caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
ec056f54589d937ae3f74d795b94de886cbb7ec7
clang-p2996/llvm/test/CodeGen/RISCV/rvv/calling-conv.ll
Philip Reames 859c871184 [RISCV] Default to MicroOpBufferSize = 1 for scheduling purposes (#126608) 
This change introduces a default schedule model for the RISCV target
which leaves everything unchanged except the MicroOpBufferSize. The
default value of this flag in NoSched is 0. Both configurations
represent in order cores (i.e. no reorder window), the difference
between them comes down to whether heuristics other than latency are
allowed to apply. (Implementation details below)

I left the processor models which explicitly set MicroOpBufferSize=0
unchanged in this patch, but strongly suspect we should change those
too. Honestly, I think the LLVM wide default for this flag should be
changed, but don't have the energy to manage the updates for all
targets.

Implementation wise, the effect of this change is that schedule units
which are ready to run *except that* one of their predecessors may not
have completed yet are added to the Available list, not the Pending one.
The result of this is that it becomes possible to chose to schedule a
node before it's ready cycle if the heuristics prefer. This is
essentially chosing to insert a resource stall instead of e.g.
increasing register pressure.

Note that I was initially concerned there might be a correctness aspect
(as in some kind of exposed pipeline design), but the generic scheduler
doesn't seem to know how to insert noop instructions. Without that, a
program wouldn't be guaranteed to schedule on an exposed pipeline
depending on the program and schedule model in question.

The effect of this is that we sometimes prefer register pressure in
codegen results. This is mostly churn (or small wins) on scalar because
we have many more registers, but is of major importance on vector -
particularly high LMUL - because we effectively have many fewer
registers and the relative cost of spilling is much higher. This is a
significant improvement on high LMUL code quality for default rva23u
configurations - or any non -mcpu vector configuration for that matter.

