caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
8fcb1263f42657ecbc355beff12500dfbcddee17
clang-p2996/llvm/test/CodeGen/AMDGPU/sra.ll
Lucas Ramirez 6206f5444f [AMDGPU] Occupancy w.r.t. workgroup size range is also a range (#123748) 
Occupancy (i.e., the number of waves per EU) depends, in addition to
register usage, on per-workgroup LDS usage as well as on the range of
possible workgroup sizes. Mirroring the latter, occupancy should
therefore be expressed as a range since different group sizes generally
yield different achievable occupancies.

`getOccupancyWithLocalMemSize` currently returns a scalar occupancy
based on the maximum workgroup size and LDS usage. With respect to the
workgroup size range, this scalar can be the minimum, the maximum, or
neither of the two of the range of achievable occupancies. This commit
fixes the function by making it compute and return the range of
achievable occupancies w.r.t. workgroup size and LDS usage; it also
renames it to `getOccupancyWithWorkGroupSizes` since it is the range of
workgroup sizes that produces the range of achievable occupancies.

Computing the achievable occupancy range is surprisingly involved.
Minimum/maximum workgroup sizes do not necessarily yield maximum/minimum
occupancies i.e., sometimes workgroup sizes inside the range yield the
occupancy bounds. The implementation finds these sizes in constant time;
heavy documentation explains the rationale behind the sometimes
relatively obscure calculations.

As a justifying example, consider a target with 10 waves / EU, 4 EUs/CU,
64-wide waves. Also consider a function with no LDS usage and a flat
workgroup size range of [513,1024].

- A group of 513 items requires 9 waves per group. Only 4 groups made up
of 9 waves each can fit fully on a CU at any given time, for a total of
36 waves on the CU, or 9 per EU. However, filling as much as possible
the remaining 40-36=4 wave slots without decreasing the number of groups
reveals that a larger group of 640 items yields 40 waves on the CU, or
10 per EU.
- Similarly, a group of 1024 items requires 16 waves per group. Only 2
groups made up of 16 waves each can fit fully on a CU ay any given time,
for a total of 32 waves on the CU, or 8 per EU. However, removing as
many waves as possible from the groups without being able to fit another
equal-sized group on the CU reveals that a smaller group of 896 items
yields 28 waves on the CU, or 7 per EU.

Therefore the achievable occupancy range for this function is not [8,9]
as the group size bounds directly yield, but [7,10].

Naturally this change causes a lot of test churn as instruction
scheduling is driven by achievable occupancy estimates. In most unit
tests the flat workgroup size range is the default [1,1024] which,
ignoring potential LDS limitations, would previously produce a scalar
occupancy of 8 (derived from 1024) on a lot of targets, whereas we now
consider the maximum occupancy to be 10 in such cases. Most tests are
updated automatically and checked manually for sanity. I also manually
changed some non-automatically generated assertions when necessary.

Fixes #118220.
