unsigned chan;
+ for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
+ LLVMValueRef reg;
+ LLVMValueRef reg_index = lp_build_const_int32(
+ base->gallivm,
+ radeon_llvm_reg_index_soa(i, chan));
+ reg = lp_build_intrinsic_unary(base->gallivm->builder,
+ "llvm.AMDGPU.reserve.reg",
+ base->elem_type, reg_index);
+ lp_build_intrinsic_unary(base->gallivm->builder,
+ "llvm.AMDGPU.export.reg",
+ LLVMVoidTypeInContext(base->gallivm->context),
+ reg);
+ }
+ }
+}
+
+static void llvm_emit_epilogue(struct lp_build_tgsi_context * bld_base)
+{
+ struct radeon_llvm_context * ctx = radeon_llvm_context(bld_base);
+ struct lp_build_context * base = &bld_base->base;
+ unsigned i;
+
+ /* Add the necessary export instructions */
+ for (i = 0; i < ctx->output_reg_count; i++) {
+ unsigned chan;
+ for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
+ LLVMValueRef output;
+ LLVMValueRef store_output;
+ unsigned adjusted_reg_idx = i +
+ ctx->reserved_reg_count;
+ LLVMValueRef reg_index = lp_build_const_int32(
+ base->gallivm,
+ radeon_llvm_reg_index_soa(adjusted_reg_idx, chan));
+
+ output = LLVMBuildLoad(base->gallivm->builder,
+ ctx->soa.outputs[i][chan], "");
+
+ store_output = lp_build_intrinsic_binary(
+ base->gallivm->builder,
+ "llvm.AMDGPU.store.output",
+ base->elem_type,
+ output, reg_index);
+
+ lp_build_intrinsic_unary(base->gallivm->builder,
+ "llvm.AMDGPU.export.reg",
+ LLVMVoidTypeInContext(base->gallivm->context),
+ store_output);
+ }
+ }
+}
+
+static void llvm_emit_tex(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct gallivm_state * gallivm = bld_base->base.gallivm;
+ LLVMValueRef args[3];
+
+ args[0] = emit_data->args[0];
+ args[1] = lp_build_const_int32(gallivm,
+ emit_data->inst->Src[1].Register.Index);
+ args[2] = lp_build_const_int32(gallivm,
+ emit_data->inst->Texture.Texture);
+ emit_data->output[0] = lp_build_intrinsic(gallivm->builder,
+ action->intr_name,
+ emit_data->dst_type, args, 3);
+}
+
+static void dp_fetch_args(
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct lp_build_context * base = &bld_base->base;
+ unsigned chan;
+ LLVMValueRef elements[2][4];
+ unsigned opcode = emit_data->inst->Instruction.Opcode;
+ unsigned dp_components = (opcode == TGSI_OPCODE_DP2 ? 2 :
+ (opcode == TGSI_OPCODE_DP3 ? 3 : 4));
+ for (chan = 0 ; chan < dp_components; chan++) {
+ elements[0][chan] = lp_build_emit_fetch(bld_base,
+ emit_data->inst, 0, chan);
+ elements[1][chan] = lp_build_emit_fetch(bld_base,
+ emit_data->inst, 1, chan);
+ }
+
+ for ( ; chan < 4; chan++) {
+ elements[0][chan] = base->zero;
+ elements[1][chan] = base->zero;
+ }
+
+ /* Fix up for DPH */
+ if (opcode == TGSI_OPCODE_DPH) {
+ elements[0][TGSI_CHAN_W] = base->one;
+ }
+
+ emit_data->args[0] = lp_build_gather_values(bld_base->base.gallivm,
+ elements[0], 4);
+ emit_data->args[1] = lp_build_gather_values(bld_base->base.gallivm,
+ elements[1], 4);
+ emit_data->arg_count = 2;
+
+ emit_data->dst_type = base->elem_type;
+}
+
+static struct lp_build_tgsi_action dot_action = {
+ .fetch_args = dp_fetch_args,
+ .emit = lp_build_tgsi_intrinsic,
+ .intr_name = "llvm.AMDGPU.dp4"
+};
+
+static void txp_fetch_args(
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ LLVMValueRef src_w;
+ unsigned chan;
+ LLVMValueRef coords[4];
+
+ emit_data->dst_type = LLVMVectorType(bld_base->base.elem_type, 4);
+ src_w = lp_build_emit_fetch(bld_base, emit_data->inst, 0, TGSI_CHAN_W);
+
+ for (chan = 0; chan < 3; chan++ ) {
+ LLVMValueRef arg = lp_build_emit_fetch(bld_base,
+ emit_data->inst, 0, chan);
+ coords[chan] = lp_build_emit_llvm_binary(bld_base,
+ TGSI_OPCODE_DIV, arg, src_w);
+ }
+ coords[3] = bld_base->base.one;
