On Mon, 2018-07-02 at 09:32 +1000, Timothy Arceri wrote:
> Reviewed-by: Timothy Arceri <tarc...@itsqueeze.com>
>
> On 02/07/18 05:50, Marek Olšák wrote:
> > From: Marek Olšák <marek.ol...@amd.com>
> >
> > Users are encouraged to switch to LLVM 6.0 released in March 2018.
> > ---
> > .travis.yml | 24 ++--
> > configure.ac | 4 +-
> > meson.build | 4 +-
> > src/amd/common/ac_llvm_build.c | 129
> > +++++-----------------
> > src/amd/common/ac_llvm_helper.cpp | 4 -
> > src/gallium/drivers/radeonsi/si_shader.c | 24 ----
> > src/gallium/drivers/radeonsi/si_shader_nir.c | 7 --
> > src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c | 8 +-
> > 8 files changed, 45 insertions(+), 159 deletions(-)
> >
> > diff --git a/.travis.yml b/.travis.yml
> > index 012cc91..2f04747 100644
> > --- a/.travis.yml
> > +++ b/.travis.yml
> > @@ -118,37 +118,37 @@ matrix:
> > - x11proto-xf86vidmode-dev
> > - libexpat1-dev
> > - libx11-xcb-dev
> > - libelf-dev
> > - libunwind8-dev
> > - env:
> > - LABEL="make Gallium Drivers RadeonSI"
> > - BUILD=make
> > - MAKEFLAGS="-j4"
> > - MAKE_CHECK_COMMAND="true"
> > - - LLVM_VERSION=5.0
> > + - LLVM_VERSION=6.0
> > - LLVM_CONFIG="llvm-config-${LLVM_VERSION}"
> > - DRI_LOADERS="--disable-glx --disable-gbm --disable-egl"
> > - DRI_DRIVERS=""
> > - GALLIUM_ST="--enable-dri --disable-opencl --disable-xa
> > --disable-nine --disable-xvmc --disable-vdpau --disable-va
> > --disable-omx-bellagio --disable-gallium-osmesa"
> > - GALLIUM_DRIVERS="radeonsi"
> > - VULKAN_DRIVERS=""
> > - LIBUNWIND_FLAGS="--enable-libunwind"
> > addons:
> > apt:
> > sources:
> > - - llvm-toolchain-trusty-5.0
> > + - llvm-toolchain-trusty-6.0
> > packages:
> > # LLVM packaging is broken and misses these dependencies
> > - libedit-dev
> > # From sources above
> > - - llvm-5.0-dev
> > + - llvm-6.0-dev
> > # Common
> > - xz-utils
> > - x11proto-xf86vidmode-dev
> > - libexpat1-dev
> > - libx11-xcb-dev
> > - libelf-dev
> > - libunwind8-dev
> > - env:
> > - LABEL="make Gallium Drivers Other"
> > - BUILD=make
> > @@ -251,47 +251,47 @@ matrix:
> > - libclang-4.0-dev
> > # Common
> > - xz-utils
> > - x11proto-xf86vidmode-dev
> > - libexpat1-dev
> > - libx11-xcb-dev
> > - libelf-dev
> > - libunwind8-dev
> > - env:
> > # NOTE: Analogous to SWR above, building Clover is quite slow.
> > - - LABEL="make Gallium ST Clover LLVM-5.0"
> > + - LABEL="make Gallium ST Clover LLVM-6.0"
> > - BUILD=make
> > - MAKEFLAGS="-j4"
> > - MAKE_CHECK_COMMAND="true"
> > - - LLVM_VERSION=5.0
> > + - LLVM_VERSION=6.0
> > - LLVM_CONFIG="llvm-config-${LLVM_VERSION}"
> > - OVERRIDE_CC=gcc-4.8
> > - OVERRIDE_CXX=g++-4.8
> > - DRI_LOADERS="--disable-glx --disable-gbm --disable-egl"
> > - DRI_DRIVERS=""
> > - GALLIUM_ST="--disable-dri --enable-opencl --enable-opencl-icd
> > --enable-llvm --disable-xa --disable-nine --disable-xvmc --disable-vdpau
> > --disable-va --disable-omx-bellagio --disable-gallium-osmesa"
> > - GALLIUM_DRIVERS="r600,radeonsi"
> > - VULKAN_DRIVERS=""
> > - LIBUNWIND_FLAGS="--enable-libunwind"
> > addons:
> > apt:
> > sources:
> > - - llvm-toolchain-trusty-5.0
> > + - llvm-toolchain-trusty-6.0
> > packages:
> > - libclc-dev
> > # LLVM packaging is broken and misses these dependencies
> > - libedit-dev
> > - g++-4.8
> > # From sources above
> > - - llvm-5.0-dev
> > - - clang-5.0
> > - - libclang-5.0-dev
> > + - llvm-6.0-dev
> > + - clang-6.0
> > + - libclang-6.0-dev
> > # Common
> > - xz-utils
> > - x11proto-xf86vidmode-dev
> > - libexpat1-dev
> > - libx11-xcb-dev
> > - libelf-dev
> > - libunwind8-dev
The clover parts are wrong. There already is LLVM-6 testing below, no
need to duplicate it. All you need to do is to drop radeonsi from
llvm-5 env.
