shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-amdgpu_kernel.ll
Fabian Ritter ef6eaa045a
[GISel] Introduce MIFlags::InBounds (#150900) 
This flag applies to G_PTR_ADD instructions and indicates that the operation
implements an inbounds getelementptr operation, i.e., the pointer operand is in
bounds wrt. the allocated object it is based on, and the arithmetic does not
change that.

It is set when the IRTranslator lowers inbounds GEPs (currently only in some
cases, to be extended with a future PR), and in the
(build|materialize)ObjectPtrOffset functions.

Inbounds information is useful in ISel when we have instructions that perform
address computations whose intermediate steps must be in the same memory region
as the final result. A follow-up patch will start using it for AMDGPU's flat
memory instructions, where the immediate offset must not affect the memory
aperture of the address.

This is analogous to a concurrent effort in SDAG: #131862
(related: #140017, #141725).

For SWDEV-516125.
2025-07-30 13:01:23 +02:00

2105 lines
132 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -amdgpu-ir-lower-kernel-arguments=0 -stop-after=irtranslator %s -o - | FileCheck -check-prefix=HSA-VI %s
; RUN: llc -global-isel -mtriple=amdgcn-- -mcpu=fiji -amdgpu-ir-lower-kernel-arguments=0 -stop-after=irtranslator %s -o - | FileCheck -check-prefix=LEGACY-MESA-VI %s
define amdgpu_kernel void @i8_arg(ptr addrspace(1) nocapture %out, i8 %in) nounwind {
; HSA-VI-LABEL: name: i8_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s8), align 8, addrspace 4)
; HSA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s8)
; HSA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i8_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s8), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s8)
; LEGACY-MESA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%ext = zext i8 %in to i32
store i32 %ext, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @i8_zext_arg(ptr addrspace(1) nocapture %out, i8 zeroext %in) nounwind {
; HSA-VI-LABEL: name: i8_zext_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s8), align 8, addrspace 4)
; HSA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s8)
; HSA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i8_zext_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s8), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s8)
; LEGACY-MESA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%ext = zext i8 %in to i32
store i32 %ext, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @i8_sext_arg(ptr addrspace(1) nocapture %out, i8 signext %in) nounwind {
; HSA-VI-LABEL: name: i8_sext_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s8), align 8, addrspace 4)
; HSA-VI-NEXT: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[LOAD1]](s8)
; HSA-VI-NEXT: G_STORE [[SEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i8_sext_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s8), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[LOAD1]](s8)
; LEGACY-MESA-VI-NEXT: G_STORE [[SEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%ext = sext i8 %in to i32
store i32 %ext, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @i16_arg(ptr addrspace(1) nocapture %out, i16 %in) nounwind {
; HSA-VI-LABEL: name: i16_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s16) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s16), align 8, addrspace 4)
; HSA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s16)
; HSA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i16_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s16) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s16), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s16)
; LEGACY-MESA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%ext = zext i16 %in to i32
store i32 %ext, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @i16_zext_arg(ptr addrspace(1) nocapture %out, i16 zeroext %in) nounwind {
; HSA-VI-LABEL: name: i16_zext_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s16) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s16), align 8, addrspace 4)
; HSA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s16)
; HSA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i16_zext_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s16) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s16), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s16)
; LEGACY-MESA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%ext = zext i16 %in to i32
store i32 %ext, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @i16_sext_arg(ptr addrspace(1) nocapture %out, i16 signext %in) nounwind {
; HSA-VI-LABEL: name: i16_sext_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s16) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s16), align 8, addrspace 4)
; HSA-VI-NEXT: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[LOAD1]](s16)
; HSA-VI-NEXT: G_STORE [[SEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i16_sext_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s16) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s16), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[LOAD1]](s16)
; LEGACY-MESA-VI-NEXT: G_STORE [[SEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%ext = sext i16 %in to i32
store i32 %ext, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @i32_arg(ptr addrspace(1) nocapture %out, i32 %in) nounwind {
; HSA-VI-LABEL: name: i32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s32), align 8, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store i32 %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @f32_arg(ptr addrspace(1) nocapture %out, float %in) nounwind {
; HSA-VI-LABEL: name: f32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s32), align 8, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: f32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store float %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v2i8_arg(ptr addrspace(1) %out, <2 x i8> %in) {
; HSA-VI-LABEL: name: v2i8_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<2 x s8>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<2 x s8>), align 8, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<2 x s8>), [[LOAD]](p1) :: (store (<2 x s8>) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v2i8_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<2 x s8>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<2 x s8>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<2 x s8>), [[LOAD]](p1) :: (store (<2 x s8>) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <2 x i8> %in, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v2i16_arg(ptr addrspace(1) %out, <2 x i16> %in) {
; HSA-VI-LABEL: name: v2i16_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<2 x s16>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<2 x s16>), align 8, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<2 x s16>), [[LOAD]](p1) :: (store (<2 x s16>) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v2i16_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<2 x s16>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<2 x s16>), addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<2 x s16>), [[LOAD]](p1) :: (store (<2 x s16>) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <2 x i16> %in, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v2i32_arg(ptr addrspace(1) nocapture %out, <2 x i32> %in) nounwind {
; HSA-VI-LABEL: name: v2i32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<2 x s32>), addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<2 x s32>), [[LOAD]](p1) :: (store (<2 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v2i32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<2 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<2 x s32>), [[LOAD]](p1) :: (store (<2 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <2 x i32> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v2f32_arg(ptr addrspace(1) nocapture %out, <2 x float> %in) nounwind {
