[Mesa-dev] [PATCH 2/4] r600g: remove TGSI->LLVM translation
Jan Vesely
jan.vesely at rutgers.edu
Fri Mar 11 18:24:25 UTC 2016
On Fri, 2016-03-11 at 18:48 +0100, Marek Olšák wrote:
> Can you paste the build failures please?
In file included from compute_memory_pool.c:42:0:
evergreen_compute_internal.h:43:2: error: unknown type name ‘LLVMContextRef’
LLVMContextRef llvm_ctx;
^
In file included from ../../../../src/util/list.h:43:0,
from compute_memory_pool.c:29:
compute_memory_pool.c: In function ‘compute_memory_move_item’:
compute_memory_pool.c:501:47: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
assert(prev->start_in_dw + prev->size_in_dw <= new_start_in_dw);
^
compute_memory_pool.c:508:55: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
if (src != dst || new_start_in_dw + item->size_in_dw <= item->start_in_dw) {
^
Makefile:814: recipe for target 'compute_memory_pool.lo' failed
make[3]: *** [compute_memory_pool.lo] Error 1
make[3]: *** Waiting for unfinished jobs....
In file included from evergreen_compute.c:46:0:
evergreen_compute_internal.h:43:2: error: unknown type name ‘LLVMContextRef’
LLVMContextRef llvm_ctx;
^
evergreen_compute.c: In function ‘evergreen_create_compute_state’:
evergreen_compute.c:213:2: error: implicit declaration of function ‘r600_create_shader’ [-Werror=implicit-function-declaration]
r600_create_shader(&shader->bc, &shader->binary, &use_kill);
^
evergreen_compute.c: In function ‘evergreen_delete_compute_state’:
evergreen_compute.c:244:2: error: implicit declaration of function ‘r600_destroy_shader’ [-Werror=implicit-function-declaration]
r600_destroy_shader(&shader->bc);
^
evergreen_compute.c: In function ‘evergreen_launch_grid’:
evergreen_compute.c:555:2: error: implicit declaration of function ‘r600_shader_binary_read_config’ [-Werror=implicit-function-declaration]
r600_shader_binary_read_config(&shader->binary, &shader->bc,
there might be more. These are just errors displayed before the build (-j8) bailed out.
It'd be nice if the removal was done in one easily revertible patch, though it's a pity this needs to go.
Jan
>
> Marek
>
> On Fri, Mar 11, 2016 at 5:11 PM, Jan Vesely <jan.vesely at rutgers.edu>
> wrote:
> >
> > On Fri, 2016-03-11 at 16:01 +0100, Marek Olšák wrote:
> > >
> > > From: Marek Olšák <marek.olsak at amd.com>
> > >
> > > It was useful for testing and as a prototype for radeonsi
> > > bringup,
> > > but it's not used anymore and doesn't support OpenGL 3.3 even.
> > This breaks OpenCL on pre-SI (does not even build).
> >
> > Jan
> >
> > >
> > > ---
> > > configure.ac | 15 +-
> > > src/gallium/drivers/r600/Makefile.am | 8 -
> > > src/gallium/drivers/r600/Makefile.sources | 4 -
> > > src/gallium/drivers/r600/r600_llvm.c | 943 ----------------
> > > ----
> > > ----------
> > > src/gallium/drivers/r600/r600_llvm.h | 42 --
> > > src/gallium/drivers/r600/r600_pipe.c | 5 -
> > > src/gallium/drivers/r600/r600_pipe.h | 3 -
> > > src/gallium/drivers/r600/r600_shader.c | 122 +---
> > > 8 files changed, 23 insertions(+), 1119 deletions(-)
> > > delete mode 100644 src/gallium/drivers/r600/r600_llvm.c
> > > delete mode 100644 src/gallium/drivers/r600/r600_llvm.h
> > >
> > > diff --git a/configure.ac b/configure.ac
> > > index 49be147..fd6ee9d 100644
> > > --- a/configure.ac
> > > +++ b/configure.ac
> > > @@ -929,12 +929,6 @@ AC_ARG_ENABLE([xlib-glx],
> > > [enable_xlib_glx="$enableval"],
> > > [enable_xlib_glx=no])
> > >
> > > -AC_ARG_ENABLE([r600-llvm-compiler],
> > > - [AS_HELP_STRING([--enable-r600-llvm-compiler],
> > > - [Enable experimental LLVM backend for graphics shaders
> > > @<:@default=disabled@:>@])],
> > > - [enable_r600_llvm="$enableval"],
> > > - [enable_r600_llvm=no])
> > > -
> > > AC_ARG_ENABLE([gallium-tests],
> > > [AS_HELP_STRING([--enable-gallium-tests],
> > > [Enable optional Gallium tests)
> > > @<:@default=disabled@:>@])],
> > > @@ -2236,14 +2230,8 @@ if test -n "$with_gallium_drivers"; then
> > > PKG_CHECK_MODULES([RADEON], [libdrm_radeon >=
> > > $LIBDRM_RADEON_REQUIRED])
> > > gallium_require_drm "Gallium R600"
> > > gallium_require_drm_loader
> > > - if test "x$enable_r600_llvm" = xyes -o
> > > "x$enable_opencl"
> > > = xyes; then
> > > - radeon_llvm_check "r600g"
> > > - LLVM_COMPONENTS="${LLVM_COMPONENTS} bitreader
> > > asmparser"
> > > - fi
> > > - if test "x$enable_r600_llvm" = xyes; then
> > > - USE_R600_LLVM_COMPILER=yes;
> > > - fi
> > > if test "x$enable_opencl" = xyes; then
> > > + radeon_llvm_check "r600g"
> > > LLVM_COMPONENTS="${LLVM_COMPONENTS} bitreader
> > > asmparser"
> > > fi
> > > ;;
> > > @@ -2414,7 +2402,6 @@ AM_CONDITIONAL(NEED_RADEON_DRM_WINSYS, test
> > > "x$HAVE_GALLIUM_R300" = xyes -o \
> > > "x$HAVE_GALLIUM_RADE
> > > ONSI
> > > " = xyes)
> > > AM_CONDITIONAL(NEED_WINSYS_XLIB, test "x$NEED_WINSYS_XLIB" =
> > > xyes)
> > > AM_CONDITIONAL(NEED_RADEON_LLVM, test x$NEED_RADEON_LLVM = xyes)
> > > -AM_CONDITIONAL(USE_R600_LLVM_COMPILER, test
> > > x$USE_R600_LLVM_COMPILER
> > > = xyes)
> > > AM_CONDITIONAL(HAVE_GALLIUM_COMPUTE, test x$enable_opencl =
> > > xyes)
> > > AM_CONDITIONAL(HAVE_MESA_LLVM, test x$MESA_LLVM = x1)
> > > AM_CONDITIONAL(USE_VC4_SIMULATOR, test x$USE_VC4_SIMULATOR =
> > > xyes)
> > > diff --git a/src/gallium/drivers/r600/Makefile.am
> > > b/src/gallium/drivers/r600/Makefile.am
> > > index 8317da7..f3bb03e 100644
> > > --- a/src/gallium/drivers/r600/Makefile.am
> > > +++ b/src/gallium/drivers/r600/Makefile.am
> > > @@ -21,14 +21,6 @@ AM_CFLAGS += \
> > > $(LLVM_CFLAGS) \
> > > -I$(top_srcdir)/src/gallium/drivers/radeon/
> > >
> > > -libr600_la_SOURCES += \
> > > - $(LLVM_C_SOURCES)
> > > -
> > > -endif
> > > -
> > > -if USE_R600_LLVM_COMPILER
> > > -AM_CFLAGS += \
> > > - -DR600_USE_LLVM
> > > endif
> > >
> > > if HAVE_GALLIUM_COMPUTE
> > > diff --git a/src/gallium/drivers/r600/Makefile.sources
> > > b/src/gallium/drivers/r600/Makefile.sources
> > > index 024dea3..8bf8083 100644
> > > --- a/src/gallium/drivers/r600/Makefile.sources
> > > +++ b/src/gallium/drivers/r600/Makefile.sources
> > > @@ -64,7 +64,3 @@ CXX_SOURCES = \
> > > sb/sb_shader.h \
> > > sb/sb_ssa_builder.cpp \
> > > sb/sb_valtable.cpp
> > > -
> > > -LLVM_C_SOURCES = \
> > > - r600_llvm.c \
> > > - r600_llvm.h
> > > diff --git a/src/gallium/drivers/r600/r600_llvm.c
> > > b/src/gallium/drivers/r600/r600_llvm.c
> > > deleted file mode 100644
> > > index 7eab29c..0000000
> > > --- a/src/gallium/drivers/r600/r600_llvm.c
> > > +++ /dev/null
> > > @@ -1,943 +0,0 @@
> > > -#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/list.h"
> > > -#include "util/u_memory.h"
> > > -
> > > -#include "evergreend.h"
> > > -#include "r600_asm.h"
> > > -#include "r600_sq.h"
> > > -#include "r600_opcodes.h"
> > > -#include "r600_shader.h"
> > > -#include "r600_pipe.h"
> > > -#include "radeon_llvm.h"
> > > -#include "radeon_llvm_emit.h"
> > > -#include "radeon_elf_util.h"
> > > -
> > > -#include <stdio.h>
> > > -
> > > -#if defined R600_USE_LLVM || defined HAVE_OPENCL
> > > -
> > > -#define CONSTANT_BUFFER_0_ADDR_SPACE 8
> > > -#define CONSTANT_BUFFER_1_ADDR_SPACE
> > > (CONSTANT_BUFFER_0_ADDR_SPACE +
> > > R600_BUFFER_INFO_CONST_BUFFER)
> > > -#define LLVM_R600_BUFFER_INFO_CONST_BUFFER \
> > > - (CONSTANT_BUFFER_0_ADDR_SPACE +
> > > R600_BUFFER_INFO_CONST_BUFFER)
> > > -
> > > -static LLVMValueRef llvm_load_const_buffer(
> > > - struct lp_build_tgsi_context * bld_base,
> > > - LLVMValueRef OffsetValue,
> > > - unsigned ConstantAddressSpace)
> > > -{
> > > - LLVMValueRef offset[2] = {
> > > - LLVMConstInt(LLVMInt64TypeInContext(bld_base-
> > > >
> > > > base.gallivm->context), 0, false),
> > > - OffsetValue
> > > - };
> > > -
> > > - LLVMTypeRef const_ptr_type =
> > > LLVMPointerType(LLVMArrayType(LLVMVectorType(bld_base-
> > > >