Jan
> > - env:
> > # NOTE: Analogous to SWR above, building Clover is quite slow.
> > - LABEL="make Gallium ST Clover LLVM-6.0"
> > @@ -359,37 +359,37 @@ matrix:
> > - x11proto-xf86vidmode-dev
> > - libexpat1-dev
> > - libx11-xcb-dev
> > - libelf-dev
> > - libunwind8-dev
> > - env:
> > - LABEL="make Vulkan"
> > - BUILD=make
> > - MAKEFLAGS="-j4"
> > - MAKE_CHECK_COMMAND="make -C src/gtest check && make -C
> > src/intel check"
> > - - LLVM_VERSION=5.0
> > + - LLVM_VERSION=6.0
> > - LLVM_CONFIG="llvm-config-${LLVM_VERSION}"
> > - DRI_LOADERS="--disable-glx --disable-gbm --disable-egl
> > --with-platforms=x11,wayland"
> > - DRI_DRIVERS=""
> > - GALLIUM_ST="--enable-dri --enable-dri3 --disable-opencl
> > --disable-xa --disable-nine --disable-xvmc --disable-vdpau --disable-va
> > --disable-omx-bellagio --disable-gallium-osmesa"
> > - GALLIUM_DRIVERS=""
> > - VULKAN_DRIVERS="intel,radeon"
> > - LIBUNWIND_FLAGS="--disable-libunwind"
> > addons:
> > apt:
> > sources:
> > - - llvm-toolchain-trusty-5.0
> > + - llvm-toolchain-trusty-6.0
> > packages:
> > # LLVM packaging is broken and misses these dependencies
> > - libedit-dev
> > # From sources above
> > - - llvm-5.0-dev
> > + - llvm-6.0-dev
> > # Common
> > - xz-utils
> > - x11proto-xf86vidmode-dev
> > - libexpat1-dev
> > - libx11-xcb-dev
> > - libelf-dev
> > - env:
> > - LABEL="scons"
> > - BUILD=scons
> > - SCONSFLAGS="-j4"
> > diff --git a/configure.ac b/configure.ac
> > index f135d05..58cf56e 100644
> > --- a/configure.ac
> > +++ b/configure.ac
> > @@ -101,22 +101,22 @@ XDAMAGE_REQUIRED=1.1
> > XSHMFENCE_REQUIRED=1.1
> > XVMC_REQUIRED=1.0.6
> > PYTHON_MAKO_REQUIRED=0.8.0
> > LIBSENSORS_REQUIRED=4.0.0
> > ZLIB_REQUIRED=1.2.3
> >
> > dnl LLVM versions
> > LLVM_REQUIRED_GALLIUM=3.3.0
> > LLVM_REQUIRED_OPENCL=3.9.0
> > LLVM_REQUIRED_R600=3.9.0
> > -LLVM_REQUIRED_RADEONSI=5.0.0
> > -LLVM_REQUIRED_RADV=5.0.0
> > +LLVM_REQUIRED_RADEONSI=6.0.0
> > +LLVM_REQUIRED_RADV=6.0.0
> > LLVM_REQUIRED_SWR=5.0.0
> >
> > dnl Check for progs
> > AC_PROG_CPP
> > AC_PROG_CC
> > AC_PROG_CXX
> > dnl add this here, so the help for this environmnet variable is close to
> > dnl other CC/CXX flags related help
> > AC_ARG_VAR([CXX11_CXXFLAGS], [Compiler flag to enable C++11 support (only
> > needed if not
> > enabled by default and different from
> > -std=c++11)])
> > diff --git a/meson.build b/meson.build
> > index b2722c7..9b5b53e 100644
> > --- a/meson.build
> > +++ b/meson.build
> > @@ -1140,21 +1140,23 @@ if with_amd_vk or with_gallium_radeonsi or
> > with_gallium_r600
> > endif
> > endif
> > if with_gallium_opencl
> > llvm_modules += [
> > 'all-targets', 'linker', 'coverage', 'instrumentation', 'ipo',
> > 'irreader',
> > 'lto', 'option', 'objcarcopts', 'profiledata',
> > ]
