Skip to content

Commit

Permalink
Auto merge of #115580 - eduardosm:stdarch-intrinsics, r=davidtwco,bjorn3
Browse files Browse the repository at this point in the history 
Update stdarch submodule and remove special handling in cranelift codegen for some AVX and SSE2 LLVM intrinsics

rust-lang/stdarch#1463 reimplemented some x86 intrinsics to avoid using some x86-specific LLVM intrinsics:

* Store unaligned (`_mm*_storeu_*`) use `<*mut _>::write_unaligned` instead of `llvm.x86.*.storeu.*`.
* Shift by immediate (`_mm*_s{ll,rl,ra}i_epi*`) use `if` (srl, sll) or `min` (sra) to simulate the behaviour when the RHS is out of range. RHS is constant, so the `if`/`min` will be optimized away.

This PR updates the stdarch submodule to pull these changes and removes special handling for those LLVM intrinsics from cranelift codegen. I left gcc codegen untouched because there are some autogenerated lists.
  • Loading branch information
@bors
bors committed Sep 6, 2023
2 parents e3abbd4 + 0b6ee86 commit b0d4553
Show file tree
Hide file tree
Showing 2 changed files with 1 addition and 249 deletions.
248 changes: 0 additions & 248 deletions compiler/rustc_codegen_cranelift/src/intrinsics/llvm_x86.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -177,244 +177,6 @@ pub(crate) fn codegen_x86_llvm_intrinsic_call<'tcx>(
bool_to_zero_or_max_uint(fx, res_lane_ty, res_lane)
});
}
"llvm.x86.sse2.psrli.d" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.sse2.psrli.d imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 32 => fx.bcx.ins().ushr_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.sse2.psrai.d" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.sse2.psrai.d imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 32 => fx.bcx.ins().sshr_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.sse2.pslli.d" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.sse2.pslli.d imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 32 => fx.bcx.ins().ishl_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.sse2.psrli.w" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.sse2.psrli.d imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 16 => fx.bcx.ins().ushr_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.sse2.psrai.w" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.sse2.psrai.d imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 16 => fx.bcx.ins().sshr_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.sse2.pslli.w" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.sse2.pslli.d imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 16 => fx.bcx.ins().ishl_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.avx.psrli.d" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.avx.psrli.d imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 32 => fx.bcx.ins().ushr_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.avx.psrai.d" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.avx.psrai.d imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 32 => fx.bcx.ins().sshr_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.sse2.psrli.q" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.avx.psrli.q imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 64 => fx.bcx.ins().ushr_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.sse2.pslli.q" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.avx.pslli.q imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 64 => fx.bcx.ins().ishl_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.avx.pslli.d" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.avx.pslli.d imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 32 => fx.bcx.ins().ishl_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.avx2.psrli.w" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.avx.psrli.w imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 16 => fx.bcx.ins().ushr_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.avx2.psrai.w" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.avx.psrai.w imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 16 => fx.bcx.ins().sshr_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.avx2.pslli.w" => {
let (a, imm8) = match args {
[a, imm8] => (a, imm8),
_ => bug!("wrong number of args for intrinsic {intrinsic}"),
};
let a = codegen_operand(fx, a);
let imm8 = crate::constant::mir_operand_get_const_val(fx, imm8)
.expect("llvm.x86.avx.pslli.w imm8 not const");

simd_for_each_lane(fx, a, ret, &|fx, _lane_ty, _res_lane_ty, lane| match imm8
.try_to_bits(Size::from_bytes(4))
.unwrap_or_else(|| panic!("imm8 not scalar: {:?}", imm8))
{
imm8 if imm8 < 16 => fx.bcx.ins().ishl_imm(lane, i64::from(imm8 as u8)),
_ => fx.bcx.ins().iconst(types::I32, 0),
});
}
"llvm.x86.ssse3.pshuf.b.128" | "llvm.x86.avx2.pshuf.b" => {
let (a, b) = match args {
[a, b] => (a, b),
Expand Down Expand Up @@ -506,14 +268,6 @@ pub(crate) fn codegen_x86_llvm_intrinsic_call<'tcx>(
ret.place_lane(fx, 2).to_ptr().store(fx, res_2, MemFlags::trusted());
ret.place_lane(fx, 3).to_ptr().store(fx, res_3, MemFlags::trusted());
}
"llvm.x86.sse2.storeu.dq" | "llvm.x86.sse2.storeu.pd" => {
intrinsic_args!(fx, args => (mem_addr, a); intrinsic);
let mem_addr = mem_addr.load_scalar(fx);