2025-01-23 16:07:57 +01:00

957 lines
36 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn-- -mcpu=verde -verify-machineinstrs < %s | FileCheck %s -check-prefixes=SI
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn-- -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck %s -check-prefixes=VI
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=r600-- -mcpu=redwood -verify-machineinstrs < %s | FileCheck %s -check-prefixes=EG
declare i32 @llvm.amdgcn.workitem.id.x() #0
define amdgpu_kernel void @ashr_v2i32(ptr addrspace(1) %out, ptr addrspace(1) %in) {
; SI-LABEL: ashr_v2i32:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, -1
; SI-NEXT: s_mov_b32 s10, s6
; SI-NEXT: s_mov_b32 s11, s7
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b32 s8, s2
; SI-NEXT: s_mov_b32 s9, s3
; SI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
; SI-NEXT: s_mov_b32 s4, s0
; SI-NEXT: s_mov_b32 s5, s1
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_ashr_i32_e32 v1, v1, v3
; SI-NEXT: v_ashr_i32_e32 v0, v0, v2
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: ashr_v2i32:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT: s_mov_b32 s7, 0xf000
; VI-NEXT: s_mov_b32 s6, -1
; VI-NEXT: s_mov_b32 s10, s6
; VI-NEXT: s_mov_b32 s11, s7
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_mov_b32 s8, s2
; VI-NEXT: s_mov_b32 s9, s3
; VI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
; VI-NEXT: s_mov_b32 s4, s0
; VI-NEXT: s_mov_b32 s5, s1
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_ashrrev_i32_e32 v1, v3, v1
; VI-NEXT: v_ashrrev_i32_e32 v0, v2, v0
; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; VI-NEXT: s_endpgm
;
; EG-LABEL: ashr_v2i32:
; EG: ; %bb.0:
; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]
; EG-NEXT: TEX 0 @6
; EG-NEXT: ALU 3, @9, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: Fetch clause starting at 6:
; EG-NEXT: VTX_READ_128 T0.XYZW, T0.X, 0, #1
; EG-NEXT: ALU clause starting at 8:
; EG-NEXT: MOV * T0.X, KC0[2].Z,
; EG-NEXT: ALU clause starting at 9:
; EG-NEXT: ASHR * T0.Y, T0.Y, T0.W,
; EG-NEXT: ASHR T0.X, T0.X, T0.Z,
; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
%b_ptr = getelementptr <2 x i32>, ptr addrspace(1) %in, i32 1
%a = load <2 x i32>, ptr addrspace(1) %in
%b = load <2 x i32>, ptr addrspace(1) %b_ptr
%result = ashr <2 x i32> %a, %b
store <2 x i32> %result, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @ashr_v4i32(ptr addrspace(1) %out, ptr addrspace(1) %in) {
; SI-LABEL: ashr_v4i32:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, -1
; SI-NEXT: s_mov_b32 s10, s6
; SI-NEXT: s_mov_b32 s11, s7
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b32 s8, s2
; SI-NEXT: s_mov_b32 s9, s3
; SI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
; SI-NEXT: buffer_load_dwordx4 v[4:7], off, s[8:11], 0 offset:16
; SI-NEXT: s_mov_b32 s4, s0
; SI-NEXT: s_mov_b32 s5, s1
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_ashr_i32_e32 v3, v3, v7
; SI-NEXT: v_ashr_i32_e32 v2, v2, v6
; SI-NEXT: v_ashr_i32_e32 v1, v1, v5
; SI-NEXT: v_ashr_i32_e32 v0, v0, v4
; SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: ashr_v4i32:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT: s_mov_b32 s7, 0xf000
; VI-NEXT: s_mov_b32 s6, -1
; VI-NEXT: s_mov_b32 s10, s6
; VI-NEXT: s_mov_b32 s11, s7
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_mov_b32 s8, s2
; VI-NEXT: s_mov_b32 s9, s3
; VI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
; VI-NEXT: buffer_load_dwordx4 v[4:7], off, s[8:11], 0 offset:16
; VI-NEXT: s_mov_b32 s4, s0
; VI-NEXT: s_mov_b32 s5, s1
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_ashrrev_i32_e32 v3, v7, v3
; VI-NEXT: v_ashrrev_i32_e32 v2, v6, v2
; VI-NEXT: v_ashrrev_i32_e32 v1, v5, v1
; VI-NEXT: v_ashrrev_i32_e32 v0, v4, v0
; VI-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0
; VI-NEXT: s_endpgm
;
; EG-LABEL: ashr_v4i32:
; EG: ; %bb.0:
; EG-NEXT: ALU 0, @10, KC0[CB0:0-32], KC1[]
; EG-NEXT: TEX 1 @6
; EG-NEXT: ALU 5, @11, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: Fetch clause starting at 6:
; EG-NEXT: VTX_READ_128 T1.XYZW, T0.X, 16, #1
; EG-NEXT: VTX_READ_128 T0.XYZW, T0.X, 0, #1
; EG-NEXT: ALU clause starting at 10:
; EG-NEXT: MOV * T0.X, KC0[2].Z,
; EG-NEXT: ALU clause starting at 11:
; EG-NEXT: ASHR * T0.W, T0.W, T1.W,
; EG-NEXT: ASHR * T0.Z, T0.Z, T1.Z,
; EG-NEXT: ASHR * T0.Y, T0.Y, T1.Y,
; EG-NEXT: ASHR T0.X, T0.X, T1.X,
; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
%b_ptr = getelementptr <4 x i32>, ptr addrspace(1) %in, i32 1
%a = load <4 x i32>, ptr addrspace(1) %in
%b = load <4 x i32>, ptr addrspace(1) %b_ptr
%result = ashr <4 x i32> %a, %b
store <4 x i32> %result, ptr addrspace(1) %out
ret void
}
; FIXME: The ashr operation is uniform, but because its operands come from a
; global load we end up with the vector instructions rather than scalar.