; HSA-VI-LABEL: name: v2f32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<2 x s32>), addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<2 x s32>), [[LOAD]](p1) :: (store (<2 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v2f32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<2 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<2 x s32>), [[LOAD]](p1) :: (store (<2 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <2 x float> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v3i8_arg(ptr addrspace(1) nocapture %out, <3 x i8> %in) nounwind {
; HSA-VI-LABEL: name: v3i8_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<3 x s8>), align 8, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<3 x s8>), [[LOAD]](p1) :: (store (<3 x s8>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v3i8_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<3 x s8>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<3 x s8>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<3 x s8>), [[LOAD]](p1) :: (store (<3 x s8>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <3 x i8> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v3i16_arg(ptr addrspace(1) nocapture %out, <3 x i16> %in) nounwind {
; HSA-VI-LABEL: name: v3i16_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<3 x s16>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<3 x s16>), align 8, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<3 x s16>), [[LOAD]](p1) :: (store (<3 x s16>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v3i16_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<3 x s16>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<3 x s16>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<3 x s16>), [[LOAD]](p1) :: (store (<3 x s16>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <3 x i16> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v3i32_arg(ptr addrspace(1) nocapture %out, <3 x i32> %in) nounwind {
; HSA-VI-LABEL: name: v3i32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<3 x s32>), align 16, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<3 x s32>), [[LOAD]](p1) :: (store (<3 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v3i32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<3 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<3 x s32>), [[LOAD]](p1) :: (store (<3 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <3 x i32> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v3f32_arg(ptr addrspace(1) nocapture %out, <3 x float> %in) nounwind {
; HSA-VI-LABEL: name: v3f32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<3 x s32>), align 16, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<3 x s32>), [[LOAD]](p1) :: (store (<3 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v3f32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<3 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<3 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<3 x s32>), [[LOAD]](p1) :: (store (<3 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <3 x float> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v4i8_arg(ptr addrspace(1) %out, <4 x i8> %in) {
; HSA-VI-LABEL: name: v4i8_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<4 x s8>), align 8, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<4 x s8>), [[LOAD]](p1) :: (store (<4 x s8>) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v4i8_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<4 x s8>), addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<4 x s8>), [[LOAD]](p1) :: (store (<4 x s8>) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <4 x i8> %in, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v4i16_arg(ptr addrspace(1) %out, <4 x i16> %in) {
; HSA-VI-LABEL: name: v4i16_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<4 x s16>), addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<4 x s16>), [[LOAD]](p1) :: (store (<4 x s16>) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v4i16_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<4 x s16>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<4 x s16>), [[LOAD]](p1) :: (store (<4 x s16>) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <4 x i16> %in, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v4i32_arg(ptr addrspace(1) nocapture %out, <4 x i32> %in) nounwind {
; HSA-VI-LABEL: name: v4i32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<4 x s32>), addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<4 x s32>), [[LOAD]](p1) :: (store (<4 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v4i32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<4 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<4 x s32>), [[LOAD]](p1) :: (store (<4 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <4 x i32> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v4f32_arg(ptr addrspace(1) nocapture %out, <4 x float> %in) nounwind {
; HSA-VI-LABEL: name: v4f32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<4 x s32>), addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<4 x s32>), [[LOAD]](p1) :: (store (<4 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v4f32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<4 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<4 x s32>), [[LOAD]](p1) :: (store (<4 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <4 x float> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v8i8_arg(ptr addrspace(1) %out, <8 x i8> %in) {
; HSA-VI-LABEL: name: v8i8_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<8 x s8>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<8 x s8>), addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<8 x s8>), [[LOAD]](p1) :: (store (<8 x s8>) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v8i8_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<8 x s8>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<8 x s8>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<8 x s8>), [[LOAD]](p1) :: (store (<8 x s8>) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <8 x i8> %in, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v8i16_arg(ptr addrspace(1) %out, <8 x i16> %in) {
; HSA-VI-LABEL: name: v8i16_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<8 x s16>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<8 x s16>), addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<8 x s16>), [[LOAD]](p1) :: (store (<8 x s16>) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v8i16_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<8 x s16>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<8 x s16>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<8 x s16>), [[LOAD]](p1) :: (store (<8 x s16>) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <8 x i16> %in, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v8i32_arg(ptr addrspace(1) nocapture %out, <8 x i32> %in) nounwind {
; HSA-VI-LABEL: name: v8i32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 32
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<8 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<8 x s32>), align 16, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<8 x s32>), [[LOAD]](p1) :: (store (<8 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v8i32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 68
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<8 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<8 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<8 x s32>), [[LOAD]](p1) :: (store (<8 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <8 x i32> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v8f32_arg(ptr addrspace(1) nocapture %out, <8 x float> %in) nounwind {
; HSA-VI-LABEL: name: v8f32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 32