> > > > base.elem_type, 4), 1024),
> > > - ConstantAdd
> > > r
> > > essSpace);
> > > - LLVMValueRef const_ptr = LLVMBuildIntToPtr(bld_base-
> > > >
> > > > base.gallivm->builder, lp_build_const_int32(bld_base-
> > > > >base.gallivm,
> > > 0), const_ptr_type, "");
> > > - LLVMValueRef ptr = LLVMBuildGEP(bld_base->base.gallivm-
> > > >
> > > > builder, const_ptr, offset, 2, "");
> > > - return LLVMBuildLoad(bld_base->base.gallivm->builder, ptr,
> > > "");
> > > -}
> > > -
> > > -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)
> > > -{
> > > - LLVMValueRef offset = lp_build_const_int32(bld_base-
> > > >
> > > > base.gallivm, reg->Register.Index);
> > > - if (reg->Register.Indirect) {
> > > - struct lp_build_tgsi_soa_context *bld =
> > > lp_soa_context(bld_base);
> > > - LLVMValueRef index = LLVMBuildLoad(bld_base-
> > > >
> > > > base.gallivm->builder, bld->addr[reg->Indirect.Index][reg-
> > > > Indirect.Swizzle], "");
> > > - offset = LLVMBuildAdd(bld_base->base.gallivm-
> > > >
> > > > builder, offset, index, "");
> > > - }
> > > - unsigned ConstantAddressSpace =
> > > CONSTANT_BUFFER_0_ADDR_SPACE
> > > ;
> > > - if (reg->Register.Dimension) {
> > > - ConstantAddressSpace += reg->Dimension.Index;
> > > - }
> > > - LLVMValueRef cvecval = llvm_load_const_buffer(bld_base,
> > > offset, ConstantAddressSpace);
> > > - LLVMValueRef cval = LLVMBuildExtractElement(bld_base-
> > > >
> > > > base.gallivm->builder, cvecval, lp_build_const_int32(bld_base-
> > > > base.gallivm, swizzle), "");
> > > - return bitcast(bld_base, type, cval);
> > > -}
> > > -
> > > -static void llvm_load_system_value(
> > > - struct radeon_llvm_context * ctx,
> > > - unsigned index,
> > > - const struct tgsi_full_declaration *decl)
> > > -{
> > > - unsigned chan;
> > > -
> > > - switch (decl->Semantic.Name) {
> > > - case TGSI_SEMANTIC_INSTANCEID: chan = 3; break;
> > > - case TGSI_SEMANTIC_VERTEXID: chan = 0; break;
> > > - default: assert(!"unknown system value");
> > > - }
> > > -
> > > - ctx->system_values[index] = LLVMBuildExtractElement(ctx-
> > > >
> > > > gallivm.builder,
> > > - LLVMGetParam(ctx->main_fn, 0),
> > > lp_build_const_int32(&(ctx->gallivm), chan),
> > > - "");
> > > -}
> > > -
> > > -static LLVMValueRef
> > > -llvm_load_input_vector(
> > > - struct radeon_llvm_context * ctx, unsigned location,
> > > unsigned ijregs,
> > > - boolean interp)
> > > -{
> > > - LLVMTypeRef VecType;
> > > - LLVMValueRef Args[3] = {
> > > - lp_build_const_int32(&(ctx->gallivm),
> > > location)
> > > - };
> > > - unsigned ArgCount = 1;
> > > - if (interp) {
> > > - VecType = LLVMVectorType(ctx-
> > > >
> > > > soa.bld_base.base.elem_type, 2);
> > > - LLVMValueRef IJIndex = LLVMGetParam(ctx-
> > > >
> > > > main_fn, ijregs / 2);
> > > - Args[ArgCount++] =
> > > LLVMBuildExtractElement(ctx->gallivm.builder, IJIndex,
> > > - lp_build_const_int32(&(ctx-
> > > >
> > > > gallivm), 2 * (ijregs % 2)), "");
> > > - Args[ArgCount++] =
> > > LLVMBuildExtractElement(ctx->gallivm.builder, IJIndex,
> > > - lp_build_const_int32(&(ctx-
> > > >
> > > > gallivm), 2 * (ijregs % 2) + 1), "");
> > > - LLVMValueRef HalfVec[2] = {
> > > - lp_build_intrinsic(ctx-
> > > >
> > > > gallivm.builder, "llvm.R600.interp.xy",
> > > - VecType, Args, ArgCount,
> > > LLVMReadNoneAttribute),
> > > - lp_build_intrinsic(ctx-
> > > >
> > > > gallivm.builder, "llvm.R600.interp.zw",
> > > - VecType, Args, ArgCount,
> > > LLVMReadNoneAttribute)
> > > - };
> > > - LLVMValueRef MaskInputs[4] = {
> > > - lp_build_const_int32(&(ctx-
> > > >
> > > > gallivm), 0),
> > > - lp_build_const_int32(&(ctx-
> > > >
> > > > gallivm), 1),
> > > - lp_build_const_int32(&(ctx-
> > > >
> > > > gallivm), 2),
> > > - lp_build_const_int32(&(ctx-
> > > >
> > > > gallivm), 3)
> > > - };
> > > - LLVMValueRef Mask =
> > > LLVMConstVector(MaskInputs, 4);
> > > - return LLVMBuildShuffleVector(ctx-
> > > >
> > > > gallivm.builder, HalfVec[0], HalfVec[1],
> > > - Mask, "");
> > > - } else {
> > > - VecType = LLVMVectorType(ctx-
> > > >
> > > > soa.bld_base.base.elem_type, 4);
> > > - return lp_build_intrinsic(ctx-
> > > >
> > > > gallivm.builder, "llvm.R600.interp.const",
> > > - VecType, Args, ArgCount,
> > > LLVMReadNoneAttribute);
> > > - }
> > > -}
> > > -
> > > -static LLVMValueRef
> > > -llvm_face_select_helper(
> > > - struct radeon_llvm_context * ctx,
> > > - LLVMValueRef face, LLVMValueRef front_color, LLVMValueRef
> > > back_color)
> > > -{
> > > - const struct lp_build_context * bb = &ctx-
> > > >
> > > > soa.bld_base.base;
> > > - LLVMValueRef is_front = LLVMBuildFCmp(
> > > - bb->gallivm->builder, LLVMRealUGT, face,
> > > - lp_build_const_float(bb->gallivm,
> > > 0.0f), "");
> > > - return LLVMBuildSelect(bb->gallivm->builder, is_front,
> > > - front_color, back_color, "");
> > > -}
> > > -
> > > -static void llvm_load_input(
> > > - struct radeon_llvm_context * ctx,
> > > - unsigned input_index,
> > > - const struct tgsi_full_declaration *decl)
> > > -{
> > > - const struct r600_shader_io * input = &ctx-
> > > >
> > > > r600_inputs[input_index];
> > > - unsigned chan;
> > > - int two_side = (ctx->two_side && input->name ==
> > > TGSI_SEMANTIC_COLOR);
> > > - LLVMValueRef v;
> > > - boolean require_interp_intrinsic = ctx->chip_class >=
> > > EVERGREEN &&
> > > - ctx->type == TGSI_PROCESSOR_FRAGMENT;
> > > -
> > > - if (require_interp_intrinsic && input->spi_sid) {
> > > - v = llvm_load_input_vector(ctx, input->lds_pos,
> > > input->ij_index,
> > > - (input->interpolate > 0));
> > > - } else
> > > - v = LLVMGetParam(ctx->main_fn, input->gpr);
> > > -
> > > - if (two_side) {
> > > - struct r600_shader_io * back_input =
> > > - &ctx->r600_inputs[input->back_color_input];
> > > - LLVMValueRef v2;
> > > - LLVMValueRef face = LLVMGetParam(ctx->main_fn, ctx-
> > > >
> > > > face_gpr);
> > > - face = LLVMBuildExtractElement(ctx-
> > > >gallivm.builder,
> > > face,
> > > - lp_build_const_int32(&(ctx->gallivm), 0),
> > > "");
> > > -
> > > - if (require_interp_intrinsic && back_input-
> > > >spi_sid)
> > > - v2 = llvm_load_input_vector(ctx,
> > > back_input-
> > > >
> > > > lds_pos,
> > > - back_input->ij_index, (back_input-
> > > >
> > > > interpolate > 0));
> > > - else
> > > - v2 = LLVMGetParam(ctx->main_fn, back_input-
> > > >
> > > > gpr);
> > > - v = llvm_face_select_helper(ctx, face, v, v2);
> > > - }
> > > -
> > > - for (chan = 0; chan < 4; chan++) {
> > > - unsigned soa_index =
> > > radeon_llvm_reg_index_soa(input_index, chan);
> > > -
> > > - ctx->inputs[soa_index] =
> > > LLVMBuildExtractElement(ctx->gallivm.builder, v,
> > > - lp_build_const_int32(&(ctx->gallivm),
> > > chan),
> > > "");
> > > -
> > > - if (input->name == TGSI_SEMANTIC_POSITION &&
> > > - ctx->type ==
> > > TGSI_PROCESSOR_FRAGMENT
> > > && chan == 3) {
> > > - /* RCP for fragcoord.w */
> > > - ctx->inputs[soa_index] = LLVMBuildFDiv(ctx-
> > > >
> > > > gallivm.builder,
> > > - lp_build_const_float(&(ctx-
> > > >
> > > > gallivm), 1.0f),
> > > - ctx->inputs[soa_index], "");
> > > - }
> > > - }
> > > -}
> > > -
> > > -static void llvm_emit_prologue(struct lp_build_tgsi_context *
> > > bld_base)
> > > -{
> > > - struct radeon_llvm_context * ctx =
> > > radeon_llvm_context(bld_base);
> > > - radeon_llvm_shader_type(ctx->main_fn, ctx->type);
> > > -
> > > -}
> > > -
> > > -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;
> > > - struct pipe_stream_output_info * so = ctx->stream_outputs;
> > > - unsigned i;
> > > - unsigned next_pos = 60;
> > > - unsigned next_param = 0;
> > > -
> > > - unsigned color_count = 0;
> > > - boolean has_color = false;
> > > -
> > > - if (ctx->type == TGSI_PROCESSOR_VERTEX && so->num_outputs)
> > > {
> > > - for (i = 0; i < so->num_outputs; i++) {
> > > - unsigned register_index = so-