+ emit_data->args[0] = lp_build_gather_values(bld_base->base.gallivm,
+ coords, 4);
+ emit_data->arg_count = 1;
+}
+
+LLVMModuleRef r600_tgsi_llvm(
+ struct radeon_llvm_context * ctx,
+ const struct tgsi_token * tokens)
+{
+ struct tgsi_shader_info shader_info;
+ struct lp_build_tgsi_context * bld_base = &ctx->soa.bld_base;
+ radeon_llvm_context_init(ctx);
+ tgsi_scan_shader(tokens, &shader_info);
+
+ bld_base->info = &shader_info;
+ bld_base->userdata = ctx;
+ bld_base->emit_fetch_funcs[TGSI_FILE_CONSTANT] = llvm_fetch_const;
+ bld_base->emit_prologue = llvm_emit_prologue;
+ bld_base->emit_epilogue = llvm_emit_epilogue;
+ ctx->userdata = ctx;
+ ctx->load_input = llvm_load_input;
+
+ bld_base->op_actions[TGSI_OPCODE_DP2] = dot_action;
+ bld_base->op_actions[TGSI_OPCODE_DP3] = dot_action;
+ bld_base->op_actions[TGSI_OPCODE_DP4] = dot_action;
+ bld_base->op_actions[TGSI_OPCODE_DPH] = dot_action;
+ bld_base->op_actions[TGSI_OPCODE_TEX].emit = llvm_emit_tex;
+ bld_base->op_actions[TGSI_OPCODE_TXB].emit = llvm_emit_tex;
+ bld_base->op_actions[TGSI_OPCODE_TXD].emit = llvm_emit_tex;
+ bld_base->op_actions[TGSI_OPCODE_TXL].emit = llvm_emit_tex;
+ bld_base->op_actions[TGSI_OPCODE_TXP].fetch_args = txp_fetch_args;
+ bld_base->op_actions[TGSI_OPCODE_TXP].emit = llvm_emit_tex;
+
+ lp_build_tgsi_llvm(bld_base, tokens);
+
+ radeon_llvm_finalize_module(ctx);
+
+ return ctx->gallivm.module;
+}
+
+const char * r600_llvm_gpu_string(enum radeon_family family)
+{
+ const char * gpu_family;
+
+ switch (family) {
+ case CHIP_R600:
+ case CHIP_RV610:
+ case CHIP_RV630:
+ case CHIP_RV620:
+ case CHIP_RV635:
+ case CHIP_RS780:
+ case CHIP_RS880:
+ case CHIP_RV710:
+ gpu_family = "rv710";
+ break;
+ case CHIP_RV730:
+ gpu_family = "rv730";
+ break;
+ case CHIP_RV670:
+ case CHIP_RV740:
+ case CHIP_RV770:
+ gpu_family = "rv770";
+ break;
+ case CHIP_PALM:
+ case CHIP_CEDAR:
+ gpu_family = "cedar";
+ break;
+ case CHIP_SUMO:
+ case CHIP_SUMO2:
+ case CHIP_REDWOOD:
+ gpu_family = "redwood";
+ break;
+ case CHIP_JUNIPER:
+ gpu_family = "juniper";
+ break;
+ case CHIP_HEMLOCK:
+ case CHIP_CYPRESS:
+ gpu_family = "cypress";
+ break;
+ case CHIP_BARTS:
+ gpu_family = "barts";
+ break;
+ case CHIP_TURKS:
+ gpu_family = "turks";
+ break;
+ case CHIP_CAICOS:
+ gpu_family = "caicos";
+ break;
+ case CHIP_CAYMAN:
+ gpu_family = "cayman";
+ break;
+ default:
+ gpu_family = "";
+ fprintf(stderr, "Chip not supported by r600 llvm "
+ "backend, please file a bug at bugs.freedesktop.org\n");
+ break;
+ }
+ return gpu_family;
+}
+
+unsigned r600_llvm_compile(
+ LLVMModuleRef mod,
+ unsigned char ** inst_bytes,
+ unsigned * inst_byte_count,
+ enum radeon_family family,
+ unsigned dump)
+{
+ const char * gpu_family = r600_llvm_gpu_string(family);
+ return radeon_llvm_compile(mod, inst_bytes, inst_byte_count,
+ gpu_family, dump);
+}
diff --git a/src/gallium/drivers/r600/r600_llvm.h b/src/gallium/drivers/r600/r600_llvm.h
new file mode 100644
index 0000000..0f6a1f8
--- /dev/null
+++ b/src/gallium/drivers/r600/r600_llvm.h
@@ -0,0 +1,29 @@
+
+#ifndef R600_LLVM_H
+#define R600_LLVM_H
+
+#ifdef R600_USE_LLVM
+
+#include "radeon_llvm.h"
+#include <llvm-c/Core.h>
+
+struct r600_shader_ctx;
+struct radeon_llvm_context;
+enum radeon_family;
+
+LLVMModuleRef r600_tgsi_llvm(
+ struct radeon_llvm_context * ctx,
+ const struct tgsi_token * tokens);
+
+const char * r600_llvm_gpu_string(enum radeon_family family);
+
+unsigned r600_llvm_compile(
+ LLVMModuleRef mod,
+ unsigned char ** inst_bytes,
+ unsigned * inst_byte_count,
+ enum radeon_family family,
+ unsigned dump);
+
+#endif /* R600_USE_LLVM */
+
+#endif /* R600_LLVM_H */
--
1.7.7.6
More information about the mesa-dev
mailing list