[Mesa-dev] [PATCH 6/6] r600g: Add hooks for the LLVM shader compiler
Tom Stellard
tstellar at gmail.com
Tue Apr 17 09:11:06 PDT 2012
The LLVM backend can now be enabled for r600g by using the
--enable-r600-llvm-compiler configure flag. If you configure with this
flag, you can still use the default compiler by setting the envrionment
variable R600_USE_LLVM=0
---
configure.ac | 14 ++
src/gallium/drivers/r600/Makefile.am | 24 +++
src/gallium/drivers/r600/Makefile.sources | 2 +
src/gallium/drivers/r600/r600_shader.c | 280 ++++++++++++++++++++++++++++-
4 files changed, 318 insertions(+), 2 deletions(-)
diff --git a/configure.ac b/configure.ac
index c5514b9..a70d1b1 100644
--- a/configure.ac
+++ b/configure.ac
@@ -637,6 +637,12 @@ AC_ARG_ENABLE([gallium_gbm],
[enable_gallium_gbm="$enableval"],
[enable_gallium_gbm=auto])
+AC_ARG_ENABLE([r600-llvm-compiler],
+ [AS_HELP_STRING([--enable-r600-llvm-compilerl],
+ [Enable experimental LLVM backend for graphics shaders @<:default=disable@:>@])],
+ [enable_r600_llvm="$enableval"],
+ [enable_r600_llvm=no])
+
# Option for Gallium drivers
GALLIUM_DRIVERS_DEFAULT="r300,r600,svga,swrast"
@@ -1906,6 +1912,13 @@ if test "x$with_gallium_drivers" != x; then
xr600)
PKG_CHECK_MODULES([RADEON], [libdrm_radeon >= $LIBDRM_RADEON_REQUIRED])
GALLIUM_DRIVERS_DIRS="$GALLIUM_DRIVERS_DIRS r600"
+ if test "x$enable_r600_llvm" = xyes; then
+ if test "x$LLVM_VERSION" != "x3.1"; then
+ AC_MSG_ERROR([LLVM 3.1 is required for the r600 llvm compiler.])
+ fi
+ NEED_RADEON_GALLIUM=yes;
+ USE_R600_LLVM_COMPILER=yes;
+ fi
gallium_check_st "radeon/drm" "dri-r600" "xorg-r600" "" "xvmc-r600" "vdpau-r600" "va-r600"
;;
xradeonsi)
@@ -1976,6 +1989,7 @@ AM_CONDITIONAL(HAVE_GALAHAD_GALLIUM, test x$HAVE_GALAHAD_GALLIUM = xyes)
AM_CONDITIONAL(HAVE_IDENTITY_GALLIUM, test x$HAVE_IDENTITY_GALLIUM = xyes)
AM_CONDITIONAL(HAVE_NOOP_GALLIUM, test x$HAVE_NOOP_GALLIUM = xyes)
AM_CONDITIONAL(NEED_RADEON_GALLIUM, test x$NEED_RADEON_GALLIUM = xyes)
+AM_CONDITIONAL(USE_R600_LLVM_COMPILER, test x$USE_R600_LLVM_COMPILER = xyes)
AC_SUBST([GALLIUM_MAKE_DIRS])
dnl prepend CORE_DIRS to SRC_DIRS
diff --git a/src/gallium/drivers/r600/Makefile.am b/src/gallium/drivers/r600/Makefile.am
index 8acd36a..a567f07 100644
--- a/src/gallium/drivers/r600/Makefile.am
+++ b/src/gallium/drivers/r600/Makefile.am
@@ -15,3 +15,27 @@ AM_CFLAGS = \
libr600_a_SOURCES = \
$(C_SOURCES)
+
+if USE_R600_LLVM_COMPILER
+
+# This is a hack until we can move the backend into the LLVM project.
+# We need to use mklib, because it splits up libradeon.a into object files
+# so that we can link it with the r600 objects.