> > > >
> > > > output[i].register_index;
> > > - unsigned start_component = so-
> > > >
> > > > output[i].start_component;
> > > - unsigned num_components = so-
> > > >
> > > > output[i].num_components;
> > > - unsigned dst_offset = so-
> > > >
> > > > output[i].dst_offset;
> > > - unsigned chan;
> > > - LLVMValueRef elements[4];
> > > - if (dst_offset < start_component) {
> > > - for (chan = 0; chan <
> > > TGSI_NUM_CHANNELS; chan++) {
> > > - elements[chan] =
> > > LLVMBuildLoad(base->gallivm->builder,
> > > - ctx-
> > > >
> > > > soa.outputs[register_index][(chan + start_component) %
> > > TGSI_NUM_CHANNELS], "");
> > > - }
> > > - start_component = 0;
> > > - } else {
> > > - for (chan = 0; chan <
> > > TGSI_NUM_CHANNELS; chan++) {
> > > - elements[chan] =
> > > LLVMBuildLoad(base->gallivm->builder,
> > > - ctx-
> > > >
> > > > soa.outputs[register_index][chan], "");
> > > - }
> > > - }
> > > - LLVMValueRef output =
> > > lp_build_gather_values(base->gallivm, elements, 4);
> > > - LLVMValueRef args[4];
> > > - args[0] = output;
> > > - args[1] = lp_build_const_int32(base-
> > > >
> > > > gallivm, dst_offset - start_component);
> > > - args[2] = lp_build_const_int32(base-
> > > >
> > > > gallivm, so->output[i].output_buffer);
> > > - args[3] = lp_build_const_int32(base-
> > > >
> > > > gallivm, ((1 << num_components) - 1) << start_component);
> > > - lp_build_intrinsic(base->gallivm->builder,
> > > "llvm.R600.store.stream.output",
> > > - LLVMVoidTypeInContext(base-
> > > >gallivm-
> > > >
> > > > context), args, 4, 0);
> > > - }
> > > - }
> > > -
> > > - /* Add the necessary export instructions */
> > > - for (i = 0; i < ctx->output_reg_count; i++) {
> > > - unsigned chan;
> > > - LLVMValueRef elements[4];
> > > - for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
> > > - elements[chan] = LLVMBuildLoad(base-
> > > >
> > > > gallivm->builder,
> > > - ctx->soa.outputs[i][chan], "");
> > > - }
> > > - if (ctx->alpha_to_one && ctx->type ==
> > > TGSI_PROCESSOR_FRAGMENT && ctx->r600_outputs[i].name ==
> > > TGSI_SEMANTIC_COLOR)
> > > - elements[3] = lp_build_const_float(base-
> > > >
> > > > gallivm, 1.0f);
> > > - LLVMValueRef output = lp_build_gather_values(base-
> > > >
> > > > gallivm, elements, 4);
> > > -
> > > - if (ctx->type == TGSI_PROCESSOR_VERTEX) {
> > > - switch (ctx->r600_outputs[i].name) {
> > > - case TGSI_SEMANTIC_POSITION:
> > > - case TGSI_SEMANTIC_PSIZE: {
> > > - LLVMValueRef args[3];
> > > - args[0] = output;
> > > - args[1] =
> > > lp_build_const_int32(base-
> > > >
> > > > gallivm, next_pos++);
> > > - args[2] =
> > > lp_build_const_int32(base-
> > > >
> > > > gallivm, V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS);
> > > - lp_build_intrinsic(
> > > - base->gallivm->builder,
> > > - "llvm.R600.store.swizzle",
> > > - LLVMVoidTypeInContext(base-
> > > >
> > > > gallivm->context),
> > > - args, 3, 0);
> > > - break;
> > > - }
> > > - case TGSI_SEMANTIC_CLIPVERTEX: {
> > > - LLVMValueRef args[3];
> > > - unsigned reg_index;
> > > - LLVMValueRef adjusted_elements[4];
> > > - for (reg_index = 0; reg_index < 2;
> > > reg_index ++) {
> > > - for (chan = 0; chan <
> > > TGSI_NUM_CHANNELS; chan++) {
> > > - LLVMValueRef offset
> > > = lp_build_const_int32(bld_base->base.gallivm, reg_index * 4 +
> > > chan);
> > > - LLVMValueRef
> > > base_vector = llvm_load_const_buffer(bld_base, offset,
> > > CONSTANT_BUFFER_1_ADDR_SPACE);
> > > - args[0] = output;
> > > - args[1] =
> > > base_vector;
> > > - adjusted_elements[c
> > > h
> > > an] = lp_build_intrinsic(base->gallivm->builder,
> > > - "llvm.AMDGP
> > > U
> > > .dp4", bld_base->base.elem_type,
> > > - args, 2,
> > > LLVMReadNoneAttribute);
> > > - }
> > > - args[0] =
> > > lp_build_gather_values(base->gallivm,
> > > - adjusted_elements,
> > > 4);
> > > - args[1] =
> > > lp_build_const_int32(base->gallivm, next_pos++);
> > > - args[2] =
> > > lp_build_const_int32(base->gallivm,
> > > V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS);
> > > - lp_build_intrinsic(
> > > - base->gallivm-
> > > >
> > > > builder,
> > > - "llvm.R600.store.sw
> > > i
> > > zzle",
> > > - LLVMVoidTypeInConte
> > > x
> > > t(base->gallivm->context),
> > > - args, 3, 0);
> > > - }
> > > - break;
> > > - }
> > > - case TGSI_SEMANTIC_CLIPDIST : {
> > > - LLVMValueRef args[3];
> > > - args[0] = output;
> > > - args[1] =
> > > lp_build_const_int32(base-
> > > >
> > > > gallivm, next_pos++);
> > > - args[2] =
> > > lp_build_const_int32(base-
> > > >
> > > > gallivm, V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS);
> > > - lp_build_intrinsic(
> > > - base->gallivm->builder,
> > > - "llvm.R600.store.swizzle",
> > > - LLVMVoidTypeInContext(base-
> > > >
> > > > gallivm->context),
> > > - args, 3, 0);
> > > - args[1] =
> > > lp_build_const_int32(base-
> > > >
> > > > gallivm, next_param++);
> > > - args[2] =
> > > lp_build_const_int32(base-
> > > >
> > > > gallivm, V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM);
> > > - lp_build_intrinsic(
> > > - base->gallivm->builder,
> > > - "llvm.R600.store.swizzle",
> > > - LLVMVoidTypeInContext(base-
> > > >
> > > > gallivm->context),
> > > - args, 3, 0);
> > > - break;
> > > - }
> > > - case TGSI_SEMANTIC_FOG: {
> > > - elements[0] = LLVMBuildLoad(base-
> > > >
> > > > gallivm->builder,
> > > - ctx->soa.outputs[i][0],
> > > "");
> > > - elements[1] = elements[2] =
> > > lp_build_const_float(base->gallivm, 0.0f);
> > > - elements[3] =
> > > lp_build_const_float(base->gallivm, 1.0f);
> > > -
> > > - LLVMValueRef args[3];
> > > - args[0] =
> > > lp_build_gather_values(base->gallivm, elements, 4);
> > > - args[1] =
> > > lp_build_const_int32(base-
> > > >
> > > > gallivm, next_param++);
> > > - args[2] =
> > > lp_build_const_int32(base-
> > > >
> > > > gallivm, V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM);
> > > - lp_build_intrinsic(
> > > - base->gallivm->builder,
> > > - "llvm.R600.store.swizzle",
> > > - LLVMVoidTypeInContext(base-
> > > >
> > > > gallivm->context),
> > > - args, 3, 0);
> > > - break;
> > > - }
> > > - default: {
> > > - LLVMValueRef args[3];
> > > - args[0] = output;
> > > - args[1] =
> > > lp_build_const_int32(base-
> > > >
> > > > gallivm, next_param++);
> > > - args[2] =
> > > lp_build_const_int32(base-
> > > >
> > > > gallivm, V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM);
> > > - lp_build_intrinsic(
> > > - base->gallivm->builder,
> > > - "llvm.R600.store.swizzle",
> > > - LLVMVoidTypeInContext(base-
> > > >
> > > > gallivm->context),
> > > - args, 3, 0);
> > > - break;
> > > - }
> > > - }
> > > - } else if (ctx->type == TGSI_PROCESSOR_FRAGMENT) {
> > > - switch (ctx->r600_outputs[i].name) {
> > > - case TGSI_SEMANTIC_COLOR:
> > > - has_color = true;
> > > - if ( color_count < ctx-
> > > >
> > > > color_buffer_count) {
> > > - LLVMValueRef args[3];
> > > - args[0] = output;
> > > - if (ctx->fs_color_all) {
> > > - for (unsigned j =
> > > 0;
> > > j < ctx->color_buffer_count; j++) {
> > > - args[1] =
> > > lp_build_const_int32(base->gallivm, j);
> > > - args[2] =
> > > lp_build_const_int32(base->gallivm,
> > > V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL);
> > > - lp_build_in
> > > t
> > > rinsic(
> > > - bas
> > > e
> > > ->gallivm->builder,
> > > - "ll
> > > v
> > > m.R600.store.swizzle",
> > > - LLV
> > > M
> > > VoidTypeInContext(base->gallivm->context),
> > > - arg
> > > s
> > > , 3, 0);
> > > - }
> > > - } else {
> > > - args[1] =
> > > lp_build_const_int32(base->gallivm, color_count++);
> > > - args[2] =
> > > lp_build_const_int32(base->gallivm,
> > > V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL);
> > > - lp_build_intrinsic(
> > > - base-
> > > >
> > > > gallivm->builder,
> > > - "llvm.R600.
