On Fri, Nov 2, 2018 at 5:21 AM Samuel Pitoiset <samuel.pitoi...@gmail.com>
wrote:
> User are encouraged to switch to LLVM 7.0 released in September 2018.
>
> Signed-off-by: Samuel Pitoiset <samuel.pitoi...@gmail.com>
> ---
> configure.ac | 4 +-
> meson.build | 2 +-
> src/amd/common/ac_llvm_build.c | 270 +-----------------
> src/amd/common/ac_llvm_build.h | 8 +-
> src/amd/common/ac_llvm_helper.cpp | 7 -
> src/amd/common/ac_llvm_util.c | 19 +-
> src/amd/common/ac_nir_to_llvm.c | 24 +-
> src/amd/vulkan/radv_cmd_buffer.c | 15 +-
> src/amd/vulkan/radv_device.c | 4 +-
> src/amd/vulkan/radv_extensions.py | 2 +-
> src/amd/vulkan/radv_nir_to_llvm.c | 13 +-
> src/amd/vulkan/radv_private.h | 2 +-
> src/gallium/drivers/radeonsi/si_descriptors.c | 49 +---
> src/gallium/drivers/radeonsi/si_get.c | 9 -
> src/gallium/drivers/radeonsi/si_shader.c | 92 +-----
> src/gallium/drivers/radeonsi/si_shader.h | 27 --
> .../drivers/radeonsi/si_shader_tgsi_alu.c | 47 ++-
> .../drivers/radeonsi/si_state_shaders.c | 7 +-
> 18 files changed, 86 insertions(+), 515 deletions(-)
>
> diff --git a/configure.ac b/configure.ac
> index d782f56205..fe63044293 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -107,8 +107,8 @@ dnl LLVM versions
> LLVM_REQUIRED_GALLIUM=3.3.0
> LLVM_REQUIRED_OPENCL=3.9.0
> LLVM_REQUIRED_R600=3.9.0
> -LLVM_REQUIRED_RADEONSI=6.0.0
> -LLVM_REQUIRED_RADV=6.0.0
> +LLVM_REQUIRED_RADEONSI=7.0.0
> +LLVM_REQUIRED_RADV=7.0.0
> LLVM_REQUIRED_SWR=6.0.0
>
> dnl Check for progs
> diff --git a/meson.build b/meson.build
> index 18667988ba..22a9c7bd9a 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -1176,7 +1176,7 @@ if with_gallium_opencl
> endif
>
> if with_amd_vk or with_gallium_radeonsi
> - _llvm_version = '>= 6.0.0'
> + _llvm_version = '>= 7.0.0'
> elif with_gallium_swr
> _llvm_version = '>= 6.0.0'
> elif with_gallium_opencl or with_gallium_r600
> diff --git a/src/amd/common/ac_llvm_build.c
> b/src/amd/common/ac_llvm_build.c
> index 1392ec0f23..108e37a3e7 100644
> --- a/src/amd/common/ac_llvm_build.c
> +++ b/src/amd/common/ac_llvm_build.c
> @@ -75,7 +75,7 @@ ac_llvm_context_init(struct ac_llvm_context *ctx,
> ctx->i16 = LLVMIntTypeInContext(ctx->context, 16);
> ctx->i32 = LLVMIntTypeInContext(ctx->context, 32);
> ctx->i64 = LLVMIntTypeInContext(ctx->context, 64);
> - ctx->intptr = HAVE_32BIT_POINTERS ? ctx->i32 : ctx->i64;
> + ctx->intptr = ctx->i32;
> ctx->f16 = LLVMHalfTypeInContext(ctx->context);
> ctx->f32 = LLVMFloatTypeInContext(ctx->context);
> ctx->f64 = LLVMDoubleTypeInContext(ctx->context);
> @@ -1342,99 +1342,28 @@ ac_build_ddxy(struct ac_llvm_context *ctx,
> int idx,
> LLVMValueRef val)
> {
> + unsigned tl_lanes[4], trbl_lanes[4];
> LLVMValueRef tl, trbl, args[2];
> LLVMValueRef result;
>
> - if (HAVE_LLVM >= 0x0700) {
> - unsigned tl_lanes[4], trbl_lanes[4];
> -
> - for (unsigned i = 0; i < 4; ++i) {
> - tl_lanes[i] = i & mask;
> - trbl_lanes[i] = (i & mask) + idx;
> - }
> -
> - tl = ac_build_quad_swizzle(ctx, val,