// FIXME correctly handle the unalignment
let dest = CPlace::for_ptr(Pointer::new(mem_addr), a.layout());
dest.write_cvalue(fx, a);
}
"llvm.x86.ssse3.pabs.b.128" | "llvm.x86.ssse3.pabs.w.128" | "llvm.x86.ssse3.pabs.d.128" => {
let a = match args {
[a] => a,
Expand Down Expand Up @@ -571,8 +325,6 @@ pub(crate) fn codegen_x86_llvm_intrinsic_call<'tcx>(
// llvm.x86.avx2.vperm2i128
// llvm.x86.ssse3.pshuf.b.128
// llvm.x86.avx2.pshuf.b
// llvm.x86.avx2.psrli.w
// llvm.x86.sse2.psrli.w

fn llvm_add_sub<'tcx>(
fx: &mut FunctionCx<'_, '_, 'tcx>,
Expand Down
2 changes: 1 addition & 1 deletion library/stdarch
Submodule stdarch updated 54 files
+20 −16 .github/workflows/main.yml
+3 −4 ci/docker/riscv64gc-unknown-linux-gnu/Dockerfile
+4 −2 ci/dox.sh
+5 −0 ci/run-docker.sh
+6 −0 ci/run.sh
+0 −23 crates/core_arch/src/aarch64/armclang.rs
+0 −14 crates/core_arch/src/aarch64/mod.rs
+525 −233 crates/core_arch/src/aarch64/neon/generated.rs
+12 −12 crates/core_arch/src/aarch64/neon/mod.rs
+2 −2 crates/core_arch/src/aarch64/tme.rs
+0 −104 crates/core_arch/src/aarch64/v8.rs
+0 −35 crates/core_arch/src/arm/armclang.rs
+0 −125 crates/core_arch/src/arm/ex.rs
+0 −14 crates/core_arch/src/arm/mod.rs
+0 −28 crates/core_arch/src/arm/neon.rs
+0 −49 crates/core_arch/src/arm/v6.rs
+0 −87 crates/core_arch/src/arm/v7.rs
+0 −3 crates/core_arch/src/arm_shared/mod.rs
+625 −221 crates/core_arch/src/arm_shared/neon/generated.rs
+0 −9 crates/core_arch/src/arm_shared/registers/aarch32.rs
+0 −121 crates/core_arch/src/arm_shared/registers/mod.rs
+0 −39 crates/core_arch/src/arm_shared/registers/v6m.rs
+0 −17 crates/core_arch/src/arm_shared/registers/v7m.rs
+2 −0 crates/core_arch/src/lib.rs
+7 −2 crates/core_arch/src/mod.rs
+106 −1 crates/core_arch/src/powerpc/altivec.rs
+5 −0 crates/core_arch/src/riscv32/mod.rs
+367 −0 crates/core_arch/src/riscv32/zk.rs
+4 −0 crates/core_arch/src/riscv64/mod.rs
+281 −0 crates/core_arch/src/riscv64/zk.rs
+52 −171 crates/core_arch/src/riscv_shared/mod.rs
+150 −0 crates/core_arch/src/riscv_shared/zb.rs
+462 −0 crates/core_arch/src/riscv_shared/zk.rs
+2 −2 crates/core_arch/src/wasm32/relaxed_simd.rs
+9 −29 crates/core_arch/src/wasm32/simd128.rs
+11 −17 crates/core_arch/src/x86/avx.rs
+33 −25 crates/core_arch/src/x86/avx2.rs
+81 −77 crates/core_arch/src/x86/avx512bw.rs
+258 −244 crates/core_arch/src/x86/avx512f.rs
+3 −3 crates/core_arch/src/x86/mod.rs
+34 −30 crates/core_arch/src/x86/sse2.rs
+6 −15 crates/intrinsic-test/missing_aarch64.txt
+0 −18 crates/intrinsic-test/missing_arm.txt
+4 −4 crates/intrinsic-test/src/argument.rs
+3 −1 crates/std_detect/src/detect/arch/arm.rs
+8 −0 crates/std_detect/src/detect/os/linux/arm.rs
+0 −2 crates/std_detect/src/lib.rs
+14 −5 crates/std_detect/tests/cpu-detection.rs
+305 −38 crates/stdarch-gen/neon.spec
+20 −1 crates/stdarch-gen/src/main.rs
+3 −1 crates/stdarch-test/Cargo.toml
+33 −25 crates/stdarch-test/src/disassembly.rs
+9 −6 crates/stdarch-test/src/lib.rs
+1 −1 examples/connect5.rs

0 comments on commit b0d4553

Please sign in to comment.