+libr600_a_AR = $(top_srcdir)/bin/mklib -o r600 -static
+
+libr600_a_SOURCES += \
+ $(LLVM_C_SOURCES)
+
+libr600_a_LIBADD = \
+ $(top_srcdir)/src/gallium/drivers/radeon/libradeon.a
+
+AM_CFLAGS += \
+ $(LLVM_CFLAGS) \
+ -I$(top_srcdir)/src/gallium/drivers/radeon/ \
+ -DR600_USE_LLVM
+
+AM_CXXFLAGS= \
+ $(LLVM_CXXFLAGS)
+else
+libr600_a_AR = $(AR) $(ARFLAGS)
+endif
diff --git a/src/gallium/drivers/r600/Makefile.sources b/src/gallium/drivers/r600/Makefile.sources
index e7813ef..a920b93 100644
--- a/src/gallium/drivers/r600/Makefile.sources
+++ b/src/gallium/drivers/r600/Makefile.sources
@@ -15,3 +15,5 @@ C_SOURCES := \
eg_asm.c \
r600_translate.c \
r600_state_common.c
+
+LLVM_C_SOURCES := r600_llvm.c
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 4932dcc..562f27b 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -21,14 +21,18 @@
* USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
#include "r600_sq.h"
+#include "r600_llvm.h"
#include "r600_formats.h"
#include "r600_opcodes.h"
#include "r600d.h"
+#include "pipe/p_shader_tokens.h"
#include "tgsi/tgsi_info.h"
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_scan.h"
#include "tgsi/tgsi_dump.h"
+#include "util/u_memory.h"
+#include <stdio.h>
#include <errno.h>
#include <byteswap.h>
@@ -205,6 +209,224 @@ struct r600_shader_tgsi_instruction {
static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[], eg_shader_tgsi_instruction[], cm_shader_tgsi_instruction[];
static int tgsi_helper_tempx_replicate(struct r600_shader_ctx *ctx);
+static inline void callstack_check_depth(struct r600_shader_ctx *ctx, unsigned reason, unsigned check_max_only);
+static void fc_pushlevel(struct r600_shader_ctx *ctx, int type);
+static int tgsi_else(struct r600_shader_ctx *ctx);
+static int tgsi_endif(struct r600_shader_ctx *ctx);
+static int tgsi_bgnloop(struct r600_shader_ctx *ctx);
+static int tgsi_endloop(struct r600_shader_ctx *ctx);
+static int tgsi_loop_brk_cont(struct r600_shader_ctx *ctx);
+
+/*
+ * bytestream -> r600 shader
+ *
+ * These functions are used to transform the output of the LLVM backend into
+ * struct r600_bytecode.
+ */
+
+static unsigned r600_src_from_byte_stream(unsigned char * bytes,
+ unsigned bytes_read, struct r600_bytecode_alu * alu, unsigned src_idx)
+{
+ unsigned i;
+ unsigned sel0, sel1;
+ sel0 = bytes[bytes_read++];
+ sel1 = bytes[bytes_read++];
+ alu->src[src_idx].sel = sel0 | (sel1 << 8);
+ alu->src[src_idx].chan = bytes[bytes_read++];
+ alu->src[src_idx].neg = bytes[bytes_read++];
+ alu->src[src_idx].abs = bytes[bytes_read++];
+ alu->src[src_idx].rel = bytes[bytes_read++];
+ alu->src[src_idx].kc_bank = bytes[bytes_read++];
+ for (i = 0; i < 4; i++) {
+ alu->src[src_idx].value |= bytes[bytes_read++] << (i * 8);
+ }
+ return bytes_read;
+}
+
+static unsigned r600_alu_from_byte_stream(struct r600_shader_ctx *ctx,
+ unsigned char * bytes, unsigned bytes_read)
+{
+ unsigned src_idx;
+ unsigned inst0, inst1;
+ struct r600_bytecode_alu alu;
+ memset(&alu, 0, sizeof(alu));
+ for(src_idx = 0; src_idx < 3; src_idx++) {
+ bytes_read = r600_src_from_byte_stream(bytes, bytes_read,
+ &alu, src_idx);
+ }
+
+ alu.dst.sel = bytes[bytes_read++];
+ alu.dst.chan = bytes[bytes_read++];
+ alu.dst.clamp = bytes[bytes_read++];
+ alu.dst.write = bytes[bytes_read++];
+ alu.dst.rel = bytes[bytes_read++];
+ inst0 = bytes[bytes_read++];
+ inst1 = bytes[bytes_read++];
+ alu.inst = inst0 | (inst1 << 8);
+ alu.last = bytes[bytes_read++];
+ alu.is_op3 = bytes[bytes_read++];
+ alu.predicate = bytes[bytes_read++];
+ alu.bank_swizzle = bytes[bytes_read++];
+ alu.bank_swizzle_force = bytes[bytes_read++];