> > llvm_optional_modules += ['coroutines', 'opencl']
> > endif
> >
> > -if with_amd_vk or with_gallium_radeonsi or with_gallium_swr
> > +if with_amd_vk or with_gallium_radeonsi
> > + _llvm_version = '>= 6.0.0'
> > +elif with_gallium_swr
> > _llvm_version = '>= 5.0.0'
> > elif with_gallium_opencl or with_gallium_r600
> > _llvm_version = '>= 3.9.0'
> > else
> > _llvm_version = '>= 3.3.0'
> > endif
> >
> > _llvm = get_option('llvm')
> > if _llvm == 'auto'
> > dep_llvm = dependency(
> > diff --git a/src/amd/common/ac_llvm_build.c b/src/amd/common/ac_llvm_build.c
> > index 93ae273..d2ada00 100644
> > --- a/src/amd/common/ac_llvm_build.c
> > +++ b/src/amd/common/ac_llvm_build.c
> > @@ -1819,189 +1819,110 @@ LLVMValueRef ac_build_cvt_pkrtz_f16(struct
> > ac_llvm_context *ctx,
> > {
> > LLVMTypeRef v2f16 =
> > LLVMVectorType(LLVMHalfTypeInContext(ctx->context), 2);
> > LLVMValueRef res =
> > ac_build_intrinsic(ctx, "llvm.amdgcn.cvt.pkrtz",
> > v2f16, args, 2,
> > AC_FUNC_ATTR_READNONE);
> > return LLVMBuildBitCast(ctx->builder, res, ctx->i32, "");
> > }
> >
> > -/* Upper 16 bits must be zero. */
> > -static LLVMValueRef ac_llvm_pack_two_int16(struct ac_llvm_context *ctx,
> > - LLVMValueRef val[2])
> > -{
> > - return LLVMBuildOr(ctx->builder, val[0],
> > - LLVMBuildShl(ctx->builder, val[1],
> > - LLVMConstInt(ctx->i32, 16, 0),
> > - ""), "");
> > -}
> > -
> > -/* Upper 16 bits are ignored and will be dropped. */
> > -static LLVMValueRef ac_llvm_pack_two_int32_as_int16(struct ac_llvm_context
> > *ctx,
> > - LLVMValueRef val[2])
> > -{
> > - LLVMValueRef v[2] = {
> > - LLVMBuildAnd(ctx->builder, val[0],
> > - LLVMConstInt(ctx->i32, 0xffff, 0), ""),
> > - val[1],
> > - };
> > - return ac_llvm_pack_two_int16(ctx, v);
> > -}
> > -
> > LLVMValueRef ac_build_cvt_pknorm_i16(struct ac_llvm_context *ctx,
> > LLVMValueRef args[2])
> > {
> > - if (HAVE_LLVM >= 0x0600) {
> > - LLVMValueRef res =
> > - ac_build_intrinsic(ctx, "llvm.amdgcn.cvt.pknorm.i16",
> > - ctx->v2i16, args, 2,
> > - AC_FUNC_ATTR_READNONE);
> > - return LLVMBuildBitCast(ctx->builder, res, ctx->i32, "");
> > - }
> > -
> > - LLVMValueRef val[2];
> > -
> > - for (int chan = 0; chan < 2; chan++) {
> > - /* Clamp between [-1, 1]. */
> > - val[chan] = ac_build_fmin(ctx, args[chan], ctx->f32_1);
> > - val[chan] = ac_build_fmax(ctx, val[chan],
> > LLVMConstReal(ctx->f32, -1));
> > - /* Convert to a signed integer in [-32767, 32767]. */
> > - val[chan] = LLVMBuildFMul(ctx->builder, val[chan],
> > - LLVMConstReal(ctx->f32, 32767), "");
> > - /* If positive, add 0.5, else add -0.5. */
> > - val[chan] = LLVMBuildFAdd(ctx->builder, val[chan],
> > - LLVMBuildSelect(ctx->builder,
> > - LLVMBuildFCmp(ctx->builder, LLVMRealOGE,
> > - val[chan], ctx->f32_0,
> > ""),
> > - LLVMConstReal(ctx->f32, 0.5),