> > > s
> > > tore.swizzle",
> > > - LLVMVoidTyp
> > > e
> > > InContext(base->gallivm->context),
> > > - args, 3,
> > > 0);
> > > - }
> > > - }
> > > - break;
> > > - case TGSI_SEMANTIC_POSITION:
> > > - lp_build_intrinsic_unary(
> > > - base->gallivm->builder,
> > > - "llvm.R600.store.pixel.dept
> > > h
> > > ",
> > > - LLVMVoidTypeInContext(base-
> > > >
> > > > gallivm->context),
> > > - LLVMBuildLoad(base-
> > > >gallivm-
> > > >
> > > > builder, ctx->soa.outputs[i][2], ""));
> > > - break;
> > > - case TGSI_SEMANTIC_STENCIL:
> > > - lp_build_intrinsic_unary(
> > > - base->gallivm->builder,
> > > - "llvm.R600.store.pixel.sten
> > > c
> > > il",
> > > - LLVMVoidTypeInContext(base-
> > > >
> > > > gallivm->context),
> > > - LLVMBuildLoad(base-
> > > >gallivm-
> > > >
> > > > builder, ctx->soa.outputs[i][1], ""));
> > > - break;
> > > - }
> > > - }
> > > - }
> > > - // Add dummy exports
> > > - if (ctx->type == TGSI_PROCESSOR_VERTEX) {
> > > - if (!next_param) {
> > > - lp_build_intrinsic_unary(base->gallivm-
> > > >
> > > > builder, "llvm.R600.store.dummy",
> > > - LLVMVoidTypeInContext(base-
> > > >gallivm-
> > > >
> > > > context),
> > > - lp_build_const_int32(base->gallivm,
> > > V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM));
> > > - }
> > > - if (!(next_pos-60)) {
> > > - lp_build_intrinsic_unary(base->gallivm-
> > > >
> > > > builder, "llvm.R600.store.dummy",
> > > - LLVMVoidTypeInContext(base-
> > > >gallivm-
> > > >
> > > > context),
> > > - lp_build_const_int32(base->gallivm,
> > > V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS));
> > > - }
> > > - }
> > > - if (ctx->type == TGSI_PROCESSOR_FRAGMENT) {
> > > - if (!has_color) {
> > > - lp_build_intrinsic_unary(base->gallivm-
> > > >
> > > > builder, "llvm.R600.store.dummy",
> > > - LLVMVoidTypeInContext(base-
> > > >gallivm-
> > > >
> > > > context),
> > > - lp_build_const_int32(base->gallivm,
> > > V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL));
> > > - }
> > > - }
> > > -
> > > -}
> > > -
> > > -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[7];
> > > - unsigned c, sampler_src;
> > > - struct radeon_llvm_context * ctx =
> > > radeon_llvm_context(bld_base);
> > > -
> > > - if (emit_data->inst->Texture.Texture ==
> > > TGSI_TEXTURE_BUFFER)
> > > {
> > > - switch (emit_data->inst->Instruction.Opcode) {
> > > - case TGSI_OPCODE_TXQ: {
> > > - struct radeon_llvm_context * ctx =
> > > radeon_llvm_context(bld_base);
> > > - ctx->uses_tex_buffers = true;
> > > - bool isEgPlus = (ctx->chip_class >=
> > > EVERGREEN);
> > > - LLVMValueRef offset =
> > > lp_build_const_int32(bld_base->base.gallivm,
> > > - isEgPlus ? 0 : 1);
> > > - LLVMValueRef cvecval =
> > > llvm_load_const_buffer(bld_base, offset,
> > > - LLVM_R600_BUFFER_INFO_CONST_BUFFER)
> > > ;
> > > - if (!isEgPlus) {
> > > - LLVMValueRef maskval[4] = {
> > > - lp_build_const_int32(galliv
> > > m
> > > , 1),
> > > - lp_build_const_int32(galliv
> > > m
> > > , 2),
> > > - lp_build_const_int32(galliv
> > > m
> > > , 3),
> > > - lp_build_const_int32(galliv
> > > m
> > > , 0),
> > > - };
> > > - LLVMValueRef mask =
> > > LLVMConstVector(maskval, 4);
> > > - cvecval =
> > > LLVMBuildShuffleVector(gallivm->builder, cvecval, cvecval,
> > > - mask, "");
> > > - }
> > > - emit_data->output[0] = cvecval;
> > > - return;
> > > - }
> > > - case TGSI_OPCODE_TXF: {
> > > - args[0] = LLVMBuildExtractElement(gallivm-
> > > >
> > > > builder, emit_data->args[0], lp_build_const_int32(gallivm, 0),
> > > > "");
> > > - args[1] = lp_build_const_int32(gallivm,
> > > R600_MAX_CONST_BUFFERS);
> > > - emit_data->output[0] =
> > > lp_build_intrinsic(gallivm->builder,
> > > - "llvm.R600.
> > > l
> > > oad.texbuf",
> > > - emit_data-
> > > >
> > > > dst_type, args, 2, LLVMReadNoneAttribute);
> > > - if (ctx->chip_class >= EVERGREEN)
> > > - return;
> > > - ctx->uses_tex_buffers = true;
> > > - LLVMDumpValue(emit_data->output[0]);
> > > - emit_data->output[0] =
> > > LLVMBuildBitCast(gallivm->builder,
> > > - emit_data->output[0],
> > > LLVMVectorType(bld_base->base.int_elem_type, 4),
> > > - "");
> > > - LLVMValueRef Mask =
> > > llvm_load_const_buffer(bld_base,
> > > - lp_build_const_int32(gallivm, 0),
> > > - LLVM_R600_BUFFER_INFO_CONST_BUFFER)
> > > ;
> > > - Mask = LLVMBuildBitCast(gallivm->builder,
> > > Mask,
> > > - LLVMVectorType(bld_base-
> > > >
> > > > base.int_elem_type, 4), "");
> > > - emit_data->output[0] =
> > > lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_AND,
> > > - emit_data->output[0],
> > > - Mask);
> > > - LLVMValueRef WComponent =
> > > LLVMBuildExtractElement(gallivm->builder,
> > > - emit_data->output[0],
> > > lp_build_const_int32(gallivm, 3), "");
> > > - Mask = llvm_load_const_buffer(bld_base,
> > > lp_build_const_int32(gallivm, 1),
> > > - LLVM_R600_BUFFER_INFO_CONST_BUFFER)
> > > ;
> > > - Mask = LLVMBuildExtractElement(gallivm-
> > > >
> > > > builder, Mask,
> > > - lp_build_const_int32(gallivm, 0),
> > > "");
> > > - Mask = LLVMBuildBitCast(gallivm->builder,
> > > Mask,
> > > - bld_base->base.int_elem_type, "");
> > > - WComponent =
> > > lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_OR,
> > > - WComponent, Mask);
> > > - emit_data->output[0] =
> > > LLVMBuildInsertElement(gallivm->builder,
> > > - emit_data->output[0], WComponent,
> > > lp_build_const_int32(gallivm, 3), "");
> > > - emit_data->output[0] =
> > > LLVMBuildBitCast(gallivm->builder,
> > > - emit_data->output[0],
> > > LLVMVectorType(bld_base->base.elem_type, 4), "");
> > > - }
> > > - return;
> > > - default:
> > > - break;
> > > - }
> > > - }
> > > -
> > > - if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_TEX
> > > >
> > > > >
> > > > >
> > > - emit_data->inst->Instruction.Opcode ==
> > > TGSI_OPCODE_TXP) {
> > > - LLVMValueRef Vector[4] = {
> > > - LLVMBuildExtractElement(gallivm->builder,
> > > emit_data->args[0],
> > > - lp_build_const_int32(gallivm, 0),
> > > ""),
> > > - LLVMBuildExtractElement(gallivm->builder,
> > > emit_data->args[0],
> > > - lp_build_const_int32(gallivm, 1),
> > > ""),
> > > - LLVMBuildExtractElement(gallivm->builder,
> > > emit_data->args[0],
> > > - lp_build_const_int32(gallivm, 2),
> > > ""),
> > > - LLVMBuildExtractElement(gallivm->builder,
> > > emit_data->args[0],
> > > - lp_build_const_int32(gallivm, 3),
> > > ""),
> > > - };
> > > - switch (emit_data->inst->Texture.Texture) {
> > > - case TGSI_TEXTURE_2D:
> > > - case TGSI_TEXTURE_RECT:
> > > - Vector[2] = Vector[3] =
> > > LLVMGetUndef(bld_base->base.elem_type);
> > > - break;
> > > - case TGSI_TEXTURE_1D:
> > > - Vector[1] = Vector[2] = Vector[3] =
> > > LLVMGetUndef(bld_base->base.elem_type);
> > > - break;
> > > - default:
> > > - break;
> > > - }
> > > - args[0] = lp_build_gather_values(gallivm, Vector,
> > > 4);
> > > - } else {
> > > - args[0] = emit_data->args[0];
> > > - }
> > > -
> > > - assert(emit_data->arg_count + 2 <= Elements(args));
> > > -
> > > - for (c = 1; c < emit_data->arg_count; ++c)
> > > - args[c] = emit_data->args[c];
> > > -
> > > - if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_TXF)