define amdgpu_kernel void @ashr_v2i16(ptr addrspace(1) %out, ptr addrspace(1) %in) {
; SI-LABEL: ashr_v2i16:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, -1
; SI-NEXT: s_mov_b32 s10, s6
; SI-NEXT: s_mov_b32 s11, s7
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b32 s8, s2
; SI-NEXT: s_mov_b32 s9, s3
; SI-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT: s_mov_b32 s4, s0
; SI-NEXT: s_mov_b32 s5, s1
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_readfirstlane_b32 s0, v0
; SI-NEXT: v_readfirstlane_b32 s1, v1
; SI-NEXT: s_sext_i32_i16 s2, s0
; SI-NEXT: s_ashr_i32 s0, s0, 16
; SI-NEXT: s_lshr_b32 s3, s1, 16
; SI-NEXT: s_ashr_i32 s0, s0, s3
; SI-NEXT: s_ashr_i32 s1, s2, s1
; SI-NEXT: s_lshl_b32 s0, s0, 16
; SI-NEXT: s_and_b32 s1, s1, 0xffff
; SI-NEXT: s_or_b32 s0, s1, s0
; SI-NEXT: v_mov_b32_e32 v0, s0
; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: ashr_v2i16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT: s_mov_b32 s7, 0xf000
; VI-NEXT: s_mov_b32 s6, -1
; VI-NEXT: s_mov_b32 s10, s6
; VI-NEXT: s_mov_b32 s11, s7
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_mov_b32 s8, s2
; VI-NEXT: s_mov_b32 s9, s3
; VI-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0
; VI-NEXT: s_mov_b32 s4, s0
; VI-NEXT: s_mov_b32 s5, s1
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_readfirstlane_b32 s0, v1
; VI-NEXT: v_readfirstlane_b32 s1, v0
; VI-NEXT: s_ashr_i32 s2, s1, 16
; VI-NEXT: s_sext_i32_i16 s1, s1
; VI-NEXT: s_ashr_i32 s3, s0, 16
; VI-NEXT: s_sext_i32_i16 s0, s0
; VI-NEXT: s_ashr_i32 s0, s1, s0
; VI-NEXT: s_ashr_i32 s1, s2, s3
; VI-NEXT: s_lshl_b32 s1, s1, 16
; VI-NEXT: s_and_b32 s0, s0, 0xffff
; VI-NEXT: s_or_b32 s0, s0, s1
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: buffer_store_dword v0, off, s[4:7], 0
; VI-NEXT: s_endpgm
;
; EG-LABEL: ashr_v2i16:
; EG: ; %bb.0:
; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]
; EG-NEXT: TEX 0 @6
; EG-NEXT: ALU 14, @9, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T6.X, T7.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: Fetch clause starting at 6:
; EG-NEXT: VTX_READ_64 T6.XY, T6.X, 0, #1
; EG-NEXT: ALU clause starting at 8:
; EG-NEXT: MOV * T6.X, KC0[2].Z,
; EG-NEXT: ALU clause starting at 9:
; EG-NEXT: LSHR * T0.W, T6.X, literal.x,
; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00)
; EG-NEXT: BFE_INT T0.Y, PV.W, 0.0, literal.x,
; EG-NEXT: LSHR T0.Z, T6.Y, literal.x,
; EG-NEXT: BFE_INT T0.W, T6.X, 0.0, literal.x,
; EG-NEXT: AND_INT * T1.W, T6.Y, literal.y,
; EG-NEXT: 16(2.242078e-44), 65535(9.183409e-41)
; EG-NEXT: ASHR T0.W, PV.W, PS,
; EG-NEXT: ASHR * T1.W, PV.Y, PV.Z,
; EG-NEXT: LSHL T1.W, PS, literal.x,
; EG-NEXT: AND_INT * T0.W, PV.W, literal.y,
