fc0653f31c
Recent upstream trends have moved away from explicitly using `-verify-machineinstrs`, as it's already covered by the expensive checks. This PR removes almost all `-verify-machineinstrs` from tests in `llvm/test/CodeGen/AMDGPU/*.ll`, leaving only those tests where its removal currently causes failures.
494 lines
18 KiB
LLVM
494 lines
18 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=kaveri < %s | FileCheck -check-prefixes=GCN,CI %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s
|
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s
|
|
|
|
define i8 @atomic_load_monotonic_i8(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_i8:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_u8 v0, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_i8:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_u8 v0, v0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: atomic_load_monotonic_i8:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: ds_load_u8_d16 v0, v0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: atomic_load_monotonic_i8:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: ds_load_u8 v0, v0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%load = load atomic i8, ptr addrspace(3) %ptr monotonic, align 1
|
|
ret i8 %load
|
|
}
|
|
|
|
define i8 @atomic_load_monotonic_i8_offset(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_i8_offset:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_u8 v0, v0 offset:16
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_i8_offset:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_u8 v0, v0 offset:16
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: atomic_load_monotonic_i8_offset:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: ds_load_u8_d16 v0, v0 offset:16
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: atomic_load_monotonic_i8_offset:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: ds_load_u8 v0, v0 offset:16
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%gep = getelementptr inbounds i8, ptr addrspace(3) %ptr, i8 16
|
|
%load = load atomic i8, ptr addrspace(3) %gep monotonic, align 1
|
|
ret i8 %load
|
|
}
|
|
|
|
define i16 @atomic_load_monotonic_i16(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_i16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_u16 v0, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_i16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_u16 v0, v0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: atomic_load_monotonic_i16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: ds_load_u16_d16 v0, v0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: atomic_load_monotonic_i16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%load = load atomic i16, ptr addrspace(3) %ptr monotonic, align 2
|
|
ret i16 %load
|
|
}
|
|
|
|
define i16 @atomic_load_monotonic_i16_offset(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_i16_offset:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_u16 v0, v0 offset:32
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_i16_offset:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_u16 v0, v0 offset:32
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: atomic_load_monotonic_i16_offset:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: ds_load_u16_d16 v0, v0 offset:32
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: atomic_load_monotonic_i16_offset:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 offset:32
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%gep = getelementptr inbounds i16, ptr addrspace(3) %ptr, i16 16
|
|
%load = load atomic i16, ptr addrspace(3) %gep monotonic, align 2
|
|
ret i16 %load
|
|
}
|
|
|
|
define i32 @atomic_load_monotonic_i32(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_i32:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_b32 v0, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_i32:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_b32 v0, v0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: atomic_load_monotonic_i32:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: ds_load_b32 v0, v0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%load = load atomic i32, ptr addrspace(3) %ptr monotonic, align 4
|
|
ret i32 %load
|
|
}
|
|
|
|
define i32 @atomic_load_monotonic_i32_offset(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_i32_offset:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_b32 v0, v0 offset:64
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_i32_offset:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_b32 v0, v0 offset:64
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: atomic_load_monotonic_i32_offset:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: ds_load_b32 v0, v0 offset:64
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%gep = getelementptr inbounds i32, ptr addrspace(3) %ptr, i32 16
|
|
%load = load atomic i32, ptr addrspace(3) %gep monotonic, align 4
|
|
ret i32 %load
|
|
}
|
|
|
|
define i64 @atomic_load_monotonic_i64(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_i64:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_b64 v[0:1], v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_i64:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_b64 v[0:1], v0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: atomic_load_monotonic_i64:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: ds_load_b64 v[0:1], v0
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%load = load atomic i64, ptr addrspace(3) %ptr monotonic, align 8
|
|
ret i64 %load
|
|
}
|
|
|
|
define i64 @atomic_load_monotonic_i64_offset(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_i64_offset:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_b64 v[0:1], v0 offset:128
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_i64_offset:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_b64 v[0:1], v0 offset:128
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: atomic_load_monotonic_i64_offset:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: ds_load_b64 v[0:1], v0 offset:128
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%gep = getelementptr inbounds i64, ptr addrspace(3) %ptr, i32 16
|
|
%load = load atomic i64, ptr addrspace(3) %gep monotonic, align 8
|
|
ret i64 %load
|
|
}
|
|
|
|
define float @atomic_load_monotonic_f32_offset(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_f32_offset:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_b32 v0, v0 offset:64