> > > {
> > > - args[1] = LLVMBuildShl(gallivm->builder, args[1],
> > > lp_build_const_int32(gallivm, 1), "");
> > > - args[2] = LLVMBuildShl(gallivm->builder, args[2],
> > > lp_build_const_int32(gallivm, 1), "");
> > > - args[3] = LLVMBuildShl(gallivm->builder, args[3],
> > > lp_build_const_int32(gallivm, 1), "");
> > > - }
> > > -
> > > - sampler_src = emit_data->inst->Instruction.NumSrcRegs-1;
> > > -
> > > - args[c++] = lp_build_const_int32(gallivm,
> > > - emit_data->inst-
> > > >
> > > > Src[sampler_src].Register.Index + R600_MAX_CONST_BUFFERS);
> > > - args[c++] = lp_build_const_int32(gallivm,
> > > - emit_data->inst-
> > > >
> > > > Src[sampler_src].Register.Index);
> > > - args[c++] = lp_build_const_int32(gallivm,
> > > - emit_data->inst-
> > > >
> > > > Texture.Texture);
> > > -
> > > - if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_TXF
> > > &&
> > > - (emit_data->inst->Texture.Texture ==
> > > TGSI_TEXTURE_2D_MSAA ||
> > > - emit_data->inst->Texture.Texture ==
> > > TGSI_TEXTURE_2D_ARRAY_MSAA)) {
> > > -
> > > - switch (emit_data->inst->Texture.Texture) {
> > > - case TGSI_TEXTURE_2D_MSAA:
> > > - args[6] = lp_build_const_int32(gallivm,
> > > TGSI_TEXTURE_2D);
> > > - break;
> > > - case TGSI_TEXTURE_2D_ARRAY_MSAA:
> > > - args[6] = lp_build_const_int32(gallivm,
> > > TGSI_TEXTURE_2D_ARRAY);
> > > - break;
> > > - default:
> > > - break;
> > > - }
> > > -
> > > - if (ctx->has_compressed_msaa_texturing) {
> > > - LLVMValueRef ldptr_args[10] = {
> > > - args[0], // Coord
> > > - args[1], // Offset X
> > > - args[2], // Offset Y
> > > - args[3], // Offset Z
> > > - args[4],
> > > - args[5],
> > > - lp_build_const_int32(gallivm, 1),
> > > - lp_build_const_int32(gallivm, 1),
> > > - lp_build_const_int32(gallivm, 1),
> > > - lp_build_const_int32(gallivm, 1)
> > > - };
> > > - LLVMValueRef ptr =
> > > lp_build_intrinsic(gallivm->builder,
> > > - "llvm.R600.ldptr",
> > > - emit_data->dst_type, ldptr_args,
> > > 10,
> > > LLVMReadNoneAttribute);
> > > - LLVMValueRef Tmp =
> > > LLVMBuildExtractElement(gallivm->builder, args[0],
> > > - lp_build_const_int32(gallivm, 3),
> > > "");
> > > - Tmp = LLVMBuildMul(gallivm->builder, Tmp,
> > > - lp_build_const_int32(gallivm, 4),
> > > "");
> > > - LLVMValueRef ResX =
> > > LLVMBuildExtractElement(gallivm->builder, ptr,
> > > - lp_build_const_int32(gallivm, 0),
> > > "");
> > > - ResX = LLVMBuildBitCast(gallivm->builder,
> > > ResX,
> > > - bld_base->base.int_elem_type, "");
> > > - Tmp = LLVMBuildLShr(gallivm->builder, ResX,
> > > Tmp, "");
> > > - Tmp = LLVMBuildAnd(gallivm->builder, Tmp,
> > > - lp_build_const_int32(gallivm, 0xF),
> > > "");
> > > - args[0] = LLVMBuildInsertElement(gallivm-
> > > >
> > > > builder, args[0], Tmp,
> > > - lp_build_const_int32(gallivm, 3),
> > > "");
> > > - args[c++] = 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,
> > > c, LLVMReadNoneAttribute);
> > > -
> > > - if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_TXQ
> > > &&
> > > - ((emit_data->inst->Texture.Texture ==
> > > TGSI_TEXTURE_CUBE_ARRAY ||
> > > - emit_data->inst->Texture.Texture ==
> > > TGSI_TEXTURE_SHADOWCUBE_ARRAY)))
> > > - if (emit_data->inst->Dst[0].Register.WriteMask & 4)
> > > {
> > > - LLVMValueRef offset =
> > > lp_build_const_int32(bld_base->base.gallivm, 0);
> > > - LLVMValueRef ZLayer =
> > > LLVMBuildExtractElement(gallivm->builder,
> > > - llvm_load_const_buffer(bld_base,
> > > offset, LLVM_R600_BUFFER_INFO_CONST_BUFFER),
> > > - lp_build_const_int32(gallivm, 0),
> > > "");
> > > -
> > > - emit_data->output[0] =
> > > LLVMBuildInsertElement(gallivm->builder, emit_data->output[0],
> > > ZLayer, lp_build_const_int32(gallivm, 2), "");
> > > - struct radeon_llvm_context * ctx =
> > > radeon_llvm_context(bld_base);
> > > - ctx->has_txq_cube_array_z_comp = true;
> > > - }
> > > -}
> > > -
> > > -static void emit_cndlt(
> > > - const struct lp_build_tgsi_action * action,
> > > - struct lp_build_tgsi_context * bld_base,
> > > - struct lp_build_emit_data * emit_data)
> > > -{
> > > - LLVMBuilderRef builder = bld_base->base.gallivm->builder;
> > > - LLVMValueRef float_zero = lp_build_const_float(
> > > - bld_base->base.gallivm, 0.0f);
> > > - LLVMValueRef cmp = LLVMBuildFCmp(
> > > - builder, LLVMRealULT, emit_data->args[0],
> > > float_zero, "");
> > > - emit_data->output[emit_data->chan] =
> > > LLVMBuildSelect(builder,
> > > - cmp, emit_data->args[1], emit_data->args[2], "");
> > > -}
> > > -
> > > -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 = build_tgsi_intrinsic_nomem,
> > > - .intr_name = "llvm.AMDGPU.dp4"
> > > -};
> > > -
> > > -static void txd_fetch_args(
> > > - struct lp_build_tgsi_context * bld_base,
> > > - struct lp_build_emit_data * emit_data)
> > > -{
> > > - const struct tgsi_full_instruction * inst = emit_data-
> > > >inst;
> > > -
> > > - LLVMValueRef coords[4];
> > > - unsigned chan, src;
> > > - for (src = 0; src < 3; src++) {
> > > - for (chan = 0; chan < 4; chan++)
> > > - coords[chan] =
> > > lp_build_emit_fetch(bld_base,
> > > inst, src, chan);
> > > -
> > > - emit_data->args[src] =
> > > lp_build_gather_values(bld_base->base.gallivm,
> > > - coords, 4);
> > > - }
> > > - emit_data->arg_count = 3;
> > > - emit_data->dst_type = LLVMVectorType(bld_base-
> > > >
> > > > base.elem_type, 4);
> > > -}
> > > -
> > > -
> > > -static void txp_fetch_args(
> > > - struct lp_build_tgsi_context * bld_base,
> > > - struct lp_build_emit_data * emit_data)
> > > -{
> > > - const struct tgsi_full_instruction * inst = emit_data-
> > > >inst;
> > > - LLVMValueRef src_w;
> > > - unsigned chan;
> > > - LLVMValueRef coords[5];
> > > -
> > > - 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;
> > > -
> > > - if ((inst->Texture.Texture == TGSI_TEXTURE_CUBE ||
> > > - inst->Texture.Texture == TGSI_TEXTURE_CUBE_ARRAY ||
> > > - inst->Texture.Texture == TGSI_TEXTURE_SHADOWCUBE ||
> > > - inst->Texture.Texture ==
> > > TGSI_TEXTURE_SHADOWCUBE_ARRAY)
> > > &&
> > > - inst->Instruction.Opcode != TGSI_OPCODE_TXQ &&
> > > - inst->Instruction.Opcode != TGSI_OPCODE_TXQ_LZ) {
> > > - radeon_llvm_emit_prepare_cube_coords(bld_base,
> > > emit_data, coords, NULL);
> > > - }
> > > -
> > > - emit_data->args[0] = lp_build_gather_values(bld_base-
> > > >
> > > > base.gallivm,
> > > - coords, 4);
> > > - emit_data->arg_count = 1;
> > > -}
> > > -
> > > -static void tex_fetch_args(
> > > - struct lp_build_tgsi_context * bld_base,
> > > - struct lp_build_emit_data * emit_data)
> > > -{
> > > - const struct tgsi_full_instruction * inst = emit_data-
> > > >inst;
> > > -
> > > - LLVMValueRef coords[5];
> > > - unsigned chan;
> > > - for (chan = 0; chan < 4; chan++) {
> > > - coords[chan] = lp_build_emit_fetch(bld_base, inst,
> > > 0, chan);
> > > - }
> > > -
> > > - if (inst->Instruction.Opcode == TGSI_OPCODE_TEX2 ||
> > > - inst->Instruction.Opcode == TGSI_OPCODE_TXB2 ||
> > > - inst->Instruction.Opcode == TGSI_OPCODE_TXL2) {
> > > - /* These instructions have additional operand that
> > > should be packed
> > > - * into the cube coord vector by
> > > radeon_llvm_emit_prepare_cube_coords.