> > - LLVMConstReal(ctx->f32, -0.5), ""), "");
> > - val[chan] = LLVMBuildFPToSI(ctx->builder, val[chan], ctx->i32,
> > "");
> > - }
> > - return ac_llvm_pack_two_int32_as_int16(ctx, val);
> > + LLVMValueRef res =
> > + ac_build_intrinsic(ctx, "llvm.amdgcn.cvt.pknorm.i16",
> > + ctx->v2i16, args, 2,
> > + AC_FUNC_ATTR_READNONE);
> > + return LLVMBuildBitCast(ctx->builder, res, ctx->i32, "");
> > }
> >
> > LLVMValueRef ac_build_cvt_pknorm_u16(struct ac_llvm_context *ctx,
> > LLVMValueRef args[2])
> > {
> > - if (HAVE_LLVM >= 0x0600) {
> > - LLVMValueRef res =
> > - ac_build_intrinsic(ctx, "llvm.amdgcn.cvt.pknorm.u16",
> > - ctx->v2i16, args, 2,
> > - AC_FUNC_ATTR_READNONE);
> > - return LLVMBuildBitCast(ctx->builder, res, ctx->i32, "");
> > - }
> > -
> > - LLVMValueRef val[2];
> > -
> > - for (int chan = 0; chan < 2; chan++) {
> > - val[chan] = ac_build_clamp(ctx, args[chan]);
> > - val[chan] = LLVMBuildFMul(ctx->builder, val[chan],
> > - LLVMConstReal(ctx->f32, 65535), "");
> > - val[chan] = LLVMBuildFAdd(ctx->builder, val[chan],
> > - LLVMConstReal(ctx->f32, 0.5), "");
> > - val[chan] = LLVMBuildFPToUI(ctx->builder, val[chan],
> > - ctx->i32, "");
> > - }
> > - return ac_llvm_pack_two_int32_as_int16(ctx, val);
> > + LLVMValueRef res =
> > + ac_build_intrinsic(ctx, "llvm.amdgcn.cvt.pknorm.u16",
> > + ctx->v2i16, args, 2,
> > + AC_FUNC_ATTR_READNONE);
> > + return LLVMBuildBitCast(ctx->builder, res, ctx->i32, "");
> > }
> >
> > /* The 8-bit and 10-bit clamping is for HW workarounds. */
> > LLVMValueRef ac_build_cvt_pk_i16(struct ac_llvm_context *ctx,
> > LLVMValueRef args[2], unsigned bits, bool hi)
> > {
> > assert(bits == 8 || bits == 10 || bits == 16);
> >
> > LLVMValueRef max_rgb = LLVMConstInt(ctx->i32,
> > bits == 8 ? 127 : bits == 10 ? 511 : 32767, 0);
> > LLVMValueRef min_rgb = LLVMConstInt(ctx->i32,
> > bits == 8 ? -128 : bits == 10 ? -512 : -32768, 0);
> > LLVMValueRef max_alpha =
> > bits != 10 ? max_rgb : ctx->i32_1;
> > LLVMValueRef min_alpha =
> > bits != 10 ? min_rgb : LLVMConstInt(ctx->i32, -2, 0);
> > - bool has_intrinsic = HAVE_LLVM >= 0x0600;
> >
> > /* Clamp. */
> > - if (!has_intrinsic || bits != 16) {
> > + if (bits != 16) {
> > for (int i = 0; i < 2; i++) {
> > bool alpha = hi && i == 1;
> > args[i] = ac_build_imin(ctx, args[i],
> > alpha ? max_alpha : max_rgb);
> > args[i] = ac_build_imax(ctx, args[i],
> > alpha ? min_alpha : min_rgb);
> > }
> > }
> >
> > - if (has_intrinsic) {
> > - LLVMValueRef res =
> > - ac_build_intrinsic(ctx, "llvm.amdgcn.cvt.pk.i16",
> > - ctx->v2i16, args, 2,
> > - AC_FUNC_ATTR_READNONE);
> > - return LLVMBuildBitCast(ctx->builder, res, ctx->i32, "");
> > - }
> > -
> > - return ac_llvm_pack_two_int32_as_int16(ctx, args);
> > + LLVMValueRef res =