; EG-NEXT: 16(2.242078e-44), 65535(9.183409e-41)
; EG-NEXT: OR_INT T6.X, PS, PV.W,
; EG-NEXT: LSHR * T7.X, KC0[2].Y, literal.x,
; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
%b_ptr = getelementptr <2 x i16>, ptr addrspace(1) %in, i16 1
%a = load <2 x i16>, ptr addrspace(1) %in
%b = load <2 x i16>, ptr addrspace(1) %b_ptr
%result = ashr <2 x i16> %a, %b
store <2 x i16> %result, ptr addrspace(1) %out
ret void
}
; FIXME: The ashr operation is uniform, but because its operands come from a
; global load we end up with the vector instructions rather than scalar.
define amdgpu_kernel void @ashr_v4i16(ptr addrspace(1) %out, ptr addrspace(1) %in) {
; SI-LABEL: ashr_v4i16:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, -1
; SI-NEXT: s_mov_b32 s10, s6
; SI-NEXT: s_mov_b32 s11, s7
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b32 s8, s2
; SI-NEXT: s_mov_b32 s9, s3
; SI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
; SI-NEXT: s_mov_b32 s4, s0
; SI-NEXT: s_mov_b32 s5, s1
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_readfirstlane_b32 s0, v3
; SI-NEXT: v_readfirstlane_b32 s1, v2
; SI-NEXT: v_readfirstlane_b32 s2, v1
; SI-NEXT: v_readfirstlane_b32 s3, v0
; SI-NEXT: s_sext_i32_i16 s8, s3
; SI-NEXT: s_ashr_i32 s3, s3, 16
; SI-NEXT: s_sext_i32_i16 s9, s2
; SI-NEXT: s_ashr_i32 s2, s2, 16
; SI-NEXT: s_lshr_b32 s10, s1, 16
; SI-NEXT: s_lshr_b32 s11, s0, 16
; SI-NEXT: s_ashr_i32 s2, s2, s11
; SI-NEXT: s_ashr_i32 s0, s9, s0
; SI-NEXT: s_ashr_i32 s3, s3, s10
; SI-NEXT: s_ashr_i32 s1, s8, s1
; SI-NEXT: s_lshl_b32 s2, s2, 16
; SI-NEXT: s_and_b32 s0, s0, 0xffff
; SI-NEXT: s_lshl_b32 s3, s3, 16
; SI-NEXT: s_and_b32 s1, s1, 0xffff
; SI-NEXT: s_or_b32 s0, s0, s2
; SI-NEXT: s_or_b32 s1, s1, s3
; SI-NEXT: v_mov_b32_e32 v0, s1
; SI-NEXT: v_mov_b32_e32 v1, s0
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: ashr_v4i16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT: s_mov_b32 s7, 0xf000
; VI-NEXT: s_mov_b32 s6, -1
; VI-NEXT: s_mov_b32 s10, s6
; VI-NEXT: s_mov_b32 s11, s7
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_mov_b32 s8, s2
; VI-NEXT: s_mov_b32 s9, s3
; VI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
; VI-NEXT: s_mov_b32 s4, s0
; VI-NEXT: s_mov_b32 s5, s1
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_readfirstlane_b32 s0, v2
; VI-NEXT: v_readfirstlane_b32 s1, v3
; VI-NEXT: v_readfirstlane_b32 s2, v0
; VI-NEXT: v_readfirstlane_b32 s3, v1
; VI-NEXT: s_ashr_i32 s8, s3, 16
; VI-NEXT: s_sext_i32_i16 s3, s3
; VI-NEXT: s_ashr_i32 s9, s2, 16
; VI-NEXT: s_sext_i32_i16 s2, s2
; VI-NEXT: s_ashr_i32 s10, s1, 16
; VI-NEXT: s_sext_i32_i16 s1, s1
; VI-NEXT: s_ashr_i32 s11, s0, 16
; VI-NEXT: s_sext_i32_i16 s0, s0
; VI-NEXT: s_ashr_i32 s0, s2, s0
; VI-NEXT: s_ashr_i32 s2, s9, s11