> > > - * That operand should be passed as a float value
> > > in
> > > the args array
> > > - * right after the coord vector. After packing it's
> > > not used anymore,
> > > - * that's why arg_count is not increased */
> > > - coords[4] = lp_build_emit_fetch(bld_base, inst, 1,
> > > TGSI_CHAN_X);
> > > - }
> > > -
> > > - if ((inst->Texture.Texture == TGSI_TEXTURE_CUBE ||
> > > - inst->Texture.Texture == TGSI_TEXTURE_CUBE_ARRAY ||
> > > - inst->Texture.Texture == TGSI_TEXTURE_SHADOWCUBE ||
> > > - inst->Texture.Texture ==
> > > TGSI_TEXTURE_SHADOWCUBE_ARRAY)
> > > &&
> > > - inst->Instruction.Opcode != TGSI_OPCODE_TXQ &&
> > > - inst->Instruction.Opcode != TGSI_OPCODE_TXQ_LZ) {
> > > - radeon_llvm_emit_prepare_cube_coords(bld_base,
> > > emit_data, coords, NULL);
> > > - }
> > > -
> > > - emit_data->arg_count = 1;
> > > - emit_data->args[0] = lp_build_gather_values(bld_base-
> > > >
> > > > base.gallivm,
> > > - coords, 4);
> > > - emit_data->dst_type = LLVMVectorType(bld_base-
> > > >
> > > > base.elem_type, 4);
> > > -}
> > > -
> > > -static void txf_fetch_args(
> > > - struct lp_build_tgsi_context * bld_base,
> > > - struct lp_build_emit_data * emit_data)
> > > -{
> > > - const struct tgsi_full_instruction * inst = emit_data-
> > > >inst;
> > > - struct lp_build_tgsi_soa_context *bld =
> > > lp_soa_context(bld_base);
> > > - const struct tgsi_texture_offset * off = inst->TexOffsets;
> > > - LLVMTypeRef offset_type = bld_base->int_bld.elem_type;
> > > -
> > > - /* fetch tex coords */
> > > - tex_fetch_args(bld_base, emit_data);
> > > -
> > > - /* fetch tex offsets */
> > > - if (inst->Texture.NumOffsets) {
> > > - assert(inst->Texture.NumOffsets == 1);
> > > -
> > > - emit_data->args[1] = LLVMConstBitCast(
> > > - bld->immediates[off->Index][off->SwizzleX],
> > > - offset_type);
> > > - emit_data->args[2] = LLVMConstBitCast(
> > > - bld->immediates[off->Index][off->SwizzleY],
> > > - offset_type);
> > > - emit_data->args[3] = LLVMConstBitCast(
> > > - bld->immediates[off->Index][off->SwizzleZ],
> > > - offset_type);
> > > - } else {
> > > - emit_data->args[1] = bld_base->int_bld.zero;
> > > - emit_data->args[2] = bld_base->int_bld.zero;
> > > - emit_data->args[3] = bld_base->int_bld.zero;
> > > - }
> > > -
> > > - emit_data->arg_count = 4;
> > > -}
> > > -
> > > -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, "r600--");
> > > - LLVMTypeRef Arguments[32];
> > > - unsigned ArgumentsCount = 0;
> > > - for (unsigned i = 0; i < ctx->inputs_count; i++)
> > > - Arguments[ArgumentsCount++] =
> > > LLVMVectorType(bld_base->base.elem_type, 4);
> > > - radeon_llvm_create_func(ctx, NULL, 0, Arguments,
> > > ArgumentsCount);
> > > - for (unsigned i = 0; i < ctx->inputs_count; i++) {
> > > - LLVMValueRef P = LLVMGetParam(ctx->main_fn, i);
> > > - LLVMAddAttribute(P, LLVMInRegAttribute);
> > > - }
> > > - 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->load_input = llvm_load_input;
> > > - ctx->load_system_value = llvm_load_system_value;
> > > -
> > > - 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_DDX].intr_name =
> > > "llvm.AMDGPU.ddx";
> > > - bld_base->op_actions[TGSI_OPCODE_DDX].fetch_args =
> > > tex_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_DDX].emit = llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_DDY].intr_name =
> > > "llvm.AMDGPU.ddy";
> > > - bld_base->op_actions[TGSI_OPCODE_DDY].fetch_args =
> > > tex_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_DDY].emit = llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_TEX].fetch_args =
> > > tex_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_TEX].intr_name =
> > > "llvm.AMDGPU.tex";
> > > - bld_base->op_actions[TGSI_OPCODE_TEX].emit = llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_TEX2].fetch_args =
> > > tex_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_TEX2].intr_name =
> > > "llvm.AMDGPU.tex";
> > > - bld_base->op_actions[TGSI_OPCODE_TEX2].emit =
> > > llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_TXB].fetch_args =
> > > tex_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_TXB].intr_name =
> > > "llvm.AMDGPU.txb";
> > > - bld_base->op_actions[TGSI_OPCODE_TXB].emit = llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_TXB2].fetch_args =
> > > tex_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_TXB2].intr_name =
> > > "llvm.AMDGPU.txb";
> > > - bld_base->op_actions[TGSI_OPCODE_TXB2].emit =
> > > llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_TXD].fetch_args =
> > > txd_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_TXD].intr_name =
> > > "llvm.AMDGPU.txd";
> > > - bld_base->op_actions[TGSI_OPCODE_TXD].emit = llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_TXF].fetch_args =
> > > txf_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_TXF].intr_name =
> > > "llvm.AMDGPU.txf";
> > > - bld_base->op_actions[TGSI_OPCODE_TXF].emit = llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_TXL].fetch_args =
> > > tex_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_TXL].intr_name =
> > > "llvm.AMDGPU.txl";
> > > - bld_base->op_actions[TGSI_OPCODE_TXL].emit = llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_TXL2].fetch_args =
> > > tex_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_TXL2].intr_name =
> > > "llvm.AMDGPU.txl";
> > > - bld_base->op_actions[TGSI_OPCODE_TXL2].emit =
> > > llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_TXP].fetch_args =
> > > txp_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_TXP].intr_name =
> > > "llvm.AMDGPU.tex";
> > > - bld_base->op_actions[TGSI_OPCODE_TXP].emit = llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_TXQ].fetch_args =
> > > tex_fetch_args;
> > > - bld_base->op_actions[TGSI_OPCODE_TXQ].intr_name =
> > > "llvm.AMDGPU.txq";
> > > - bld_base->op_actions[TGSI_OPCODE_TXQ].emit = llvm_emit_tex;
> > > - bld_base->op_actions[TGSI_OPCODE_CMP].emit = emit_cndlt;
> > > -
> > > - lp_build_tgsi_llvm(bld_base, tokens);
> > > -
> > > - LLVMBuildRetVoid(bld_base->base.gallivm->builder);
> > > - radeon_llvm_finalize_module(ctx);
> > > -
> > > - return ctx->gallivm.module;
> > > -}
> > > -
> > > -/* We need to define these R600 registers here, because we can't
> > > include
> > > - * evergreend.h and r600d.h.