> > + ac_build_intrinsic(ctx, "llvm.amdgcn.cvt.pk.i16",
> > + ctx->v2i16, args, 2,
> > + AC_FUNC_ATTR_READNONE);
> > + return LLVMBuildBitCast(ctx->builder, res, ctx->i32, "");
> > }
> >
> > /* The 8-bit and 10-bit clamping is for HW workarounds. */
> > LLVMValueRef ac_build_cvt_pk_u16(struct ac_llvm_context *ctx,
> > LLVMValueRef args[2], unsigned bits, bool hi)
> > {
> > assert(bits == 8 || bits == 10 || bits == 16);
> >
> > LLVMValueRef max_rgb = LLVMConstInt(ctx->i32,
> > bits == 8 ? 255 : bits == 10 ? 1023 : 65535, 0);
> > LLVMValueRef max_alpha =
> > bits != 10 ? max_rgb : LLVMConstInt(ctx->i32, 3, 0);
> > - bool has_intrinsic = HAVE_LLVM >= 0x0600;
> >
> > /* Clamp. */
> > - if (!has_intrinsic || bits != 16) {
> > + if (bits != 16) {
> > for (int i = 0; i < 2; i++) {
> > bool alpha = hi && i == 1;
> > args[i] = ac_build_umin(ctx, args[i],
> > alpha ? max_alpha : max_rgb);
> > }
> > }
> >
> > - if (has_intrinsic) {
> > - LLVMValueRef res =
> > - ac_build_intrinsic(ctx, "llvm.amdgcn.cvt.pk.u16",
> > - ctx->v2i16, args, 2,
> > - AC_FUNC_ATTR_READNONE);
> > - return LLVMBuildBitCast(ctx->builder, res, ctx->i32, "");
> > - }
> > -
> > - return ac_llvm_pack_two_int16(ctx, args);
> > + LLVMValueRef res =
> > + ac_build_intrinsic(ctx, "llvm.amdgcn.cvt.pk.u16",
> > + ctx->v2i16, args, 2,
> > + AC_FUNC_ATTR_READNONE);
> > + return LLVMBuildBitCast(ctx->builder, res, ctx->i32, "");
> > }
> >
> > LLVMValueRef ac_build_wqm_vote(struct ac_llvm_context *ctx, LLVMValueRef
> > i1)
> > {
> > - assert(HAVE_LLVM >= 0x0600);
> > return ac_build_intrinsic(ctx, "llvm.amdgcn.wqm.vote", ctx->i1,
> > &i1, 1, AC_FUNC_ATTR_READNONE);
> > }
> >
> > void ac_build_kill_if_false(struct ac_llvm_context *ctx, LLVMValueRef i1)
> > {
> > - if (HAVE_LLVM >= 0x0600) {
> > - ac_build_intrinsic(ctx, "llvm.amdgcn.kill", ctx->voidt,
> > - &i1, 1, 0);
> > - return;
> > - }
> > -
> > - LLVMValueRef value = LLVMBuildSelect(ctx->builder, i1,
> > - LLVMConstReal(ctx->f32, 1),
> > - LLVMConstReal(ctx->f32, -1), "");
> > - ac_build_intrinsic(ctx, "llvm.AMDGPU.kill", ctx->voidt,
> > - &value, 1, AC_FUNC_ATTR_LEGACY);
> > + ac_build_intrinsic(ctx, "llvm.amdgcn.kill", ctx->voidt,
> > + &i1, 1, 0);
> > }
> >
> > LLVMValueRef ac_build_bfe(struct ac_llvm_context *ctx, LLVMValueRef input,
> > LLVMValueRef offset, LLVMValueRef width,
> > bool is_signed)
> > {
> > LLVMValueRef args[] = {
> > input,
> > offset,
> > width,
> > @@ -3196,21 +3117,21 @@ ac_build_reduce(struct ac_llvm_context *ctx,
> > LLVMValueRef src, nir_op op, unsign
> > result = ac_build_alu_op(ctx, result, swap, op);
> > return ac_build_wwm(ctx, result);
> > }
> > }
> >
> > LLVMValueRef
> > ac_build_quad_swizzle(struct ac_llvm_context *ctx, LLVMValueRef src,
> > unsigned lane0, unsigned lane1, unsigned lane2, unsigned lane3)