> - tl_lanes[0], tl_lanes[1],
> - tl_lanes[2], tl_lanes[3]);
> - trbl = ac_build_quad_swizzle(ctx, val,
> - trbl_lanes[0], trbl_lanes[1],
> - trbl_lanes[2], trbl_lanes[3]);
> - } else if (ctx->chip_class >= VI) {
> - LLVMValueRef thread_id, tl_tid, trbl_tid;
> - thread_id = ac_get_thread_id(ctx);
> -
> - tl_tid = LLVMBuildAnd(ctx->builder, thread_id,
> - LLVMConstInt(ctx->i32, mask, false),
> "");
> -
> - trbl_tid = LLVMBuildAdd(ctx->builder, tl_tid,
> - LLVMConstInt(ctx->i32, idx,
> false), "");
> -
> - args[0] = LLVMBuildMul(ctx->builder, tl_tid,
> - LLVMConstInt(ctx->i32, 4, false),
> "");
> - args[1] = val;
> - tl = ac_build_intrinsic(ctx,
> - "llvm.amdgcn.ds.bpermute",
> ctx->i32,
> - args, 2,
> - AC_FUNC_ATTR_READNONE |
> - AC_FUNC_ATTR_CONVERGENT);
> -
> - args[0] = LLVMBuildMul(ctx->builder, trbl_tid,
> - LLVMConstInt(ctx->i32, 4, false),
> "");
> - trbl = ac_build_intrinsic(ctx,
> - "llvm.amdgcn.ds.bpermute",
> ctx->i32,
> - args, 2,
> - AC_FUNC_ATTR_READNONE |
> - AC_FUNC_ATTR_CONVERGENT);
> - } else {
> - uint32_t masks[2] = {};
> -
> - switch (mask) {
> - case AC_TID_MASK_TOP_LEFT:
> - masks[0] = 0x8000;
> - if (idx == 1)
> - masks[1] = 0x8055;
> - else
> - masks[1] = 0x80aa;
> -
> - break;
> - case AC_TID_MASK_TOP:
> - masks[0] = 0x8044;
> - masks[1] = 0x80ee;
> - break;
> - case AC_TID_MASK_LEFT:
> - masks[0] = 0x80a0;
> - masks[1] = 0x80f5;
> - break;
> - default:
> - assert(0);
> - }
> -
> - args[0] = val;
> - args[1] = LLVMConstInt(ctx->i32, masks[0], false);
> -
> - tl = ac_build_intrinsic(ctx,
> - "llvm.amdgcn.ds.swizzle", ctx->i32,
> - args, 2,
> - AC_FUNC_ATTR_READNONE |
> - AC_FUNC_ATTR_CONVERGENT);
> -
> - args[1] = LLVMConstInt(ctx->i32, masks[1], false);
> - trbl = ac_build_intrinsic(ctx,
> - "llvm.amdgcn.ds.swizzle", ctx->i32,
> - args, 2,
> - AC_FUNC_ATTR_READNONE |
> - AC_FUNC_ATTR_CONVERGENT);
> + for (unsigned i = 0; i < 4; ++i) {
> + tl_lanes[i] = i & mask;
> + trbl_lanes[i] = (i & mask) + idx;
> }
>
> + tl = ac_build_quad_swizzle(ctx, val,
> + tl_lanes[0], tl_lanes[1],
> + tl_lanes[2], tl_lanes[3]);
> + trbl = ac_build_quad_swizzle(ctx, val,
> + trbl_lanes[0], trbl_lanes[1],
> + trbl_lanes[2], trbl_lanes[3]);
> +
> tl = LLVMBuildBitCast(ctx->builder, tl, ctx->f32, "");
> trbl = LLVMBuildBitCast(ctx->builder, trbl, ctx->f32, "");
> result = LLVMBuildFSub(ctx->builder, trbl, tl, "");
>
> - if (HAVE_LLVM >= 0x0700) {
> - result = ac_build_intrinsic(ctx,
> - "llvm.amdgcn.wqm.f32", ctx->f32,
> - &result, 1, 0);
> - }
> + result = ac_build_intrinsic(ctx, "llvm.amdgcn.wqm.f32", ctx->f32,
> + &result, 1, 0);
>
> return result;
> }
> @@ -1679,171 +1608,6 @@ static const char *get_atomic_name(enum
> ac_atomic_op op)
> unreachable("bad atomic op");
> }
>
> -/* LLVM 6 and older */
> -static LLVMValueRef ac_build_image_opcode_llvm6(struct ac_llvm_context
> *ctx,
> - struct ac_image_args *a)
> -{
> - LLVMValueRef args[16];
> - LLVMTypeRef retty = ctx->v4f32;
> - const char *name = NULL;