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<8 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<8 x s32>), align 16, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<8 x s32>), [[LOAD]](p1) :: (store (<8 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v8f32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 68
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<8 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<8 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<8 x s32>), [[LOAD]](p1) :: (store (<8 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <8 x float> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v16i8_arg(ptr addrspace(1) %out, <16 x i8> %in) {
; HSA-VI-LABEL: name: v16i8_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<16 x s8>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<16 x s8>), addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<16 x s8>), [[LOAD]](p1) :: (store (<16 x s8>) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v16i8_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<16 x s8>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<16 x s8>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<16 x s8>), [[LOAD]](p1) :: (store (<16 x s8>) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <16 x i8> %in, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v16i16_arg(ptr addrspace(1) %out, <16 x i16> %in) {
; HSA-VI-LABEL: name: v16i16_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 32
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<16 x s16>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<16 x s16>), align 16, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<16 x s16>), [[LOAD]](p1) :: (store (<16 x s16>) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v16i16_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 68
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<16 x s16>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<16 x s16>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<16 x s16>), [[LOAD]](p1) :: (store (<16 x s16>) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <16 x i16> %in, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v16i32_arg(ptr addrspace(1) nocapture %out, <16 x i32> %in) nounwind {
; HSA-VI-LABEL: name: v16i32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 64
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<16 x s32>), align 16, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<16 x s32>), [[LOAD]](p1) :: (store (<16 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v16i32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 100
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<16 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<16 x s32>), [[LOAD]](p1) :: (store (<16 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <16 x i32> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @v16f32_arg(ptr addrspace(1) nocapture %out, <16 x float> %in) nounwind {
; HSA-VI-LABEL: name: v16f32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 64
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<16 x s32>), align 16, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<16 x s32>), [[LOAD]](p1) :: (store (<16 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v16f32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 100
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<16 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<16 x s32>), [[LOAD]](p1) :: (store (<16 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store <16 x float> %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @kernel_arg_i64(ptr addrspace(1) %out, i64 %a) nounwind {
; HSA-VI-LABEL: name: kernel_arg_i64
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s64), addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s64), [[LOAD]](p1) :: (store (s64) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: kernel_arg_i64
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s64), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s64), [[LOAD]](p1) :: (store (s64) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
store i64 %a, ptr addrspace(1) %out, align 8
ret void
}
define amdgpu_kernel void @f64_kernel_arg(ptr addrspace(1) %out, double %in) {
; HSA-VI-LABEL: name: f64_kernel_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s64), addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s64), [[LOAD]](p1) :: (store (s64) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: f64_kernel_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s64), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s64), [[LOAD]](p1) :: (store (s64) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
store double %in, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @i1_arg(ptr addrspace(1) %out, i1 %x) nounwind {
; HSA-VI-LABEL: name: i1_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s1), align 8, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s1), [[LOAD]](p1) :: (store (s1) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i1_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s1), [[LOAD]](p1) :: (store (s1) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
store i1 %x, ptr addrspace(1) %out, align 1
ret void
}
define amdgpu_kernel void @i1_arg_zext_i32(ptr addrspace(1) %out, i1 %x) nounwind {
; HSA-VI-LABEL: name: i1_arg_zext_i32
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s1), align 8, addrspace 4)
; HSA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s1)
; HSA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i1_arg_zext_i32
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s1)
; LEGACY-MESA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%ext = zext i1 %x to i32
store i32 %ext, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @i1_arg_zext_i64(ptr addrspace(1) %out, i1 %x) nounwind {
; HSA-VI-LABEL: name: i1_arg_zext_i64
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s1), align 8, addrspace 4)
; HSA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[LOAD1]](s1)
; HSA-VI-NEXT: G_STORE [[ZEXT]](s64), [[LOAD]](p1) :: (store (s64) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i1_arg_zext_i64
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[LOAD1]](s1)
; LEGACY-MESA-VI-NEXT: G_STORE [[ZEXT]](s64), [[LOAD]](p1) :: (store (s64) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%ext = zext i1 %x to i64
store i64 %ext, ptr addrspace(1) %out, align 8
ret void
}
define amdgpu_kernel void @i1_arg_sext_i32(ptr addrspace(1) %out, i1 %x) nounwind {
; HSA-VI-LABEL: name: i1_arg_sext_i32
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s1), align 8, addrspace 4)
; HSA-VI-NEXT: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[LOAD1]](s1)
; HSA-VI-NEXT: G_STORE [[SEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i1_arg_sext_i32
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[LOAD1]](s1)
; LEGACY-MESA-VI-NEXT: G_STORE [[SEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%ext = sext i1 %x to i32
store i32 %ext, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @i1_arg_sext_i64(ptr addrspace(1) %out, i1 %x) nounwind {
; HSA-VI-LABEL: name: i1_arg_sext_i64
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s1), align 8, addrspace 4)
; HSA-VI-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[LOAD1]](s1)
; HSA-VI-NEXT: G_STORE [[SEXT]](s64), [[LOAD]](p1) :: (store (s64) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: i1_arg_sext_i64
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[LOAD1]](s1)
; LEGACY-MESA-VI-NEXT: G_STORE [[SEXT]](s64), [[LOAD]](p1) :: (store (s64) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%ext = sext i1 %x to i64
store i64 %ext, ptr addrspace(1) %out, align 8
ret void
}
; 0-sized arguments do not add a slot to the argument register set, so
; waste an index in the virtual register array.