> > {
> > unsigned mask = dpp_quad_perm(lane0, lane1, lane2, lane3);
> > - if (ctx->chip_class >= VI && HAVE_LLVM >= 0x0600) {
> > + if (ctx->chip_class >= VI) {
> > return ac_build_dpp(ctx, src, src, mask, 0xf, 0xf, false);
> > } else {
> > return ac_build_ds_swizzle(ctx, src, (1 << 15) | mask);
> > }
> > }
> >
> > LLVMValueRef
> > ac_build_shuffle(struct ac_llvm_context *ctx, LLVMValueRef src,
> > LLVMValueRef index)
> > {
> > index = LLVMBuildMul(ctx->builder, index, LLVMConstInt(ctx->i32, 4, 0),
> > "");
> > diff --git a/src/amd/common/ac_llvm_helper.cpp
> > b/src/amd/common/ac_llvm_helper.cpp
> > index 1a2aee3..f715c8f 100644
> > --- a/src/amd/common/ac_llvm_helper.cpp
> > +++ b/src/amd/common/ac_llvm_helper.cpp
> > @@ -69,21 +69,17 @@ LLVMBuilderRef ac_create_builder(LLVMContextRef ctx,
> > llvm::FastMathFlags flags;
> >
> > switch (float_mode) {
> > case AC_FLOAT_MODE_DEFAULT:
> > break;
> > case AC_FLOAT_MODE_NO_SIGNED_ZEROS_FP_MATH:
> > flags.setNoSignedZeros();
> > llvm::unwrap(builder)->setFastMathFlags(flags);
> > break;
> > case AC_FLOAT_MODE_UNSAFE_FP_MATH:
> > -#if HAVE_LLVM >= 0x0600
> > flags.setFast();
> > -#else
> > - flags.setUnsafeAlgebra();
> > -#endif
> > llvm::unwrap(builder)->setFastMathFlags(flags);
> > break;
> > }
> >
> > return builder;
> > }
> > diff --git a/src/gallium/drivers/radeonsi/si_shader.c
> > b/src/gallium/drivers/radeonsi/si_shader.c
> > index 32db089..7d2fc44 100644
> > --- a/src/gallium/drivers/radeonsi/si_shader.c
> > +++ b/src/gallium/drivers/radeonsi/si_shader.c
> > @@ -2443,44 +2443,20 @@ static LLVMValueRef fetch_constant(
> > idx = reg->Register.Index * 4 + swizzle;
> > if (reg->Register.Indirect) {
> > addr = si_get_indirect_index(ctx, ireg, 16, idx * 4);
> > } else {
> > addr = LLVMConstInt(ctx->i32, idx * 4, 0);
> > }
> >
> > /* Fast path when user data SGPRs point to constant buffer 0 directly.
> > */
> > if (sel->info.const_buffers_declared == 1 &&
> > sel->info.shader_buffers_declared == 0) {
> > -
> > - /* This enables use of s_load_dword and flat_load_dword for
> > const buffer 0
> > - * loads, and up to x4 load opcode merging. However, it leads
> > to horrible
> > - * code reducing SIMD wave occupancy from 8 to 2 in many cases.
> > - *
> > - * Using s_buffer_load_dword (x1) seems to be the best option
> > right now.
> > - *
> > - * LLVM 5.0 on SI doesn't insert a required s_nop between SALU
> > setting
> > - * a descriptor and s_buffer_load_dword using it, so we can't
> > expand
> > - * the pointer into a full descriptor like below. We have to use
> > - * s_load_dword instead. The only case when LLVM 5.0 would
> > select
> > - * s_buffer_load_dword (that we have to prevent) is when we use
> > use
> > - * a literal offset where we don't need bounds checking.