> - const char *atomic_subop = "";
> - char intr_name[128], coords_type[64];
> -
> - bool sample = a->opcode == ac_image_sample ||
> - a->opcode == ac_image_gather4 ||
> - a->opcode == ac_image_get_lod;
> - bool atomic = a->opcode == ac_image_atomic ||
> - a->opcode == ac_image_atomic_cmpswap;
> - bool da = a->dim == ac_image_cube ||
> - a->dim == ac_image_1darray ||
> - a->dim == ac_image_2darray ||
> - a->dim == ac_image_2darraymsaa;
> - if (a->opcode == ac_image_get_lod)
> - da = false;
> -
> - unsigned num_coords =
> - a->opcode != ac_image_get_resinfo ? ac_num_coords(a->dim)
> : 0;
> - LLVMValueRef addr;
> - unsigned num_addr = 0;
> -
> - if (a->opcode == ac_image_get_lod) {
> - switch (a->dim) {
> - case ac_image_1darray:
> - num_coords = 1;
> - break;
> - case ac_image_2darray:
> - case ac_image_cube:
> - num_coords = 2;
> - break;
> - default:
> - break;
> - }
> - }
> -
> - if (a->offset)
> - args[num_addr++] = ac_to_integer(ctx, a->offset);
> - if (a->bias)
> - args[num_addr++] = ac_to_integer(ctx, a->bias);
> - if (a->compare)
> - args[num_addr++] = ac_to_integer(ctx, a->compare);
> - if (a->derivs[0]) {
> - unsigned num_derivs = ac_num_derivs(a->dim);
> - for (unsigned i = 0; i < num_derivs; ++i)
> - args[num_addr++] = ac_to_integer(ctx,
> a->derivs[i]);
> - }
> - for (unsigned i = 0; i < num_coords; ++i)
> - args[num_addr++] = ac_to_integer(ctx, a->coords[i]);
> - if (a->lod)
> - args[num_addr++] = ac_to_integer(ctx, a->lod);
> -
> - unsigned pad_goal = util_next_power_of_two(num_addr);
> - while (num_addr < pad_goal)
> - args[num_addr++] = LLVMGetUndef(ctx->i32);
> -
> - addr = ac_build_gather_values(ctx, args, num_addr);
> -
> - unsigned num_args = 0;
> - if (atomic || a->opcode == ac_image_store || a->opcode ==
> ac_image_store_mip) {
> - args[num_args++] = a->data[0];
> - if (a->opcode == ac_image_atomic_cmpswap)
> - args[num_args++] = a->data[1];
> - }
> -
> - unsigned coords_arg = num_args;
> - if (sample)
> - args[num_args++] = ac_to_float(ctx, addr);
> - else
> - args[num_args++] = ac_to_integer(ctx, addr);
> -
> - args[num_args++] = a->resource;
> - if (sample)
> - args[num_args++] = a->sampler;
> - if (!atomic) {
> - args[num_args++] = LLVMConstInt(ctx->i32, a->dmask, 0);
> - if (sample)
> - args[num_args++] = LLVMConstInt(ctx->i1, a->unorm,
> 0);
> - args[num_args++] = a->cache_policy & ac_glc ? ctx->i1true
> : ctx->i1false;
> - args[num_args++] = a->cache_policy & ac_slc ? ctx->i1true
> : ctx->i1false;
> - args[num_args++] = ctx->i1false; /* lwe */
> - args[num_args++] = LLVMConstInt(ctx->i1, da, 0);
> - } else {
> - args[num_args++] = ctx->i1false; /* r128 */
> - args[num_args++] = LLVMConstInt(ctx->i1, da, 0);
> - args[num_args++] = a->cache_policy & ac_slc ? ctx->i1true
> : ctx->i1false;
> - }
> -
> - switch (a->opcode) {
> - case ac_image_sample:
> - name = "llvm.amdgcn.image.sample";
> - break;
> - case ac_image_gather4:
> - name = "llvm.amdgcn.image.gather4";
> - break;
> - case ac_image_load:
> - name = "llvm.amdgcn.image.load";
> - break;
> - case ac_image_load_mip:
> - name = "llvm.amdgcn.image.load.mip";
> - break;
> - case ac_image_store:
> - name = "llvm.amdgcn.image.store";
> - retty = ctx->voidt;
> - break;
> - case ac_image_store_mip:
> - name = "llvm.amdgcn.image.store.mip";
> - retty = ctx->voidt;
> - break;
> - case ac_image_atomic:
> - case ac_image_atomic_cmpswap:
> - name = "llvm.amdgcn.image.atomic.";
> - retty = ctx->i32;
> - if (a->opcode == ac_image_atomic_cmpswap) {
> - atomic_subop = "cmpswap";
> - } else {
> - atomic_subop = get_atomic_name(a->atomic);
> - }
> - break;
> - case ac_image_get_lod:
> - name = "llvm.amdgcn.image.getlod";
> - break;
> - case ac_image_get_resinfo:
> - name = "llvm.amdgcn.image.getresinfo";
> - break;
> - default:
> - unreachable("invalid image opcode");
> - }
> -
> - ac_build_type_name_for_intr(LLVMTypeOf(args[coords_arg]),
> coords_type,
> - sizeof(coords_type));
> -
> - if (atomic) {
> - snprintf(intr_name, sizeof(intr_name),
> "llvm.amdgcn.image.atomic.%s.%s",
> - atomic_subop, coords_type);
> - } else {
> - bool lod_suffix =
> - a->lod && (a->opcode == ac_image_sample ||
> a->opcode == ac_image_gather4);
> -
> - snprintf(intr_name, sizeof(intr_name),
> "%s%s%s%s.v4f32.%s.v8i32",
> - name,
> - a->compare ? ".c" : "",
> - a->bias ? ".b" :
> - lod_suffix ? ".l" :
> - a->derivs[0] ? ".d" :
> - a->level_zero ? ".lz" : "",
> - a->offset ? ".o" : "",
> - coords_type);
> - }
> -
> - LLVMValueRef result =
> - ac_build_intrinsic(ctx, intr_name, retty, args, num_args,
> - a->attributes);
> - if (!sample && retty == ctx->v4f32) {
> - result = LLVMBuildBitCast(ctx->builder, result,
> - ctx->v4i32, "");
> - }
> - return result;
> -}
> -
> LLVMValueRef ac_build_image_opcode(struct ac_llvm_context *ctx,
> struct ac_image_args *a)
> {
> @@ -1868,9 +1632,6 @@ LLVMValueRef ac_build_image_opcode(struct
> ac_llvm_context *ctx,
> (a->level_zero ? 1 : 0) +
> (a->derivs[0] ? 1 : 0) <= 1);
>
> - if (HAVE_LLVM < 0x0700)
> - return ac_build_image_opcode_llvm6(ctx, a);
> -
> if (a->opcode == ac_image_get_lod) {
> switch (dim) {
> case ac_image_1darray:
> @@ -2659,9 +2420,6 @@ LLVMTypeRef ac_array_in_const_addr_space(LLVMTypeRef
> elem_type)
>
> LLVMTypeRef ac_array_in_const32_addr_space(LLVMTypeRef elem_type)
> {
> - if (!HAVE_32BIT_POINTERS)
> - return ac_array_in_const_addr_space(elem_type);
> -
> return LLVMPointerType(LLVMArrayType(elem_type, 0),
> AC_ADDR_SPACE_CONST_32BIT);
> }
> diff --git a/src/amd/common/ac_llvm_build.h
> b/src/amd/common/ac_llvm_build.h
> index 1275e4fb69..26399b7832 100644
> --- a/src/amd/common/ac_llvm_build.h
> +++ b/src/amd/common/ac_llvm_build.h
> @@ -34,14 +34,12 @@
> extern "C" {
> #endif
>
> -#define HAVE_32BIT_POINTERS (HAVE_LLVM >= 0x0700)
> -
> enum {
> - AC_ADDR_SPACE_FLAT = HAVE_LLVM >= 0x0700 ? 0 : 4, /* Slower than
> global. */
> + AC_ADDR_SPACE_FLAT = 0, /* Slower than global. */
> AC_ADDR_SPACE_GLOBAL = 1,
> - AC_ADDR_SPACE_GDS = HAVE_LLVM >= 0x0700 ? 2 : 5,