define amdgpu_kernel void @empty_struct_arg({} %arg0, i32 %arg1) nounwind {
; HSA-VI-LABEL: name: empty_struct_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32), align 16, addrspace 4)
; HSA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; HSA-VI-NEXT: G_STORE [[LOAD]](s32), [[DEF]](p1) :: (store (s32) into `ptr addrspace(1) poison`, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: empty_struct_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD]](s32), [[DEF]](p1) :: (store (s32) into `ptr addrspace(1) poison`, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
store i32 %arg1, ptr addrspace(1) poison
ret void
}
define amdgpu_kernel void @empty_array_arg([0 x i8] %arg0, i32 %arg1) nounwind {
; HSA-VI-LABEL: name: empty_array_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32), align 16, addrspace 4)
; HSA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; HSA-VI-NEXT: G_STORE [[LOAD]](s32), [[DEF]](p1) :: (store (s32) into `ptr addrspace(1) poison`, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: empty_array_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD]](s32), [[DEF]](p1) :: (store (s32) into `ptr addrspace(1) poison`, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
store i32 %arg1, ptr addrspace(1) poison
ret void
}
; The correct load offsets for these:
; load 4 from 0,
; load 8 from 8
; load 4 from 24
; load 8 from 32
; With the SelectionDAG argument lowering, the alignments for the
; struct members is not properly considered, making these wrong.
define amdgpu_kernel void @struct_argument_alignment({i32, i64} %arg0, i8 %pad, {i32, i64} %arg1) {
; HSA-VI-LABEL: name: struct_argument_alignment
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s64), addrspace 4)
; HSA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; HSA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s8), align 16, addrspace 4)
; HSA-VI-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 24
; HSA-VI-NEXT: [[PTR_ADD3:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C3]](s64)
; HSA-VI-NEXT: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p4) :: (dereferenceable invariant load (s32), align 8, addrspace 4)
; HSA-VI-NEXT: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 32
; HSA-VI-NEXT: [[PTR_ADD4:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C4]](s64)
; HSA-VI-NEXT: [[LOAD4:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD4]](p4) :: (dereferenceable invariant load (s64), addrspace 4)
; HSA-VI-NEXT: [[C5:%[0-9]+]]:_(p1) = G_CONSTANT i64 0
; HSA-VI-NEXT: G_STORE [[LOAD]](s32), [[C5]](p1) :: (volatile store (s32) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s64), [[C5]](p1) :: (volatile store (s64) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD2]](s8), [[C5]](p1) :: (volatile store (s8) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD3]](s32), [[C5]](p1) :: (volatile store (s32) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD4]](s64), [[C5]](p1) :: (volatile store (s64) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: struct_argument_alignment
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s64), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s8), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 60
; LEGACY-MESA-VI-NEXT: [[PTR_ADD3:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C3]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 68
; LEGACY-MESA-VI-NEXT: [[PTR_ADD4:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C4]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD4:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD4]](p4) :: (dereferenceable invariant load (s64), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C5:%[0-9]+]]:_(p1) = G_CONSTANT i64 0
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD]](s32), [[C5]](p1) :: (volatile store (s32) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s64), [[C5]](p1) :: (volatile store (s64) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD2]](s8), [[C5]](p1) :: (volatile store (s8) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD3]](s32), [[C5]](p1) :: (volatile store (s32) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD4]](s64), [[C5]](p1) :: (volatile store (s64) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%val0 = extractvalue {i32, i64} %arg0, 0
%val1 = extractvalue {i32, i64} %arg0, 1
%val2 = extractvalue {i32, i64} %arg1, 0
%val3 = extractvalue {i32, i64} %arg1, 1
store volatile i32 %val0, ptr addrspace(1) null
store volatile i64 %val1, ptr addrspace(1) null
store volatile i8 %pad, ptr addrspace(1) null
store volatile i32 %val2, ptr addrspace(1) null
store volatile i64 %val3, ptr addrspace(1) null
ret void
}
define amdgpu_kernel void @pointer_in_struct_argument({ptr addrspace(3), ptr addrspace(1)} %arg0, i8 %pad, {ptr addrspace(3), ptr addrspace(1234)} %arg1) {
; HSA-VI-LABEL: name: pointer_in_struct_argument
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s64), addrspace 4)
; HSA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; HSA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s8), align 16, addrspace 4)
; HSA-VI-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 24
; HSA-VI-NEXT: [[PTR_ADD3:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C3]](s64)
; HSA-VI-NEXT: [[LOAD3:%[0-9]+]]:_(p3) = G_LOAD [[PTR_ADD3]](p4) :: (dereferenceable invariant load (s32), align 8, addrspace 4)
; HSA-VI-NEXT: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 32
; HSA-VI-NEXT: [[PTR_ADD4:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C4]](s64)
; HSA-VI-NEXT: [[LOAD4:%[0-9]+]]:_(p1234) = G_LOAD [[PTR_ADD4]](p4) :: (dereferenceable invariant load (s64), addrspace 4)
; HSA-VI-NEXT: [[C5:%[0-9]+]]:_(p1) = G_CONSTANT i64 0
; HSA-VI-NEXT: G_STORE [[LOAD]](p3), [[C5]](p1) :: (volatile store (p3) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD1]](p1), [[C5]](p1) :: (volatile store (p1) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD2]](s8), [[C5]](p1) :: (volatile store (s8) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD3]](p3), [[C5]](p1) :: (volatile store (p3) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD4]](p1234), [[C5]](p1) :: (volatile store (p1234) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: pointer_in_struct_argument
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s64), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s8), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 60
; LEGACY-MESA-VI-NEXT: [[PTR_ADD3:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C3]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD3:%[0-9]+]]:_(p3) = G_LOAD [[PTR_ADD3]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 68
; LEGACY-MESA-VI-NEXT: [[PTR_ADD4:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C4]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD4:%[0-9]+]]:_(p1234) = G_LOAD [[PTR_ADD4]](p4) :: (dereferenceable invariant load (s64), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C5:%[0-9]+]]:_(p1) = G_CONSTANT i64 0