> > - */
> > - if (ctx->screen->info.chip_class == SI && HAVE_LLVM < 0x0600 &&
> > - !reg->Register.Indirect) {
> > - LLVMValueRef ptr =
> > - LLVMGetParam(ctx->main_fn,
> > ctx->param_const_and_shader_buffers);
> > -
> > - addr = LLVMBuildLShr(ctx->ac.builder, addr,
> > LLVMConstInt(ctx->i32, 2, 0), "");
> > - LLVMValueRef result = ac_build_load_invariant(&ctx->ac,
> > ptr, addr);
> > - return bitcast(bld_base, type, result);
> > - }
> > -
> > LLVMValueRef desc = load_const_buffer_desc_fast_path(ctx);
> > LLVMValueRef result = buffer_load_const(ctx, desc, addr);
> > return bitcast(bld_base, type, result);
> > }
> >
> > assert(reg->Register.Dimension);
> > buf = reg->Dimension.Index;
> >
> > if (reg->Dimension.Indirect) {
> > LLVMValueRef ptr = LLVMGetParam(ctx->main_fn,
> > ctx->param_const_and_shader_buffers);
> > diff --git a/src/gallium/drivers/radeonsi/si_shader_nir.c
> > b/src/gallium/drivers/radeonsi/si_shader_nir.c
> > index 6eb114a..914b901 100644
> > --- a/src/gallium/drivers/radeonsi/si_shader_nir.c
> > +++ b/src/gallium/drivers/radeonsi/si_shader_nir.c
> > @@ -741,27 +741,20 @@ void si_nir_scan_shader(const struct nir_shader *nir,
> > }
> > }
> >
> > /**
> > * Perform "lowering" operations on the NIR that are run once when the
> > shader
> > * selector is created.
> > */
> > void
> > si_lower_nir(struct si_shader_selector* sel)
> > {
> > - /* Disable const buffer fast path for old LLVM versions */
> > - if (sel->screen->info.chip_class == SI && HAVE_LLVM < 0x0600 &&
> > - sel->info.const_buffers_declared == 1 &&
> > - sel->info.shader_buffers_declared == 0) {
> > - sel->info.const_buffers_declared |= 0x2;
> > - }
> > -
> > /* Adjust the driver location of inputs and outputs. The state tracker
> > * interprets them as slots, while the ac/nir backend interprets them
> > * as individual components.
> > */
> > nir_foreach_variable(variable, &sel->nir->inputs)
> > variable->data.driver_location *= 4;
> >
> > nir_foreach_variable(variable, &sel->nir->outputs) {
> > variable->data.driver_location *= 4;
> >
> > diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
> > b/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
> > index 43922dc..694f934 100644
> > --- a/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
> > +++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
> > @@ -49,25 +49,23 @@ static void kill_if_fetch_args(struct
> > lp_build_tgsi_context *bld_base,
> > emit_data->arg_count = 1;
> > emit_data->args[0] = conds[0];
> > }
> >
> > void si_llvm_emit_kill(struct ac_shader_abi *abi, LLVMValueRef visible)
> > {
> > struct si_shader_context *ctx = si_shader_context_from_abi(abi);
> > LLVMBuilderRef builder = ctx->ac.builder;
> >
> > if (ctx->shader->selector->force_correct_derivs_after_kill) {
> > - /* LLVM 6.0 can kill immediately while maintaining WQM. */
> > - if (HAVE_LLVM >= 0x0600) {
> > - ac_build_kill_if_false(&ctx->ac,
> > - ac_build_wqm_vote(&ctx->ac,
> > visible));
> > - }
> > + /* Kill immediately while maintaining WQM. */
> > + ac_build_kill_if_false(&ctx->ac,
> > + ac_build_wqm_vote(&ctx->ac, visible));
> >
> > LLVMValueRef mask = LLVMBuildLoad(builder, ctx->postponed_kill,
> > "");
> > mask = LLVMBuildAnd(builder, mask, visible, "");
> > LLVMBuildStore(builder, mask, ctx->postponed_kill);
> > return;
> > }
> >
> > ac_build_kill_if_false(&ctx->ac, visible);
> > }
> >
> >
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
--
Jan Vesely <jan.ves...@rutgers.edu>
signature.asc
Description: This is a digitally signed message part
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