> + AC_ADDR_SPACE_GDS = 2,
> AC_ADDR_SPACE_LDS = 3,
> - AC_ADDR_SPACE_CONST = HAVE_LLVM >= 0x0700 ? 4 : 2, /* Global
> allowing SMEM. */
> + AC_ADDR_SPACE_CONST = 4, /* Global allowing SMEM. */
> AC_ADDR_SPACE_CONST_32BIT = 6, /* same as CONST, but the pointer
> type has 32 bits */
> };
>
> diff --git a/src/amd/common/ac_llvm_helper.cpp
> b/src/amd/common/ac_llvm_helper.cpp
> index e022e12c7f..dcfb800854 100644
> --- a/src/amd/common/ac_llvm_helper.cpp
> +++ b/src/amd/common/ac_llvm_helper.cpp
> @@ -39,9 +39,6 @@
> #include <llvm/Transforms/IPO.h>
>
> #include <llvm/IR/LegacyPassManager.h>
> -#if HAVE_LLVM < 0x0700
> -#include "llvm/Support/raw_ostream.h"
> -#endif
>
> void ac_add_attr_dereferenceable(LLVMValueRef val, uint64_t bytes)
> {
> @@ -132,9 +129,7 @@ struct ac_compiler_passes
> *ac_create_llvm_passes(LLVMTargetMachineRef tm)
> llvm::TargetMachine *TM =
> reinterpret_cast<llvm::TargetMachine*>(tm);
>
> if (TM->addPassesToEmitFile(p->passmgr, p->ostream,
> -#if HAVE_LLVM >= 0x0700
> nullptr,
> -#endif
> llvm::TargetMachine::CGFT_ObjectFile))
> {
> fprintf(stderr, "amd: TargetMachine can't emit a file of
> this type!\n");
> delete p;
> @@ -170,7 +165,5 @@ void ac_llvm_add_barrier_noop_pass(LLVMPassManagerRef
> passmgr)
>
> void ac_enable_global_isel(LLVMTargetMachineRef tm)
> {
> -#if HAVE_LLVM >= 0x0700
> reinterpret_cast<llvm::TargetMachine*>(tm)->setGlobalISel(true);
> -#endif
> }
> diff --git a/src/amd/common/ac_llvm_util.c b/src/amd/common/ac_llvm_util.c
> index 69d9f7b9f3..0c8565f95a 100644
> --- a/src/amd/common/ac_llvm_util.c
> +++ b/src/amd/common/ac_llvm_util.c
> @@ -30,9 +30,7 @@
> #include <llvm-c/Support.h>
> #include <llvm-c/Transforms/IPO.h>
> #include <llvm-c/Transforms/Scalar.h>
> -#if HAVE_LLVM >= 0x0700
> #include <llvm-c/Transforms/Utils.h>
> -#endif
> #include "c11/threads.h"
> #include "gallivm/lp_bld_misc.h"
> #include "util/u_math.h"
> @@ -132,9 +130,9 @@ const char *ac_get_llvm_processor_name(enum
> radeon_family family)
> case CHIP_RAVEN:
> return "gfx902";
> case CHIP_VEGA12:
> - return HAVE_LLVM >= 0x0700 ? "gfx904" : "gfx902";
> + return "gfx904";
> case CHIP_VEGA20:
> - return HAVE_LLVM >= 0x0700 ? "gfx906" : "gfx902";
> + return "gfx906";
> case CHIP_RAVEN2:
> return "gfx902"; /* TODO: use gfx909 when it's available */
> default:
> @@ -323,12 +321,10 @@ ac_init_llvm_compiler(struct ac_llvm_compiler
> *compiler,
> goto fail;
> }
>
> - if (okay_to_leak_target_library_info || (HAVE_LLVM >= 0x0700)) {
> - compiler->target_library_info =
> - ac_create_target_library_info(triple);
> - if (!compiler->target_library_info)
> - goto fail;
> - }
> + compiler->target_library_info =
> + ac_create_target_library_info(triple);
> + if (!compiler->target_library_info)
> + goto fail;
>
You can also remove the "okay_to_leak_target_library_info" parameter. Other
than that:
Reviewed-by: Marek Olšák <marek.ol...@amd.com>
Marek
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