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD]](p3), [[C5]](p1) :: (volatile store (p3) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](p1), [[C5]](p1) :: (volatile store (p1) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD2]](s8), [[C5]](p1) :: (volatile store (s8) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD3]](p3), [[C5]](p1) :: (volatile store (p3) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD4]](p1234), [[C5]](p1) :: (volatile store (p1234) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%val0 = extractvalue {ptr addrspace(3), ptr addrspace(1)} %arg0, 0
%val1 = extractvalue {ptr addrspace(3), ptr addrspace(1)} %arg0, 1
%val2 = extractvalue {ptr addrspace(3), ptr addrspace(1234)} %arg1, 0
%val3 = extractvalue {ptr addrspace(3), ptr addrspace(1234)} %arg1, 1
store volatile ptr addrspace(3) %val0, ptr addrspace(1) null
store volatile ptr addrspace(1) %val1, ptr addrspace(1) null
store volatile i8 %pad, ptr addrspace(1) null
store volatile ptr addrspace(3) %val2, ptr addrspace(1) null
store volatile ptr addrspace(1234) %val3, ptr addrspace(1) null
ret void
}
; No padding between i8 and next struct, but round up at end to 4 byte
; multiple.
define amdgpu_kernel void @packed_struct_argument_alignment(<{i32, i64}> %arg0, i8, <{i32, i64}> %arg1) {
; HSA-VI-LABEL: name: packed_struct_argument_alignment
; HSA-VI: bb.1 (%ir-block.1):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 4
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s64), align 4, addrspace 4)
; HSA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 13
; HSA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; HSA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32), align 1, addrspace 4)
; HSA-VI-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 17
; HSA-VI-NEXT: [[PTR_ADD3:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C3]](s64)
; HSA-VI-NEXT: [[LOAD3:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD3]](p4) :: (dereferenceable invariant load (s64), align 1, addrspace 4)
; HSA-VI-NEXT: [[C4:%[0-9]+]]:_(p1) = G_CONSTANT i64 0
; HSA-VI-NEXT: G_STORE [[LOAD]](s32), [[C4]](p1) :: (volatile store (s32) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s64), [[C4]](p1) :: (volatile store (s64) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD2]](s32), [[C4]](p1) :: (volatile store (s32) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD3]](s64), [[C4]](p1) :: (volatile store (s64) into `ptr addrspace(1) null`, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: packed_struct_argument_alignment
; LEGACY-MESA-VI: bb.1 (%ir-block.1):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 40
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s64), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 49
; LEGACY-MESA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32), align 1, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 53
; LEGACY-MESA-VI-NEXT: [[PTR_ADD3:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C3]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD3:%[0-9]+]]:_(s64) = G_LOAD [[PTR_ADD3]](p4) :: (dereferenceable invariant load (s64), align 1, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C4:%[0-9]+]]:_(p1) = G_CONSTANT i64 0
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD]](s32), [[C4]](p1) :: (volatile store (s32) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s64), [[C4]](p1) :: (volatile store (s64) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD2]](s32), [[C4]](p1) :: (volatile store (s32) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD3]](s64), [[C4]](p1) :: (volatile store (s64) into `ptr addrspace(1) null`, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%val0 = extractvalue <{i32, i64}> %arg0, 0
%val1 = extractvalue <{i32, i64}> %arg0, 1
%val2 = extractvalue <{i32, i64}> %arg1, 0
%val3 = extractvalue <{i32, i64}> %arg1, 1
store volatile i32 %val0, ptr addrspace(1) null
store volatile i64 %val1, ptr addrspace(1) null
store volatile i32 %val2, ptr addrspace(1) null
store volatile i64 %val3, ptr addrspace(1) null
ret void
}
define amdgpu_kernel void @unused_i32_arg(ptr addrspace(1) nocapture %out, i32 %unused, i32 %in) nounwind {
; HSA-VI-LABEL: name: unused_i32_arg
; HSA-VI: bb.1.entry:
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: unused_i32_arg
; LEGACY-MESA-VI: bb.1.entry:
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
entry:
ret void
}
; Byref pointers should only be treated as offsets from kernarg
define amdgpu_kernel void @byref_constant_i8_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(i8) %in.byref) {
; HSA-VI-LABEL: name: byref_constant_i8_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s8) from %ir.in.byref, addrspace 4)
; HSA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s8)
; HSA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_constant_i8_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s8) from %ir.in.byref, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s8)
; LEGACY-MESA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load i8, ptr addrspace(4) %in.byref
%ext = zext i8 %in to i32
store i32 %ext, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @byref_constant_i16_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(i16) align 2 %in.byref) {
; HSA-VI-LABEL: name: byref_constant_i16_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s16) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s16) from %ir.in.byref, addrspace 4)
; HSA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s16)
; HSA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_constant_i16_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s16) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s16) from %ir.in.byref, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[LOAD1]](s16)
; LEGACY-MESA-VI-NEXT: G_STORE [[ZEXT]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load i16, ptr addrspace(4) %in.byref
%ext = zext i16 %in to i32
store i32 %ext, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @byref_constant_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(i32) align 4 %in.byref, i32 %after.offset) {
; HSA-VI-LABEL: name: byref_constant_i32_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 12
; HSA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; HSA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s32) from %ir.in.byref, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD2]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_constant_i32_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 48
; LEGACY-MESA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32), align 16, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s32) from %ir.in.byref, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD2]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load i32, ptr addrspace(4) %in.byref
store volatile i32 %in, ptr addrspace(1) %out, align 4
store volatile i32 %after.offset, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @byref_constant_v4i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(<4 x i32>) align(16) %in.byref, i32 %after.offset) {