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_f32_offset:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_b32 v0, v0 offset:64
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: atomic_load_monotonic_f32_offset:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: ds_load_b32 v0, v0 offset:64
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%gep = getelementptr inbounds float, ptr addrspace(3) %ptr, i32 16
|
|
%load = load atomic float, ptr addrspace(3) %gep monotonic, align 4
|
|
ret float %load
|
|
}
|
|
|
|
define double @atomic_load_monotonic_f64_offset(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_f64_offset:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_b64 v[0:1], v0 offset:128
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_f64_offset:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_b64 v[0:1], v0 offset:128
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: atomic_load_monotonic_f64_offset:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: ds_load_b64 v[0:1], v0 offset:128
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%gep = getelementptr inbounds double, ptr addrspace(3) %ptr, i32 16
|
|
%load = load atomic double, ptr addrspace(3) %gep monotonic, align 8
|
|
ret double %load
|
|
}
|
|
|
|
define ptr @atomic_load_monotonic_p0i8_offset(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_p0i8_offset:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_b64 v[0:1], v0 offset:128
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_p0i8_offset:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_b64 v[0:1], v0 offset:128
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: atomic_load_monotonic_p0i8_offset:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: ds_load_b64 v[0:1], v0 offset:128
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%gep = getelementptr inbounds ptr, ptr addrspace(3) %ptr, i32 16
|
|
%load = load atomic ptr, ptr addrspace(3) %gep monotonic, align 8
|
|
ret ptr %load
|
|
}
|
|
|
|
define ptr addrspace(3) @atomic_load_monotonic_p3i8_offset(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_p3i8_offset:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_b32 v0, v0 offset:64
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_p3i8_offset:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_b32 v0, v0 offset:64
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-LABEL: atomic_load_monotonic_p3i8_offset:
|
|
; GFX11: ; %bb.0:
|
|
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-NEXT: ds_load_b32 v0, v0 offset:64
|
|
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-NEXT: s_setpc_b64 s[30:31]
|
|
%gep = getelementptr inbounds ptr addrspace(3), ptr addrspace(3) %ptr, i32 16
|
|
%load = load atomic ptr addrspace(3), ptr addrspace(3) %gep monotonic, align 4
|
|
ret ptr addrspace(3) %load
|
|
}
|
|
|
|
define i16 @atomic_load_monotonic_f16(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_f16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_u16 v0, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_f16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_u16 v0, v0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: atomic_load_monotonic_f16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: ds_load_u16_d16 v0, v0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: atomic_load_monotonic_f16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%load = load atomic half, ptr addrspace(3) %ptr monotonic, align 2
|
|
%ret = bitcast half %load to i16
|
|
ret i16 %ret
|
|
}
|
|
|
|
define i16 @atomic_load_monotonic_f16_offset(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_f16_offset:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_u16 v0, v0 offset:32
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_f16_offset:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_u16 v0, v0 offset:32
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: atomic_load_monotonic_f16_offset:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: ds_load_u16_d16 v0, v0 offset:32
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: atomic_load_monotonic_f16_offset:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 offset:32
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%gep = getelementptr inbounds half, ptr addrspace(3) %ptr, i32 16
|
|
%load = load atomic half, ptr addrspace(3) %gep monotonic, align 2
|
|
%ret = bitcast half %load to i16
|
|
ret i16 %ret
|
|
}
|
|
|
|
define i16 @atomic_load_monotonic_bf16(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_bf16:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_u16 v0, v0
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_bf16:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_u16 v0, v0
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: atomic_load_monotonic_bf16:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: ds_load_u16_d16 v0, v0
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: atomic_load_monotonic_bf16:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%load = load atomic bfloat, ptr addrspace(3) %ptr monotonic, align 2
|
|
%ret = bitcast bfloat %load to i16
|
|
ret i16 %ret
|
|
}
|
|
|
|
define i16 @atomic_load_monotonic_bf16_offset(ptr addrspace(3) %ptr) {
|
|
; CI-LABEL: atomic_load_monotonic_bf16_offset:
|
|
; CI: ; %bb.0:
|
|
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; CI-NEXT: s_mov_b32 m0, -1
|
|
; CI-NEXT: ds_read_u16 v0, v0 offset:32
|
|
; CI-NEXT: s_waitcnt lgkmcnt(0)
|
|
; CI-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX9-LABEL: atomic_load_monotonic_bf16_offset:
|
|
; GFX9: ; %bb.0:
|
|
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX9-NEXT: ds_read_u16 v0, v0 offset:32
|
|
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-TRUE16-LABEL: atomic_load_monotonic_bf16_offset:
|
|
; GFX11-TRUE16: ; %bb.0:
|
|
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: ds_load_u16_d16 v0, v0 offset:32
|
|
; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
|
|
;
|
|
; GFX11-FAKE16-LABEL: atomic_load_monotonic_bf16_offset:
|
|
; GFX11-FAKE16: ; %bb.0:
|
|
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 offset:32
|
|
; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
|
|
%gep = getelementptr inbounds bfloat, ptr addrspace(3) %ptr, i32 16
|
|
%load = load atomic bfloat, ptr addrspace(3) %gep monotonic, align 2
|
|
%ret = bitcast bfloat %load to i16
|
|
ret i16 %ret
|
|
}
|
|
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
|
|
; GCN: {{.*}}
|