Fixes #107532
2025-02-12 12:31:39 -08:00

185 lines
8.0 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=riscv32 -mattr=+m,+v < %s | FileCheck %s --check-prefixes=CHECK,RV32
; RUN: llc -mtriple=riscv64 -mattr=+m,+v < %s | FileCheck %s --check-prefixes=CHECK,RV64
; Check that we correctly scale the split part indirect offsets by VSCALE.
define <vscale x 32 x i32> @callee_scalable_vector_split_indirect(<vscale x 32 x i32> %x, <vscale x 32 x i32> %y) {
; CHECK-LABEL: callee_scalable_vector_split_indirect:
; CHECK: # %bb.0:
; CHECK-NEXT: csrr a1, vlenb
; CHECK-NEXT: slli a1, a1, 3
; CHECK-NEXT: add a1, a0, a1
; CHECK-NEXT: vl8re32.v v24, (a1)
; CHECK-NEXT: vl8re32.v v0, (a0)
; CHECK-NEXT: vsetvli a0, zero, e32, m8, ta, ma
; CHECK-NEXT: vadd.vv v8, v8, v0
; CHECK-NEXT: vadd.vv v16, v16, v24
; CHECK-NEXT: ret
%a = add <vscale x 32 x i32> %x, %y
ret <vscale x 32 x i32> %a
}
; Call the function above. Check that we set the arguments correctly.
define <vscale x 32 x i32> @caller_scalable_vector_split_indirect(<vscale x 32 x i32> %x) {
; RV32-LABEL: caller_scalable_vector_split_indirect:
; RV32: # %bb.0:
; RV32-NEXT: addi sp, sp, -144
; RV32-NEXT: .cfi_def_cfa_offset 144
; RV32-NEXT: sw ra, 140(sp) # 4-byte Folded Spill
; RV32-NEXT: sw s0, 136(sp) # 4-byte Folded Spill
; RV32-NEXT: .cfi_offset ra, -4
; RV32-NEXT: .cfi_offset s0, -8
; RV32-NEXT: addi s0, sp, 144
; RV32-NEXT: .cfi_def_cfa s0, 0
; RV32-NEXT: csrr a0, vlenb
; RV32-NEXT: slli a0, a0, 4
; RV32-NEXT: sub sp, sp, a0
; RV32-NEXT: andi sp, sp, -128
; RV32-NEXT: addi a0, sp, 128
; RV32-NEXT: csrr a1, vlenb
; RV32-NEXT: vs8r.v v8, (a0)
; RV32-NEXT: vsetvli a2, zero, e32, m8, ta, ma
; RV32-NEXT: vmv.v.i v8, 0
; RV32-NEXT: slli a1, a1, 3
; RV32-NEXT: add a0, a0, a1
; RV32-NEXT: vs8r.v v16, (a0)
; RV32-NEXT: addi a0, sp, 128
; RV32-NEXT: vmv.v.i v16, 0
; RV32-NEXT: call callee_scalable_vector_split_indirect
; RV32-NEXT: addi sp, s0, -144
; RV32-NEXT: .cfi_def_cfa sp, 144
; RV32-NEXT: lw ra, 140(sp) # 4-byte Folded Reload
; RV32-NEXT: lw s0, 136(sp) # 4-byte Folded Reload
; RV32-NEXT: .cfi_restore ra
; RV32-NEXT: .cfi_restore s0
; RV32-NEXT: addi sp, sp, 144
; RV32-NEXT: .cfi_def_cfa_offset 0
; RV32-NEXT: ret
;
; RV64-LABEL: caller_scalable_vector_split_indirect:
; RV64: # %bb.0:
; RV64-NEXT: addi sp, sp, -144
; RV64-NEXT: .cfi_def_cfa_offset 144
; RV64-NEXT: sd ra, 136(sp) # 8-byte Folded Spill
; RV64-NEXT: sd s0, 128(sp) # 8-byte Folded Spill
; RV64-NEXT: .cfi_offset ra, -8
; RV64-NEXT: .cfi_offset s0, -16
; RV64-NEXT: addi s0, sp, 144
; RV64-NEXT: .cfi_def_cfa s0, 0
; RV64-NEXT: csrr a0, vlenb
; RV64-NEXT: slli a0, a0, 4
; RV64-NEXT: sub sp, sp, a0
; RV64-NEXT: andi sp, sp, -128
; RV64-NEXT: addi a0, sp, 128
; RV64-NEXT: csrr a1, vlenb
; RV64-NEXT: vs8r.v v8, (a0)
; RV64-NEXT: vsetvli a2, zero, e32, m8, ta, ma
; RV64-NEXT: vmv.v.i v8, 0
; RV64-NEXT: slli a1, a1, 3
; RV64-NEXT: add a0, a0, a1
; RV64-NEXT: vs8r.v v16, (a0)
; RV64-NEXT: addi a0, sp, 128
; RV64-NEXT: vmv.v.i v16, 0
; RV64-NEXT: call callee_scalable_vector_split_indirect
; RV64-NEXT: addi sp, s0, -144
; RV64-NEXT: .cfi_def_cfa sp, 144
; RV64-NEXT: ld ra, 136(sp) # 8-byte Folded Reload
; RV64-NEXT: ld s0, 128(sp) # 8-byte Folded Reload
; RV64-NEXT: .cfi_restore ra
; RV64-NEXT: .cfi_restore s0
; RV64-NEXT: addi sp, sp, 144
; RV64-NEXT: .cfi_def_cfa_offset 0
; RV64-NEXT: ret
%c = alloca i64
%a = call <vscale x 32 x i32> @callee_scalable_vector_split_indirect(<vscale x 32 x i32> zeroinitializer, <vscale x 32 x i32> %x)
ret <vscale x 32 x i32> %a
}
define target("riscv.vector.tuple", <vscale x 16 x i8>, 2) @caller_tuple_return() {
; RV32-LABEL: caller_tuple_return:
; RV32: # %bb.0:
; RV32-NEXT: addi sp, sp, -16
; RV32-NEXT: .cfi_def_cfa_offset 16
; RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32-NEXT: .cfi_offset ra, -4
; RV32-NEXT: call callee_tuple_return
; RV32-NEXT: vsetivli zero, 1, e8, m1, ta, ma
; RV32-NEXT: vmv2r.v v6, v8