; HSA-VI-LABEL: name: byref_constant_v4i32_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 32
; HSA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32), align 16, addrspace 4)
; HSA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<4 x s32>) from %ir.in.byref, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD2]](<4 x s32>), [[LOAD]](p1) :: (volatile store (<4 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_constant_v4i32_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 68
; LEGACY-MESA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<4 x s32>) from %ir.in.byref, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD2]](<4 x s32>), [[LOAD]](p1) :: (volatile store (<4 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load <4 x i32>, ptr addrspace(4) %in.byref
store volatile <4 x i32> %in, ptr addrspace(1) %out, align 4
store volatile i32 %after.offset, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @byref_align_constant_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(i32) align(256) %in.byref, i32 %after.offset) {
; HSA-VI-LABEL: name: byref_align_constant_i32_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 256
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 260
; HSA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; HSA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s32) from %ir.in.byref, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD2]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_align_constant_i32_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 292
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 296
; LEGACY-MESA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32), align 8, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s32) from %ir.in.byref, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD2]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load i32, ptr addrspace(4) %in.byref
store volatile i32 %in, ptr addrspace(1) %out, align 4
store volatile i32 %after.offset, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @byref_natural_align_constant_v16i32_arg(ptr addrspace(1) nocapture %out, i8, ptr addrspace(4) byref(<16 x i32>) align(64) %in.byref, i32 %after.offset) {
; HSA-VI-LABEL: name: byref_natural_align_constant_v16i32_arg
; HSA-VI: bb.1 (%ir-block.1):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 64
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 128
; HSA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32), align 16, addrspace 4)
; HSA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<16 x s32>) from %ir.in.byref, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD2]](<16 x s32>), [[LOAD]](p1) :: (volatile store (<16 x s32>) into %ir.out, align 4, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_natural_align_constant_v16i32_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.1):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 100
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 164
; LEGACY-MESA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<16 x s32>) from %ir.in.byref, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD2]](<16 x s32>), [[LOAD]](p1) :: (volatile store (<16 x s32>) into %ir.out, align 4, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load <16 x i32>, ptr addrspace(4) %in.byref
store volatile <16 x i32> %in, ptr addrspace(1) %out, align 4
store volatile i32 %after.offset, ptr addrspace(1) %out, align 4
ret void
}
; Also accept byref kernel arguments with other global address spaces.
define amdgpu_kernel void @byref_global_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(1) byref(i32) align(4) %in.byref) {
; HSA-VI-LABEL: name: byref_global_i32_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[ADDRSPACE_CAST:%[0-9]+]]:_(p1) = G_ADDRSPACE_CAST [[PTR_ADD1]](p4)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[ADDRSPACE_CAST]](p1) :: (dereferenceable "amdgpu-noclobber" load (s32) from %ir.in.byref, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_global_i32_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[ADDRSPACE_CAST:%[0-9]+]]:_(p1) = G_ADDRSPACE_CAST [[PTR_ADD1]](p4)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[ADDRSPACE_CAST]](p1) :: (dereferenceable "amdgpu-noclobber" load (s32) from %ir.in.byref, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load i32, ptr addrspace(1) %in.byref
store i32 %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @byref_flat_i32_arg(ptr addrspace(1) nocapture %out, ptr byref(i32) align(4) %in.byref) {
; HSA-VI-LABEL: name: byref_flat_i32_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[ADDRSPACE_CAST:%[0-9]+]]:_(p0) = G_ADDRSPACE_CAST [[PTR_ADD1]](p4)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[ADDRSPACE_CAST]](p0) :: (dereferenceable load (s32) from %ir.in.byref)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_flat_i32_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[ADDRSPACE_CAST:%[0-9]+]]:_(p0) = G_ADDRSPACE_CAST [[PTR_ADD1]](p4)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[ADDRSPACE_CAST]](p0) :: (dereferenceable load (s32) from %ir.in.byref)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load i32, ptr %in.byref
store i32 %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @byref_constant_32bit_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(6) byref(i32) align(4) %in.byref) {
; HSA-VI-LABEL: name: byref_constant_32bit_i32_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[ADDRSPACE_CAST:%[0-9]+]]:_(p6) = G_ADDRSPACE_CAST [[PTR_ADD1]](p4)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[ADDRSPACE_CAST]](p6) :: (dereferenceable invariant load (s32) from %ir.in.byref, addrspace 6)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_constant_32bit_i32_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[ADDRSPACE_CAST:%[0-9]+]]:_(p6) = G_ADDRSPACE_CAST [[PTR_ADD1]](p4)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[ADDRSPACE_CAST]](p6) :: (dereferenceable invariant load (s32) from %ir.in.byref, addrspace 6)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load i32, ptr addrspace(6) %in.byref
store i32 %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @byref_unknown_as_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(999) byref(i32) align(4) %in.byref) {
; HSA-VI-LABEL: name: byref_unknown_as_i32_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[ADDRSPACE_CAST:%[0-9]+]]:_(p999) = G_ADDRSPACE_CAST [[PTR_ADD1]](p4)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[ADDRSPACE_CAST]](p999) :: (dereferenceable load (s32) from %ir.in.byref, addrspace 999)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_unknown_as_i32_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[ADDRSPACE_CAST:%[0-9]+]]:_(p999) = G_ADDRSPACE_CAST [[PTR_ADD1]](p4)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[ADDRSPACE_CAST]](p999) :: (dereferenceable load (s32) from %ir.in.byref, addrspace 999)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load i32, ptr addrspace(999) %in.byref
store i32 %in, ptr addrspace(1) %out, align 4
ret void
}
; Invalid, but should not crash.