+ alu.omod = bytes[bytes_read++];
+ alu.index_mode = bytes[bytes_read++];
+ r600_bytecode_add_alu(ctx->bc, &alu);
+
+ /* XXX: Handle other KILL instructions */
+ if (alu.inst == CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLGT)) {
+ ctx->shader->uses_kill = 1;
+ /* XXX: This should be enforced in the LLVM backend. */
+ ctx->bc->force_add_cf = 1;
+ }
+ return bytes_read;
+}
+
+static void llvm_if(struct r600_shader_ctx *ctx, struct r600_bytecode_alu * alu,
+ unsigned pred_inst)
+{
+ alu->inst = pred_inst;
+ alu->predicate = 1;
+ alu->src[1].sel = V_SQ_ALU_SRC_0;
+ alu->src[1].chan = 0;
+ alu->last = 1;
+ r600_bytecode_add_alu_type(ctx->bc, alu,
+ CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE));
+
+ r600_bytecode_add_cfinst(ctx->bc, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_JUMP));
+ fc_pushlevel(ctx, FC_IF);
+ callstack_check_depth(ctx, FC_PUSH_VPM, 0);
+}
+
+static void r600_break_from_byte_stream(struct r600_shader_ctx *ctx,
+ struct r600_bytecode_alu *alu, unsigned compare_opcode)
+{
+ unsigned opcode = TGSI_OPCODE_BRK;
+ if (ctx->bc->chip_class == CAYMAN)
+ ctx->inst_info = &cm_shader_tgsi_instruction[opcode];
+ else if (ctx->bc->chip_class >= EVERGREEN)
+ ctx->inst_info = &eg_shader_tgsi_instruction[opcode];
+ else
+ ctx->inst_info = &r600_shader_tgsi_instruction[opcode];
+ llvm_if(ctx, alu, compare_opcode);
+ tgsi_loop_brk_cont(ctx);
+ tgsi_endif(ctx);
+}
+
+static unsigned r600_fc_from_byte_stream(struct r600_shader_ctx *ctx,
+ unsigned char * bytes, unsigned bytes_read)
+{
+ struct r600_bytecode_alu alu;
+ unsigned inst;
+ memset(&alu, 0, sizeof(alu));
+ bytes_read = r600_src_from_byte_stream(bytes, bytes_read, &alu, 0);
+ inst = bytes[bytes_read++];
+ switch (inst) {
+ case 0:
+ llvm_if(ctx, &alu,
+ CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETNE));
+ break;
+ case 1:
+ tgsi_else(ctx);
+ break;
+ case 2:
+ tgsi_endif(ctx);
+ break;
+ case 3:
+ tgsi_bgnloop(ctx);
+ break;
+ case 4:
+ tgsi_endloop(ctx);
+ break;
+ case 5:
+ r600_break_from_byte_stream(ctx, &alu,
+ CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETE));
+ break;
+ case 6:
+ r600_break_from_byte_stream(ctx, &alu,
+ CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETNE_INT));
+ break;
+ case 7:
+ {
+ unsigned opcode = TGSI_OPCODE_CONT;
+ if (ctx->bc->chip_class == CAYMAN) {
+ ctx->inst_info =
+ &cm_shader_tgsi_instruction[opcode];
+ } else if (ctx->bc->chip_class >= EVERGREEN) {
+ ctx->inst_info =
+ &eg_shader_tgsi_instruction[opcode];
+ } else {
+ ctx->inst_info =
+ &r600_shader_tgsi_instruction[opcode];
+ }
+ tgsi_loop_brk_cont(ctx);
+ }
+ break;
+ }
+
+ return bytes_read;
+}
+
+static unsigned r600_tex_from_byte_stream(struct r600_shader_ctx *ctx,
+ unsigned char * bytes, unsigned bytes_read)
+{
+ struct r600_bytecode_tex tex;
+
+ tex.inst = bytes[bytes_read++];
+ tex.resource_id = bytes[bytes_read++];
+ tex.src_gpr = bytes[bytes_read++];
+ tex.src_rel = bytes[bytes_read++];
+ tex.dst_gpr = bytes[bytes_read++];
+ tex.dst_rel = bytes[bytes_read++];
+ tex.dst_sel_x = bytes[bytes_read++];
+ tex.dst_sel_y = bytes[bytes_read++];
+ tex.dst_sel_z = bytes[bytes_read++];
+ tex.dst_sel_w = bytes[bytes_read++];
+ tex.lod_bias = bytes[bytes_read++];
+ tex.coord_type_x = bytes[bytes_read++];
+ tex.coord_type_y = bytes[bytes_read++];
+ tex.coord_type_z = bytes[bytes_read++];
+ tex.coord_type_w = bytes[bytes_read++];
+ tex.offset_x = bytes[bytes_read++];
+ tex.offset_y = bytes[bytes_read++];
+ tex.offset_z = bytes[bytes_read++];
+ tex.sampler_id = bytes[bytes_read++];
+ tex.src_sel_x = bytes[bytes_read++];