; RV32-NEXT: vmv2r.v v8, v10
; RV32-NEXT: vmv2r.v v10, v6
; RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32-NEXT: .cfi_restore ra
; RV32-NEXT: addi sp, sp, 16
; RV32-NEXT: .cfi_def_cfa_offset 0
; RV32-NEXT: ret
;
; RV64-LABEL: caller_tuple_return:
; RV64: # %bb.0:
; RV64-NEXT: addi sp, sp, -16
; RV64-NEXT: .cfi_def_cfa_offset 16
; RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
; RV64-NEXT: .cfi_offset ra, -8
; RV64-NEXT: call callee_tuple_return
; RV64-NEXT: vsetivli zero, 1, e8, m1, ta, ma
; RV64-NEXT: vmv2r.v v6, v8
; RV64-NEXT: vmv2r.v v8, v10
; RV64-NEXT: vmv2r.v v10, v6
; RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
; RV64-NEXT: .cfi_restore ra
; RV64-NEXT: addi sp, sp, 16
; RV64-NEXT: .cfi_def_cfa_offset 0
; RV64-NEXT: ret
%a = call target("riscv.vector.tuple", <vscale x 16 x i8>, 2) @callee_tuple_return()
%b = call <vscale x 4 x i32> @llvm.riscv.tuple.extract.nxv4i32.triscv.vector.tuple_nxv16i8_2t(target("riscv.vector.tuple", <vscale x 16 x i8>, 2) %a, i32 0)
%c = call <vscale x 4 x i32> @llvm.riscv.tuple.extract.nxv4i32.triscv.vector.tuple_nxv16i8_2t(target("riscv.vector.tuple", <vscale x 16 x i8>, 2) %a, i32 1)
%d = call target("riscv.vector.tuple", <vscale x 16 x i8>, 2) @llvm.riscv.tuple.insert.triscv.vector.tuple_nxv16i8_2t.nxv4i32(target("riscv.vector.tuple", <vscale x 16 x i8>, 2) poison, <vscale x 4 x i32> %c, i32 0)
%e = call target("riscv.vector.tuple", <vscale x 16 x i8>, 2) @llvm.riscv.tuple.insert.triscv.vector.tuple_nxv16i8_2t.nxv4i32(target("riscv.vector.tuple", <vscale x 16 x i8>, 2) %d, <vscale x 4 x i32> %b, i32 1)
ret target("riscv.vector.tuple", <vscale x 16 x i8>, 2) %e
}
declare target("riscv.vector.tuple", <vscale x 16 x i8>, 2) @callee_tuple_return()
define void @caller_tuple_argument(target("riscv.vector.tuple", <vscale x 16 x i8>, 2) %x) {
; RV32-LABEL: caller_tuple_argument:
; RV32: # %bb.0:
; RV32-NEXT: addi sp, sp, -16
; RV32-NEXT: .cfi_def_cfa_offset 16
; RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
; RV32-NEXT: .cfi_offset ra, -4
; RV32-NEXT: vsetivli zero, 1, e8, m1, ta, ma
; RV32-NEXT: vmv2r.v v6, v8
; RV32-NEXT: vmv2r.v v8, v10
; RV32-NEXT: vmv2r.v v10, v6
; RV32-NEXT: call callee_tuple_argument
; RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
; RV32-NEXT: .cfi_restore ra
; RV32-NEXT: addi sp, sp, 16
; RV32-NEXT: .cfi_def_cfa_offset 0
; RV32-NEXT: ret
;
; RV64-LABEL: caller_tuple_argument:
; RV64: # %bb.0:
; RV64-NEXT: addi sp, sp, -16
; RV64-NEXT: .cfi_def_cfa_offset 16
; RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
; RV64-NEXT: .cfi_offset ra, -8
; RV64-NEXT: vsetivli zero, 1, e8, m1, ta, ma
; RV64-NEXT: vmv2r.v v6, v8
; RV64-NEXT: vmv2r.v v8, v10
; RV64-NEXT: vmv2r.v v10, v6
; RV64-NEXT: call callee_tuple_argument
; RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
; RV64-NEXT: .cfi_restore ra
; RV64-NEXT: addi sp, sp, 16
; RV64-NEXT: .cfi_def_cfa_offset 0
; RV64-NEXT: ret
%a = call <vscale x 4 x i32> @llvm.riscv.tuple.extract.nxv4i32.triscv.vector.tuple_nxv16i8_2t(target("riscv.vector.tuple", <vscale x 16 x i8>, 2) %x, i32 0)
%b = call <vscale x 4 x i32> @llvm.riscv.tuple.extract.nxv4i32.triscv.vector.tuple_nxv16i8_2t(target("riscv.vector.tuple", <vscale x 16 x i8>, 2) %x, i32 1)
%c = call target("riscv.vector.tuple", <vscale x 16 x i8>, 2) @llvm.riscv.tuple.insert.triscv.vector.tuple_nxv16i8_2t.nxv4i32(target("riscv.vector.tuple", <vscale x 16 x i8>, 2) poison, <vscale x 4 x i32> %b, i32 0)
%d = call target("riscv.vector.tuple", <vscale x 16 x i8>, 2) @llvm.riscv.tuple.insert.triscv.vector.tuple_nxv16i8_2t.nxv4i32(target("riscv.vector.tuple", <vscale x 16 x i8>, 2) %c, <vscale x 4 x i32> %a, i32 1)
call void @callee_tuple_argument(target("riscv.vector.tuple", <vscale x 16 x i8>, 2) %d)
ret void
}
declare void @callee_tuple_argument(target("riscv.vector.tuple", <vscale x 16 x i8>, 2))
Reference in New Issue View Git Blame Copy Permalink