define amdgpu_kernel void @byref_local_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(3) byref(i32) align(4) %in.byref) {
; HSA-VI-LABEL: name: byref_local_i32_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[ADDRSPACE_CAST:%[0-9]+]]:_(p3) = G_ADDRSPACE_CAST [[PTR_ADD1]](p4)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[ADDRSPACE_CAST]](p3) :: (dereferenceable load (s32) from %ir.in.byref, addrspace 3)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_local_i32_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[ADDRSPACE_CAST:%[0-9]+]]:_(p3) = G_ADDRSPACE_CAST [[PTR_ADD1]](p4)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[ADDRSPACE_CAST]](p3) :: (dereferenceable load (s32) from %ir.in.byref, addrspace 3)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load i32, ptr addrspace(3) %in.byref
store i32 %in, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @multi_byref_constant_i32_arg(ptr addrspace(1) nocapture %out, ptr addrspace(4) byref(i32) align(4) %in0.byref, ptr addrspace(4) byref(i32) align(4) %in1.byref, i32 %after.offset) {
; HSA-VI-LABEL: name: multi_byref_constant_i32_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 12
; HSA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; HSA-VI-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD3:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C3]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p4) :: (dereferenceable invariant load (s32), align 16, addrspace 4)
; HSA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s32) from %ir.in0.byref, addrspace 4)
; HSA-VI-NEXT: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32) from %ir.in1.byref, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD2]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD3]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: multi_byref_constant_i32_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p1) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p1), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 44
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 48
; LEGACY-MESA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C2]](s64)
; LEGACY-MESA-VI-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD3:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C3]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD3]](p4) :: (dereferenceable invariant load (s32), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[LOAD2:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (s32) from %ir.in0.byref, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[LOAD3:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD2]](p4) :: (dereferenceable invariant load (s32) from %ir.in1.byref, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD2]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD3]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](s32), [[LOAD]](p1) :: (volatile store (s32) into %ir.out, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in0 = load i32, ptr addrspace(4) %in0.byref
%in1 = load i32, ptr addrspace(4) %in1.byref
store volatile i32 %in0, ptr addrspace(1) %out, align 4
store volatile i32 %in1, ptr addrspace(1) %out, align 4
store volatile i32 %after.offset, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @byref_constant_i32_arg_offset0(ptr addrspace(4) byref(i32) align(4) %in.byref) {
; HSA-VI-LABEL: name: byref_constant_i32_arg_offset0
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32) from %ir.in.byref, addrspace 4)
; HSA-VI-NEXT: G_STORE [[LOAD]](s32), [[DEF]](p1) :: (store (s32) into `ptr addrspace(1) poison`, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: byref_constant_i32_arg_offset0
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (s32) from %ir.in.byref, addrspace 4)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD]](s32), [[DEF]](p1) :: (store (s32) into `ptr addrspace(1) poison`, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
%in = load i32, ptr addrspace(4) %in.byref
store i32 %in, ptr addrspace(1) poison, align 4
ret void
}
define amdgpu_kernel void @p3i8_arg(ptr addrspace(3) %arg) nounwind {
; HSA-VI-LABEL: name: p3i8_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p3), align 16, addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s8) = G_CONSTANT i8 9
; HSA-VI-NEXT: G_STORE [[C1]](s8), [[LOAD]](p3) :: (store (s8) into %ir.arg, align 4, addrspace 3)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: p3i8_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(p3) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (p3), addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s8) = G_CONSTANT i8 9
; LEGACY-MESA-VI-NEXT: G_STORE [[C1]](s8), [[LOAD]](p3) :: (store (s8) into %ir.arg, align 4, addrspace 3)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
store i8 9, ptr addrspace(3) %arg, align 4
ret void
}
define amdgpu_kernel void @p1i8_arg(ptr addrspace(1) %arg) nounwind {
; HSA-VI-LABEL: name: p1i8_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 9
