fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-05 07:28:11 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

ac/llvm: fix overloading of intrinsic names

Browse source 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Backport-to: 25.1
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35489>
This commit is contained in:
Rhys Perry 2025-06-12 10:06:05 +01:00 • committed by Marge Bot
parent ac2e36b377
commit 3c2b3fbd03
2 changed files with 31 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

29
src/amd/llvm/ac_llvm_build.c
View file

@ -1591,7 +1591,7 @@ static const char *get_atomic_name(enum ac_atomic_op op)
LLVMValueRef ac_build_image_opcode(struct ac_llvm_context *ctx, struct ac_image_args *a)
{
const char *overload[3] = {"", "", ""};
const char *overload[5] = {"", "", "", "", ""};
unsigned num_overloads = 0;
LLVMValueRef args[18];
unsigned num_args = 0;
@ -1709,9 +1709,15 @@ LLVMValueRef ac_build_image_opcode(struct ac_llvm_context *ctx, struct ac_image_
overload[num_overloads++] = sample ? (a->a16 ? ".f16" : ".f32") : (a->a16 ? ".i16" : ".i32");
args[num_args++] = a->resource;
#if LLVM_VERSION_MAJOR >= 20
overload[num_overloads++] = ".v8i32";
#endif
if (sample) {
args[num_args++] = a->sampler;
args[num_args++] = LLVMConstInt(ctx->i1, a->unorm, false);
#if LLVM_VERSION_MAJOR >= 20
overload[num_overloads++] = ".v4i32";
#endif
}
args[num_args++] = a->tfe ? ctx->i32_1 : ctx->i32_0; /* texfailctrl */
@ -1798,11 +1804,11 @@ LLVMValueRef ac_build_image_opcode(struct ac_llvm_context *ctx, struct ac_image_
snprintf(intr_name, sizeof(intr_name),
"llvm.amdgcn.image.%s%s" /* base name */
"%s%s%s%s" /* sample/gather modifiers */
".%s.%s%s%s%s", /* dimension and type overloads */
".%s.%s%s%s%s%s%s", /* dimension and type overloads */
name, atomic_subop, a->compare ? ".c" : "",
a->bias ? ".b" : lod_suffix ? ".l" : a->derivs[0] ? ".d" : a->level_zero ? ".lz" : "",
a->min_lod ? ".cl" : "", a->offset ? ".o" : "", dimname,
data_type_str, overload[0], overload[1], overload[2]);
data_type_str, overload[0], overload[1], overload[2], overload[3], overload[4]);
LLVMTypeRef retty;
if (a->opcode == ac_image_store || a->opcode == ac_image_store_mip)
@ -2557,8 +2563,13 @@ static LLVMValueRef _ac_build_readlane(struct ac_llvm_context *ctx, LLVMValueRef
if (lane)
lane = LLVMBuildZExt(ctx->builder, lane, ctx->i32, "");
#if LLVM_VERSION_MAJOR >= 19
const char *intr_name = lane == NULL ? "llvm.amdgcn.readfirstlane.i32" : "llvm.amdgcn.readlane.i32";
#else
const char *intr_name = lane == NULL ? "llvm.amdgcn.readfirstlane" : "llvm.amdgcn.readlane";
#endif
result =
ac_build_intrinsic(ctx, lane == NULL ? "llvm.amdgcn.readfirstlane" : "llvm.amdgcn.readlane",
ac_build_intrinsic(ctx, intr_name,
ctx->i32, (LLVMValueRef[]){src, lane}, lane == NULL ? 1 : 2, 0);
return LLVMBuildTrunc(ctx->builder, result, type, "");
@ -2604,7 +2615,8 @@ LLVMValueRef ac_build_readlane(struct ac_llvm_context *ctx, LLVMValueRef src, LL
LLVMValueRef ac_build_writelane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef value,
LLVMValueRef lane)
{
return ac_build_intrinsic(ctx, "llvm.amdgcn.writelane", ctx->i32,
const char *intr_name = LLVM_VERSION_MAJOR >= 19 ? "llvm.amdgcn.writelane.i32" : "llvm.amdgcn.writelane";
return ac_build_intrinsic(ctx, intr_name, ctx->i32,
(LLVMValueRef[]){value, lane, src}, 3, 0);
}
@ -2745,8 +2757,13 @@ static LLVMValueRef _ac_build_permlane16(struct ac_llvm_context *ctx, LLVMValueR
bound_ctrl ? ctx->i1true : ctx->i1false,
};
#if LLVM_VERSION_MAJOR >= 19
const char *intr_name = exchange_rows ? "llvm.amdgcn.permlanex16.i32" : "llvm.amdgcn.permlane16.i32";
#else
const char *intr_name = exchange_rows ? "llvm.amdgcn.permlanex16" : "llvm.amdgcn.permlane16";
#endif
result =
ac_build_intrinsic(ctx, exchange_rows ? "llvm.amdgcn.permlanex16" : "llvm.amdgcn.permlane16",
ac_build_intrinsic(ctx, intr_name,
ctx->i32, args, 6, 0);
return LLVMBuildTrunc(ctx->builder, result, type, "");

12
src/amd/llvm/ac_nir_to_llvm.c
View file

@ -2739,7 +2739,8 @@ static bool visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
LLVMTypeRef src_type = LLVMIntTypeInContext(ctx->ac.context, ctx->ac.wave_size);
src = LLVMBuildTrunc(ctx->ac.builder, src, src_type, "");
}
result = ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.inverse.ballot", ctx->ac.i1, &src, 1, 0);
const char *intr_name = ctx->ac.wave_size == 64 ? "llvm.amdgcn.inverse.ballot.i64" : "llvm.amdgcn.inverse.ballot.i32";
result = ac_build_intrinsic(&ctx->ac, intr_name, ctx->ac.i1, &src, 1, 0);
break;
}
case nir_intrinsic_read_invocation:
@ -2947,7 +2948,8 @@ static bool visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
src = LLVMBuildZExt(ctx->ac.builder, src, ctx->ac.i32, "");
result = ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.readlane", ctx->ac.i32,
const char *intr_name = LLVM_VERSION_MAJOR >= 19 ? "llvm.amdgcn.readlane.i32" : "llvm.amdgcn.readlane";
result = ac_build_intrinsic(&ctx->ac, intr_name, ctx->ac.i32,
(LLVMValueRef[]){src, index_val}, 2, 0);
result = LLVMBuildTrunc(ctx->ac.builder, result, type, "");
@ -3180,8 +3182,9 @@ static bool visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
result = LLVMBuildICmp(ctx->ac.builder, LLVMIntEQ, visit_first_invocation(ctx),
ac_get_thread_id(&ctx->ac), "");
break;
case nir_intrinsic_lane_permute_16_amd:
result = ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.permlane16", ctx->ac.i32,
case nir_intrinsic_lane_permute_16_amd: {
const char *intr_name = LLVM_VERSION_MAJOR >= 19 ? "llvm.amdgcn.permlane16.i32" : "llvm.amdgcn.permlane16";
result = ac_build_intrinsic(&ctx->ac, intr_name, ctx->ac.i32,
(LLVMValueRef[]){get_src(ctx, instr->src[0]),
get_src(ctx, instr->src[0]),
get_src(ctx, instr->src[1]),
@ -3189,6 +3192,7 @@ static bool visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
ctx->ac.i1false,
ctx->ac.i1false}, 6, 0);
break;
}
case nir_intrinsic_load_scalar_arg_amd:
case nir_intrinsic_load_vector_arg_amd: {
assert(nir_intrinsic_base(instr) < AC_MAX_ARGS);