> > > - */
> > > -#define R_028868_SQ_PGM_RESOURCES_VS 0x028868
> > > -#define R_028850_SQ_PGM_RESOURCES_PS 0x028850
> > > -
> > > -void r600_shader_binary_read_config(const struct
> > > radeon_shader_binary *binary,
> > > - struct r600_bytecode *bc,
> > > - uint64_t symbol_offset,
> > > - boolean *use_kill)
> > > -{
> > > - unsigned i;
> > > - const unsigned char *config =
> > > - radeon_shader_binary_config_start(binary,
> > > symbol_offset);
> > > -
> > > - for (i = 0; i < binary->config_size_per_symbol; i+= 8) {
> > > - unsigned reg =
> > > - util_le32_to_cpu(*(uint32_t*)(config + i));
> > > - unsigned value =
> > > - util_le32_to_cpu(*(uint32_t*)(config + i +
> > > 4));
> > > - switch (reg) {
> > > - /* R600 / R700 */
> > > - case R_028850_SQ_PGM_RESOURCES_PS:
> > > - case R_028868_SQ_PGM_RESOURCES_VS:
> > > - /* Evergreen / Northern Islands */
> > > - case R_028844_SQ_PGM_RESOURCES_PS:
> > > - case R_028860_SQ_PGM_RESOURCES_VS:
> > > - case R_0288D4_SQ_PGM_RESOURCES_LS:
> > > - bc->ngpr = MAX2(bc->ngpr,
> > > G_028844_NUM_GPRS(value));
> > > - bc->nstack = MAX2(bc->nstack,
> > > G_028844_STACK_SIZE(value));
> > > - break;
> > > - case R_02880C_DB_SHADER_CONTROL:
> > > - *use_kill = G_02880C_KILL_ENABLE(value);
> > > - break;
> > > - case R_0288E8_SQ_LDS_ALLOC:
> > > - bc->nlds_dw = value;
> > > - break;
> > > - }
> > > - }
> > > -
> > > -}
> > > -
> > > -unsigned r600_create_shader(struct r600_bytecode *bc,
> > > - const struct radeon_shader_binary *binary,
> > > - boolean *use_kill)
> > > -
> > > -{
> > > - assert(binary->code_size % 4 == 0);
> > > - bc->bytecode = CALLOC(1, binary->code_size);
> > > - memcpy(bc->bytecode, binary->code, binary->code_size);
> > > - bc->ndw = binary->code_size / 4;
> > > -
> > > - r600_shader_binary_read_config(binary, bc, 0, use_kill);
> > > -
> > > - return 0;
> > > -}
> > > -
> > > -void r600_destroy_shader(struct r600_bytecode *bc)
> > > -{
> > > - FREE(bc->bytecode);
> > > -}
> > > -
> > > -unsigned r600_llvm_compile(
> > > - LLVMModuleRef mod,
> > > - enum radeon_family family,
> > > - struct r600_bytecode *bc,
> > > - boolean *use_kill,
> > > - unsigned dump,
> > > - struct pipe_debug_callback *debug)
> > > -{
> > > - unsigned r;
> > > - struct radeon_shader_binary binary;
> > > - const char * gpu_family =
> > > r600_get_llvm_processor_name(family);
> > > -
> > > - radeon_shader_binary_init(&binary);
> > > - if (dump)
> > > - LLVMDumpModule(mod);
> > > - r = radeon_llvm_compile(mod, &binary, gpu_family, NULL,
> > > debug);
> > > -
> > > - r = r600_create_shader(bc, &binary, use_kill);
> > > -
> > > - radeon_shader_binary_clean(&binary);
> > > -
> > > - return r;
> > > -}
> > > -
> > > -#endif
> > > diff --git a/src/gallium/drivers/r600/r600_llvm.h
> > > b/src/gallium/drivers/r600/r600_llvm.h
> > > deleted file mode 100644
> > > index 3f7fc4b..0000000
> > > --- a/src/gallium/drivers/r600/r600_llvm.h
> > > +++ /dev/null
> > > @@ -1,42 +0,0 @@
> > > -
> > > -#ifndef R600_LLVM_H
> > > -#define R600_LLVM_H
> > > -
> > > -#if defined R600_USE_LLVM || defined HAVE_OPENCL
> > > -
> > > -#include "radeon/radeon_llvm.h"
> > > -#include <llvm-c/Core.h>
> > > -
> > > -struct pipe_debug_callback;
> > > -struct r600_bytecode;
> > > -struct r600_shader_ctx;
> > > -struct radeon_llvm_context;
> > > -struct radeon_shader_binary;
> > > -enum radeon_family;
> > > -
> > > -LLVMModuleRef r600_tgsi_llvm(
> > > - struct radeon_llvm_context * ctx,
> > > - const struct tgsi_token * tokens);
> > > -
> > > -unsigned r600_llvm_compile(
> > > - LLVMModuleRef mod,
> > > - enum radeon_family family,
> > > - struct r600_bytecode *bc,
> > > - boolean *use_kill,
> > > - unsigned dump,
> > > - struct pipe_debug_callback *debug);
> > > -
> > > -unsigned r600_create_shader(struct r600_bytecode *bc,
> > > - const struct radeon_shader_binary *binary,
> > > - boolean *use_kill);
> > > -
> > > -void r600_destroy_shader(struct r600_bytecode *bc);
> > > -
> > > -void r600_shader_binary_read_config(const struct
> > > radeon_shader_binary *binary,
> > > - struct r600_bytecode *bc,
> > > - uint64_t symbol_offset,
> > > - boolean *use_kill);
> > > -
> > > -#endif /* defined R600_USE_LLVM || defined HAVE_OPENCL */
> > > -
> > > -#endif /* R600_LLVM_H */
> > > diff --git a/src/gallium/drivers/r600/r600_pipe.c
> > > b/src/gallium/drivers/r600/r600_pipe.c
> > > index 88c500a..b801191 100644
> > > --- a/src/gallium/drivers/r600/r600_pipe.c
> > > +++ b/src/gallium/drivers/r600/r600_pipe.c
> > > @@ -43,9 +43,6 @@
> > >
> > > static const struct debug_named_value r600_debug_options[] = {
> > > /* features */
> > > -#if defined(R600_USE_LLVM)
> > > - { "llvm", DBG_LLVM, "Enable the LLVM shader compiler" },
> > > -#endif
> > > { "nocpdma", DBG_NO_CP_DMA, "Disable CP DMA" },
> > >
> > > /* shader backend */
> > > @@ -620,8 +617,6 @@ struct pipe_screen *r600_screen_create(struct
> > > radeon_winsys *ws)
> > > rscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS
> > > |
> > > DBG_PS | DBG_CS | DBG_TCS | DBG_TES;
> > > if (!debug_get_bool_option("R600_HYPERZ", TRUE))
> > > rscreen->b.debug_flags |= DBG_NO_HYPERZ;
> > > - if (debug_get_bool_option("R600_LLVM", FALSE))
> > > - rscreen->b.debug_flags |= DBG_LLVM;
> > >
> > > if (rscreen->b.family == CHIP_UNKNOWN) {
> > > fprintf(stderr, "r600: Unknown chipset 0x%04X\n",
> > > rscreen->b.info.pci_id);
> > > diff --git a/src/gallium/drivers/r600/r600_pipe.h
> > > b/src/gallium/drivers/r600/r600_pipe.h
> > > index 72aa642..cd0052a 100644
> > > --- a/src/gallium/drivers/r600/r600_pipe.h
> > > +++ b/src/gallium/drivers/r600/r600_pipe.h
> > > @@ -28,8 +28,6 @@
> > >
> > > #include "radeon/r600_pipe_common.h"
> > > #include "radeon/r600_cs.h"
> > > -
> > > -#include "r600_llvm.h"
> > > #include "r600_public.h"
> > >
> > > #include "util/u_suballoc.h"
> > > @@ -243,7 +241,6 @@ struct r600_gs_rings_state {
> > >
> > > /* This must start from 16. */
> > > /* features */
> > > -#define DBG_LLVM (1 << 29)
> > > #define DBG_NO_CP_DMA (1 << 30)
> > > /* shader backend */
> > > #define DBG_NO_SB (1 << 21)
> > > diff --git a/src/gallium/drivers/r600/r600_shader.c
> > > b/src/gallium/drivers/r600/r600_shader.c
> > > index df40f94..49a1876 100644
> > > --- a/src/gallium/drivers/r600/r600_shader.c
> > > +++ b/src/gallium/drivers/r600/r600_shader.c
> > > @@ -21,7 +21,6 @@
> > > * USE OR OTHER DEALINGS IN THE SOFTWARE.
> > > */
> > > #include "r600_sq.h"
> > > -#include "r600_llvm.h"
> > > #include "r600_formats.h"
> > > #include "r600_opcodes.h"
> > > #include "r600_shader.h"
> > > @@ -194,10 +193,7 @@ int r600_pipe_shader_create(struct
> > > pipe_context
> > > *ctx,
> > > /* disable SB for shaders using doubles */
> > > use_sb &= !shader->shader.uses_doubles;
> > >
> > > - /* Check if the bytecode has already been built. When
> > > using
> > > the llvm
> > > - * backend, r600_shader_from_tgsi() will take care of
> > > building the
> > > - * bytecode.