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(p3) = G_CONSTANT i32 0
; HSA-VI-NEXT: G_STORE [[C]](s8), [[C1]](p3) :: (store (s8) into `ptr addrspace(3) null`, addrspace 3)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: p1i8_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 9
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(p3) = G_CONSTANT i32 0
; LEGACY-MESA-VI-NEXT: G_STORE [[C]](s8), [[C1]](p3) :: (store (s8) into `ptr addrspace(3) null`, addrspace 3)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
store i8 9, ptr addrspace(3) null
ret void
}
define amdgpu_kernel void @v2p1i8_arg(<2 x ptr addrspace(1)> %arg) nounwind {
; HSA-VI-LABEL: name: v2p1i8_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(<2 x p1>) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (<2 x p1>), addrspace 4)
; HSA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; HSA-VI-NEXT: G_STORE [[LOAD]](<2 x p1>), [[DEF]](p1) :: (store (<2 x p1>) into `ptr addrspace(1) poison`, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v2p1i8_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(<2 x p1>) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (<2 x p1>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD]](<2 x p1>), [[DEF]](p1) :: (store (<2 x p1>) into `ptr addrspace(1) poison`, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
store <2 x ptr addrspace(1)> %arg, ptr addrspace(1) poison
ret void
}
define amdgpu_kernel void @v2p3i8_arg(<2 x ptr addrspace(3)> %arg) nounwind {
; HSA-VI-LABEL: name: v2p3i8_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(<2 x p3>) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (<2 x p3>), align 16, addrspace 4)
; HSA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; HSA-VI-NEXT: G_STORE [[LOAD]](<2 x p3>), [[DEF]](p1) :: (store (<2 x p3>) into `ptr addrspace(1) poison`, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v2p3i8_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(<2 x p3>) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (<2 x p3>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD]](<2 x p3>), [[DEF]](p1) :: (store (<2 x p3>) into `ptr addrspace(1) poison`, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
store <2 x ptr addrspace(3)> %arg, ptr addrspace(1) poison
ret void
}
define amdgpu_kernel void @v2p1i8_in_struct_arg({ <2 x ptr addrspace(1)>, <2 x ptr addrspace(3)> } %arg) nounwind {
; HSA-VI-LABEL: name: v2p1i8_in_struct_arg
; HSA-VI: bb.1 (%ir-block.0):
; HSA-VI-NEXT: liveins: $sgpr8_sgpr9
; HSA-VI-NEXT: {{ $}}
; HSA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr8_sgpr9
; HSA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; HSA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; HSA-VI-NEXT: [[LOAD:%[0-9]+]]:_(<2 x p1>) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (<2 x s64>), addrspace 4)
; HSA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; HSA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<2 x p3>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<2 x s32>), align 16, addrspace 4)
; HSA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; HSA-VI-NEXT: G_STORE [[LOAD]](<2 x p1>), [[DEF]](p1) :: (store (<2 x p1>) into `ptr addrspace(1) poison`, addrspace 1)
; HSA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; HSA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p1) = nuw inbounds G_PTR_ADD [[DEF]], [[C2]](s64)
; HSA-VI-NEXT: G_STORE [[LOAD1]](<2 x p3>), [[PTR_ADD2]](p1) :: (store (<2 x p3>) into `ptr addrspace(1) poison` + 16, align 16, addrspace 1)
; HSA-VI-NEXT: S_ENDPGM 0
;
; LEGACY-MESA-VI-LABEL: name: v2p1i8_in_struct_arg
; LEGACY-MESA-VI: bb.1 (%ir-block.0):
; LEGACY-MESA-VI-NEXT: liveins: $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: {{ $}}
; LEGACY-MESA-VI-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY $sgpr4_sgpr5
; LEGACY-MESA-VI-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; LEGACY-MESA-VI-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD:%[0-9]+]]:_(<2 x p1>) = G_LOAD [[PTR_ADD]](p4) :: (dereferenceable invariant load (<2 x s64>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 52
; LEGACY-MESA-VI-NEXT: [[PTR_ADD1:%[0-9]+]]:_(p4) = G_PTR_ADD [[COPY]], [[C1]](s64)
; LEGACY-MESA-VI-NEXT: [[LOAD1:%[0-9]+]]:_(<2 x p3>) = G_LOAD [[PTR_ADD1]](p4) :: (dereferenceable invariant load (<2 x s32>), align 4, addrspace 4)
; LEGACY-MESA-VI-NEXT: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD]](<2 x p1>), [[DEF]](p1) :: (store (<2 x p1>) into `ptr addrspace(1) poison`, addrspace 1)
; LEGACY-MESA-VI-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; LEGACY-MESA-VI-NEXT: [[PTR_ADD2:%[0-9]+]]:_(p1) = nuw inbounds G_PTR_ADD [[DEF]], [[C2]](s64)
; LEGACY-MESA-VI-NEXT: G_STORE [[LOAD1]](<2 x p3>), [[PTR_ADD2]](p1) :: (store (<2 x p3>) into `ptr addrspace(1) poison` + 16, align 16, addrspace 1)
; LEGACY-MESA-VI-NEXT: S_ENDPGM 0
store { <2 x ptr addrspace(1)>, <2 x ptr addrspace(3)> } %arg, ptr addrspace(1) poison
ret void
}
Reference in New Issue View Git Blame Copy Permalink