; VI-NEXT: s_ashr_i32 s1, s3, s1
; VI-NEXT: s_ashr_i32 s3, s8, s10
; VI-NEXT: s_lshl_b32 s3, s3, 16
; VI-NEXT: s_and_b32 s1, s1, 0xffff
; VI-NEXT: s_lshl_b32 s2, s2, 16
; VI-NEXT: s_and_b32 s0, s0, 0xffff
; VI-NEXT: s_or_b32 s1, s1, s3
; VI-NEXT: s_or_b32 s0, s0, s2
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; VI-NEXT: s_endpgm
;
; EG-LABEL: ashr_v4i16:
; EG: ; %bb.0:
; EG-NEXT: ALU 1, @8, KC0[CB0:0-32], KC1[]
; EG-NEXT: TEX 0 @6
; EG-NEXT: ALU 48, @10, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T10.XY, T9.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: Fetch clause starting at 6:
; EG-NEXT: VTX_READ_128 T9.XYZW, T9.X, 0, #1
; EG-NEXT: ALU clause starting at 8:
; EG-NEXT: MOV * T0.Y, T6.X,
; EG-NEXT: MOV * T9.X, KC0[2].Z,
; EG-NEXT: ALU clause starting at 10:
; EG-NEXT: BFE_INT T0.W, T9.X, 0.0, literal.x,
; EG-NEXT: AND_INT * T1.W, T9.Z, literal.y,
; EG-NEXT: 16(2.242078e-44), 65535(9.183409e-41)
; EG-NEXT: ASHR * T0.W, PV.W, PS,
; EG-NEXT: AND_INT T0.W, PV.W, literal.x,
; EG-NEXT: AND_INT * T1.W, T0.Y, literal.y,
; EG-NEXT: 65535(9.183409e-41), -65536(nan)
; EG-NEXT: OR_INT * T0.W, PS, PV.W,
; EG-NEXT: MOV * T6.X, PV.W,
; EG-NEXT: MOV T0.Y, PV.X,
; EG-NEXT: LSHR * T0.W, T9.X, literal.x,
; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00)
; EG-NEXT: BFE_INT T0.W, PV.W, 0.0, literal.x,
; EG-NEXT: LSHR * T1.W, T9.Z, literal.x,
; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00)
; EG-NEXT: ASHR T0.W, PV.W, PS,
; EG-NEXT: AND_INT * T1.W, T0.Y, literal.x,
; EG-NEXT: 65535(9.183409e-41), 0(0.000000e+00)
; EG-NEXT: LSHL * T0.W, PV.W, literal.x,
; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00)
; EG-NEXT: OR_INT * T0.W, T1.W, PV.W,
; EG-NEXT: MOV T6.X, PV.W,
; EG-NEXT: MOV T0.Y, T7.X,
; EG-NEXT: BFE_INT T0.W, T9.Y, 0.0, literal.x,
; EG-NEXT: AND_INT * T1.W, T9.W, literal.y,
; EG-NEXT: 16(2.242078e-44), 65535(9.183409e-41)
; EG-NEXT: ASHR T0.W, PV.W, PS,
; EG-NEXT: AND_INT * T1.W, PV.Y, literal.x,
; EG-NEXT: -65536(nan), 0(0.000000e+00)
; EG-NEXT: AND_INT * T0.W, PV.W, literal.x,
; EG-NEXT: 65535(9.183409e-41), 0(0.000000e+00)
; EG-NEXT: OR_INT * T0.W, T1.W, PV.W,
; EG-NEXT: MOV * T7.X, PV.W,
; EG-NEXT: MOV T0.Y, PV.X,
; EG-NEXT: LSHR * T0.W, T9.Y, literal.x,
; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00)
; EG-NEXT: BFE_INT T0.W, PV.W, 0.0, literal.x,
; EG-NEXT: LSHR * T1.W, T9.W, literal.x,
; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00)
; EG-NEXT: ASHR T0.W, PV.W, PS,
; EG-NEXT: AND_INT * T1.W, T0.Y, literal.x,
; EG-NEXT: 65535(9.183409e-41), 0(0.000000e+00)
; EG-NEXT: LSHL * T0.W, PV.W, literal.x,
; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00)
; EG-NEXT: LSHR T9.X, KC0[2].Y, literal.x,