+ tex.src_sel_y = bytes[bytes_read++];
+ tex.src_sel_z = bytes[bytes_read++];
+ tex.src_sel_w = bytes[bytes_read++];
+
+ r600_bytecode_add_tex(ctx->bc, &tex);
+
+ return bytes_read;
+}
+
+static void r600_bytecode_from_byte_stream(struct r600_shader_ctx *ctx,
+ unsigned char * bytes, unsigned num_bytes)
+{
+ unsigned bytes_read = 0;
+ while (bytes_read < num_bytes) {
+ char inst_type = bytes[bytes_read++];
+ switch (inst_type) {
+ case 0:
+ bytes_read = r600_alu_from_byte_stream(ctx, bytes,
+ bytes_read);
+ break;
+ case 1:
+ bytes_read = r600_tex_from_byte_stream(ctx, bytes,
+ bytes_read);
+ break;
+ case 2:
+ bytes_read = r600_fc_from_byte_stream(ctx, bytes,
+ bytes_read);
+ break;
+ default:
+ /* XXX: Error here */
+ break;
+ }
+ }
+}
+
+/* End bytestream -> r600 shader functions*/
static int tgsi_is_supported(struct r600_shader_ctx *ctx)
{
@@ -815,7 +1037,14 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
unsigned opcode;
int i, j, k, r = 0;
int next_pixel_base = 0, next_pos_base = 60, next_param_base = 0;
+ /* Declarations used by llvm code */
+ bool use_llvm = false;
+ unsigned char * inst_bytes;
+ unsigned inst_byte_count;
+#ifdef R600_USE_LLVM
+ use_llvm = debug_get_bool_option("R600_LLVM", TRUE);
+#endif
ctx.bc = &shader->bc;
ctx.shader = shader;
r600_bytecode_init(ctx.bc, rctx->chip_class, rctx->family);
@@ -869,8 +1098,46 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
if (ctx.type == TGSI_PROCESSOR_FRAGMENT && ctx.bc->chip_class >= EVERGREEN) {
ctx.file_offset[TGSI_FILE_INPUT] = evergreen_gpr_count(&ctx);
}
- ctx.file_offset[TGSI_FILE_OUTPUT] = ctx.file_offset[TGSI_FILE_INPUT] +
- ctx.info.file_max[TGSI_FILE_INPUT] + 1;
+
+ /* LLVM backend setup */
+#ifdef R600_USE_LLVM
+ if (use_llvm && ctx.info.indirect_files) {
+ fprintf(stderr, "Warning: R600 LLVM backend does not support "
+ "indirect adressing. Falling back to TGSI "
+ "backend.\n");
+ use_llvm = 0;
+ }
+ if (use_llvm) {
+ struct radeon_llvm_context radeon_llvm_ctx;
+ LLVMModuleRef mod;
+ unsigned dump = 0;
+ memset(&radeon_llvm_ctx, 0, sizeof(radeon_llvm_ctx));
+ radeon_llvm_ctx.reserved_reg_count = ctx.file_offset[TGSI_FILE_INPUT];
+ mod = r600_tgsi_llvm(&radeon_llvm_ctx, tokens);
+ if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE)) {
+ dump = 1;
+ }
+ if (r600_llvm_compile(mod, &inst_bytes, &inst_byte_count,
+ rctx->family, dump)) {
+ FREE(inst_bytes);
+ radeon_llvm_dispose(&radeon_llvm_ctx);
+ use_llvm = 0;
+ fprintf(stderr, "R600 LLVM backend failed to compile "
+ "shader. Falling back to TGSI\n");
+ } else {
+ ctx.file_offset[TGSI_FILE_OUTPUT] =
+ ctx.file_offset[TGSI_FILE_INPUT];
+ }
+ radeon_llvm_dispose(&radeon_llvm_ctx);
+ }
+#endif
+ /* End of LLVM backend setup */
+
+ if (!use_llvm) {
+ ctx.file_offset[TGSI_FILE_OUTPUT] =
+ ctx.file_offset[TGSI_FILE_INPUT] +
+ ctx.info.file_max[TGSI_FILE_INPUT] + 1;
+ }
ctx.file_offset[TGSI_FILE_TEMPORARY] = ctx.file_offset[TGSI_FILE_OUTPUT] +
ctx.info.file_max[TGSI_FILE_OUTPUT] + 1;
@@ -971,6 +1238,9 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
tgsi_parse_token(&ctx.parse);
switch (ctx.parse.FullToken.Token.Type) {
case TGSI_TOKEN_TYPE_INSTRUCTION:
+ if (use_llvm) {
+ continue;
+ }
r = tgsi_is_supported(&ctx);
if (r)
goto out_err;
@@ -998,6 +1268,12 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
}
}
+ /* Get instructions if we are using the LLVM backend. */
+ if (use_llvm) {
+ r600_bytecode_from_byte_stream(&ctx, inst_bytes, inst_byte_count);
+ FREE(inst_bytes);
+ }
+
noutput = shader->noutput;
if (ctx.clip_vertex_write) {
--
1.7.7.6
More information about the mesa-dev
mailing list