> > > - */
> > > + /* Check if the bytecode has already been built. */
> > > if (!shader->shader.bc.bytecode) {
> > > r = r600_bytecode_build(&shader->shader.bc);
> > > if (r) {
> > > @@ -332,7 +328,6 @@ struct r600_shader_ctx {
> > > uint32_t *literals;
> > > uint32_t nliterals;
> > > uint32_t max_driver_temp_use
> > > d
> > > ;
> > > - boolean use_llvm;
> > > /* needed for evergreen interpolation */
> > > struct eg_interp eg_interpolators[6]; //
> > > indexed by Persp/Linear * 3 + sample/center/centroid
> > > /* evergreen/cayman also store sample mask in face register
> > > */
> > > @@ -661,11 +656,9 @@ static int evergreen_interp_input(struct
> > > r600_shader_ctx *ctx, int index)
> > > ctx->shader->input[index].lds_pos = ctx->shader-
> > > >
> > > > nlds++;
> > > if (ctx->shader->input[index].interpolate > 0) {
> > > evergreen_interp_assign_ij_index(ctx,
> > > index);
> > > - if (!ctx->use_llvm)
> > > - r = evergreen_interp_alu(ctx,
> > > index);
> > > + r = evergreen_interp_alu(ctx, index);
> > > } else {
> > > - if (!ctx->use_llvm)
> > > - r = evergreen_interp_flat(ctx,
> > > index);
> > > + r = evergreen_interp_flat(ctx, index);
> > > }
> > > }
> > > return r;
> > > @@ -2936,22 +2929,16 @@ static int r600_shader_from_tgsi(struct
> > > r600_context *rctx,
> > > int i, j, k, r = 0;
> > > int next_param_base = 0, next_clip_base;
> > > int max_color_exports = MAX2(key.ps.nr_cbufs, 1);
> > > - /* Declarations used by llvm code */
> > > - bool use_llvm = false;
> > > bool indirect_gprs;
> > > bool ring_outputs = false;
> > > bool lds_outputs = false;
> > > bool lds_inputs = false;
> > > bool pos_emitted = false;
> > >
> > > -#ifdef R600_USE_LLVM
> > > - use_llvm = rscreen->b.debug_flags & DBG_LLVM;
> > > -#endif
> > > ctx.bc = &shader->bc;
> > > ctx.shader = shader;
> > > ctx.native_integers = true;
> > >
> > > -
> > > r600_bytecode_init(ctx.bc, rscreen->b.chip_class, rscreen-
> > > >
> > > > b.family,
> > > rscreen->has_compressed_msaa_texturing);
> > > ctx.tokens = tokens;
> > > @@ -3043,19 +3030,9 @@ static int r600_shader_from_tgsi(struct
> > > r600_context *rctx,
> > > ctx.file_offset[i] = 0;
> > > }
> > >
> > > -#ifdef R600_USE_LLVM
> > > - if (use_llvm && ctx.info.indirect_files &&
> > > (ctx.info.indirect_files & (1 << TGSI_FILE_CONSTANT)) !=
> > > ctx.info.indirect_files) {
> > > - fprintf(stderr, "Warning: R600 LLVM backend does
> > > not
> > > support "
> > > - "indirect adressing. Falling back
> > > to TGSI "
> > > - "backend.\n");
> > > - use_llvm = 0;
> > > - }
> > > -#endif
> > > if (ctx.type == TGSI_PROCESSOR_VERTEX) {
> > > ctx.file_offset[TGSI_FILE_INPUT] = 1;
> > > - if (!use_llvm) {
> > > - r600_bytecode_add_cfinst(ctx.bc,
> > > CF_OP_CALL_FS);
> > > - }
> > > + r600_bytecode_add_cfinst(ctx.bc, CF_OP_CALL_FS);
> > > }
> > > if (ctx.type == TGSI_PROCESSOR_FRAGMENT) {
> > > if (ctx.bc->chip_class >= EVERGREEN)
> > > @@ -3085,16 +3062,10 @@ static int r600_shader_from_tgsi(struct
> > > r600_context *rctx,
> > > if (add_tess_inout)
> > > ctx.file_offset[TGSI_FILE_INPUT]+=2;
> > > }
> > > - ctx.use_llvm = use_llvm;
> > >
> > > - if (use_llvm) {
> > > - ctx.file_offset[TGSI_FILE_OUTPUT] =
> > > - ctx.file_offset[TGSI_FILE_INPUT];
> > > - } else {
> > > - ctx.file_offset[TGSI_FILE_OUTPUT] =
> > > + 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[T
> > > G
> > > SI_FILE_OUTPUT] + 1;
> > >
> > > @@ -3234,54 +3205,10 @@ static int r600_shader_from_tgsi(struct
> > > r600_context *rctx,
> > > }
> > > }
> > >
> > > -/* LLVM backend setup */
> > > -#ifdef R600_USE_LLVM
> > > - if (use_llvm) {
> > > - struct radeon_llvm_context radeon_llvm_ctx;
> > > - LLVMModuleRef mod;
> > > - bool dump = r600_can_dump_shader(&rscreen->b,
> > > - tgsi_get_processor
> > > _
> > > type(tokens));
> > > - boolean use_kill = false;
> > > -
> > > - memset(&radeon_llvm_ctx, 0,
> > > sizeof(radeon_llvm_ctx));
> > > - radeon_llvm_ctx.type = ctx.type;
> > > - radeon_llvm_ctx.two_side = shader->two_side;
> > > - radeon_llvm_ctx.face_gpr = ctx.face_gpr;
> > > - radeon_llvm_ctx.inputs_count = ctx.shader->ninput +
> > > 1;
> > > - radeon_llvm_ctx.r600_inputs = ctx.shader->input;
> > > - radeon_llvm_ctx.r600_outputs = ctx.shader->output;
> > > - radeon_llvm_ctx.color_buffer_count =
> > > max_color_exports;
> > > - radeon_llvm_ctx.chip_class = ctx.bc->chip_class;
> > > - radeon_llvm_ctx.fs_color_all = shader->fs_write_all
> > > && (rscreen->b.chip_class >= EVERGREEN);
> > > - radeon_llvm_ctx.stream_outputs = &so;
> > > - radeon_llvm_ctx.alpha_to_one = key.ps.alpha_to_one;
> > > - radeon_llvm_ctx.has_compressed_msaa_texturing =
> > > - ctx.bc->has_compressed_msaa_texturing;
> > > - mod = r600_tgsi_llvm(&radeon_llvm_ctx, tokens);
> > > - ctx.shader->has_txq_cube_array_z_comp =
> > > radeon_llvm_ctx.has_txq_cube_array_z_comp;
> > > - ctx.shader->uses_tex_buffers =
> > > radeon_llvm_ctx.uses_tex_buffers;
> > > -
> > > - if (r600_llvm_compile(mod, rscreen->b.family,
> > > ctx.bc, &use_kill,
> > > - dump, &rctx->b.debug)) {
> > > - 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_I
> > > N
> > > PUT];
> > > - }
> > > - if (use_kill)
> > > - ctx.shader->uses_kill = use_kill;
> > > - radeon_llvm_dispose(&radeon_llvm_ctx);
> > > - }
> > > -#endif
> > > -/* End of LLVM backend setup */
> > > -
> > > if (shader->fs_write_all && rscreen->b.chip_class >=
> > > EVERGREEN)
> > > shader->nr_ps_max_color_exports = 8;
> > >
> > > - if (!use_llvm) {
> > > + if (1) {
> > > if (ctx.fragcoord_input >= 0) {
> > > if (ctx.bc->chip_class == CAYMAN) {
> > > for (j = 0 ; j < 4; j++) {
> > > @@ -3437,8 +3364,7 @@ static int r600_shader_from_tgsi(struct
> > > r600_context *rctx,
> > > alu.dst.write = (j == ochan);
> > > if (j == 3)
> > > alu.last = 1;
> > > - if (!use_llvm)
> > > - r =
> > > r600_bytecode_add_alu(ctx.bc, &alu);
> > > + r = r600_bytecode_add_alu(ctx.bc,
> > > &alu);
> > > if (r)
> > > return r;
> > > }
> > > @@ -3446,7 +3372,7 @@ static int r600_shader_from_tgsi(struct
> > > r600_context *rctx,
> > > }
> > >
> > > /* Add stream outputs. */
> > > - if (!use_llvm && so.num_outputs) {
> > > + if (so.num_outputs) {
> > > bool emit = false;
> > > if (!lds_outputs && !ring_outputs && ctx.type ==
> > > TGSI_PROCESSOR_VERTEX)
> > > emit = true;
> > > @@ -3709,31 +3635,27 @@ static int r600_shader_from_tgsi(struct
> > > r600_context *rctx,
> > > }
> > > }
> > > /* add output to bytecode */
> > > - if (!use_llvm) {
> > > - for (i = 0; i < noutput; i++) {
> > > - r =
> > > r600_bytecode_add_output(ctx.bc,
> > > &output[i]);
> > > - if (r)
> > > - goto out_err;
> > > - }
> > > + for (i = 0; i < noutput; i++) {
> > > + r = r600_bytecode_add_output(ctx.bc,
> > > &output[i]);
> > > + if (r)
> > > + goto out_err;
> > > }
> > > }
> > >
> > > /* add program end */
> > > - if (!use_llvm) {
> > > - if (ctx.bc->chip_class == CAYMAN)
> > > - cm_bytecode_add_cf_end(ctx.bc);
> > > - else {
> > > - const struct cf_op_info *last = NULL;
> > > + if (ctx.bc->chip_class == CAYMAN)
> > > + cm_bytecode_add_cf_end(ctx.bc);
> > > + else {
> > > + const struct cf_op_info *last = NULL;
> > >
> > > - if (ctx.bc->cf_last)
> > > - last = r600_isa_cf(ctx.bc->cf_last-
> > > >
> > > > op);
> > > + if (ctx.bc->cf_last)
> > > + last = r600_isa_cf(ctx.bc->cf_last->op);
> > >
> > > - /* alu clause instructions don't have EOP
> > > bit, so add NOP */
> > > - if (!last || last->flags & CF_ALU ||
> > > ctx.bc-
> > > >
> > > > cf_last->op == CF_OP_LOOP_END || ctx.bc->cf_last->op ==
> > > CF_OP_CALL_FS || ctx.bc->cf_last->op == CF_OP_POP || ctx.bc-
> > > >cf_last-
> > > >
> > > > op == CF_OP_GDS)
> > > - r600_bytecode_add_cfinst(ctx.bc,
> > > CF_OP_NOP);
> > > + /* alu clause instructions don't have EOP bit, so
> > > add NOP */
> > > + if (!last || last->flags & CF_ALU || ctx.bc-
> > > >
> > > > cf_last->op == CF_OP_LOOP_END || ctx.bc->cf_last->op ==
> > > CF_OP_CALL_FS || ctx.bc->cf_last->op == CF_OP_POP || ctx.bc-
> > > >cf_last-
> > > >
> > > > op == CF_OP_GDS)
> > > + r600_bytecode_add_cfinst(ctx.bc,
> > > CF_OP_NOP);
> > >
> > > - ctx.bc->cf_last->end_of_program = 1;
> > > - }
> > > + ctx.bc->cf_last->end_of_program = 1;
> > > }
> > >
> > > /* check GPR limit - we have 124 = 128 - 4
> > --
> > Jan Vesely <jan.vesely at rutgers.edu>
--
Jan Vesely <jan.vesely at rutgers.edu>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: This is a digitally signed message part
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160311/25912655/attachment-0001.sig>
More information about the mesa-dev
mailing list