; EG-NEXT: OR_INT * T10.Y, T1.W, PV.W,
; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
; EG-NEXT: MOV T7.X, PV.Y,
; EG-NEXT: MOV * T10.X, T6.X,
%b_ptr = getelementptr <4 x i16>, ptr addrspace(1) %in, i16 1
%a = load <4 x i16>, ptr addrspace(1) %in
%b = load <4 x i16>, ptr addrspace(1) %b_ptr
%result = ashr <4 x i16> %a, %b
store <4 x i16> %result, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @s_ashr_i64(ptr addrspace(1) %out, i32 %in) {
; SI-LABEL: s_ashr_i64:
; SI: ; %bb.0: ; %entry
; SI-NEXT: s_load_dword s6, s[4:5], 0xb
; SI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_ashr_i32 s7, s6, 31
; SI-NEXT: s_ashr_i64 s[4:5], s[6:7], 8
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: s_ashr_i64:
; VI: ; %bb.0: ; %entry
; VI-NEXT: s_load_dword s6, s[4:5], 0x2c
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-NEXT: s_mov_b32 s3, 0xf000
; VI-NEXT: s_mov_b32 s2, -1
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_ashr_i32 s7, s6, 31
; VI-NEXT: s_ashr_i64 s[4:5], s[6:7], 8
; VI-NEXT: v_mov_b32_e32 v0, s4
; VI-NEXT: v_mov_b32_e32 v1, s5
; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT: s_endpgm
;
; EG-LABEL: s_ashr_i64:
; EG: ; %bb.0: ; %entry
; EG-NEXT: ALU 4, @4, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: ALU clause starting at 4:
; EG-NEXT: ASHR * T0.Y, KC0[2].Z, literal.x,
; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00)
; EG-NEXT: BIT_ALIGN_INT T0.X, PV.Y, KC0[2].Z, literal.x,
; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y,
; EG-NEXT: 8(1.121039e-44), 2(2.802597e-45)
entry:
%in.ext = sext i32 %in to i64
%ashr = ashr i64 %in.ext, 8
store i64 %ashr, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @ashr_i64_2(ptr addrspace(1) %out, ptr addrspace(1) %in) {
; SI-LABEL: ashr_i64_2:
; SI: ; %bb.0: ; %entry
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, -1
; SI-NEXT: s_mov_b32 s10, s6
; SI-NEXT: s_mov_b32 s11, s7
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b32 s8, s2
; SI-NEXT: s_mov_b32 s9, s3
; SI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
; SI-NEXT: s_mov_b32 s4, s0
; SI-NEXT: s_mov_b32 s5, s1
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_ashr_i64 v[0:1], v[0:1], v2
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: ashr_i64_2:
; VI: ; %bb.0: ; %entry
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT: s_mov_b32 s7, 0xf000
; VI-NEXT: s_mov_b32 s6, -1
; VI-NEXT: s_mov_b32 s10, s6
; VI-NEXT: s_mov_b32 s11, s7
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_mov_b32 s8, s2
; VI-NEXT: s_mov_b32 s9, s3
; VI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
; VI-NEXT: s_mov_b32 s4, s0
; VI-NEXT: s_mov_b32 s5, s1
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_ashrrev_i64 v[0:1], v2, v[0:1]
; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; VI-NEXT: s_endpgm
;
; EG-LABEL: ashr_i64_2:
; EG: ; %bb.0: ; %entry
; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]
; EG-NEXT: TEX 0 @6
; EG-NEXT: ALU 10, @9, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: Fetch clause starting at 6:
; EG-NEXT: VTX_READ_128 T0.XYZW, T0.X, 0, #1
; EG-NEXT: ALU clause starting at 8:
; EG-NEXT: MOV * T0.X, KC0[2].Z,
; EG-NEXT: ALU clause starting at 9:
; EG-NEXT: AND_INT * T0.W, T0.Z, literal.x,
; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00)
; EG-NEXT: ASHR T1.Z, T0.Y, PV.W,
; EG-NEXT: BIT_ALIGN_INT T0.W, T0.Y, T0.X, T0.Z,
; EG-NEXT: AND_INT * T1.W, T0.Z, literal.x,
; EG-NEXT: 32(4.484155e-44), 0(0.000000e+00)
; EG-NEXT: CNDE_INT T0.X, PS, PV.W, PV.Z,
; EG-NEXT: ASHR T0.W, T0.Y, literal.x,
; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y,
; EG-NEXT: 31(4.344025e-44), 2(2.802597e-45)
; EG-NEXT: CNDE_INT * T0.Y, T1.W, T1.Z, PV.W,
entry:
%b_ptr = getelementptr i64, ptr addrspace(1) %in, i64 1
%a = load i64, ptr addrspace(1) %in
%b = load i64, ptr addrspace(1) %b_ptr
%result = ashr i64 %a, %b
store i64 %result, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @ashr_v2i64(ptr addrspace(1) %out, ptr addrspace(1) %in) {
; SI-LABEL: ashr_v2i64:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, -1
; SI-NEXT: s_mov_b32 s10, s6
; SI-NEXT: s_mov_b32 s11, s7
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b32 s8, s2
; SI-NEXT: s_mov_b32 s9, s3
; SI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
; SI-NEXT: buffer_load_dwordx4 v[4:7], off, s[8:11], 0 offset:16
; SI-NEXT: s_mov_b32 s4, s0
; SI-NEXT: s_mov_b32 s5, s1
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_ashr_i64 v[2:3], v[2:3], v6
; SI-NEXT: v_ashr_i64 v[0:1], v[0:1], v4
; SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: ashr_v2i64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT: s_mov_b32 s7, 0xf000
; VI-NEXT: s_mov_b32 s6, -1
; VI-NEXT: s_mov_b32 s10, s6
; VI-NEXT: s_mov_b32 s11, s7
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_mov_b32 s8, s2
; VI-NEXT: s_mov_b32 s9, s3
; VI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0
; VI-NEXT: buffer_load_dwordx4 v[4:7], off, s[8:11], 0 offset:16
; VI-NEXT: s_mov_b32 s4, s0
; VI-NEXT: s_mov_b32 s5, s1
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_ashrrev_i64 v[2:3], v6, v[2:3]
; VI-NEXT: v_ashrrev_i64 v[0:1], v4, v[0:1]
; VI-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0
; VI-NEXT: s_endpgm
;
; EG-LABEL: ashr_v2i64:
; EG: ; %bb.0:
; EG-NEXT: ALU 0, @10, KC0[CB0:0-32], KC1[]
; EG-NEXT: TEX 1 @6
; EG-NEXT: ALU 19, @11, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: Fetch clause starting at 6:
; EG-NEXT: VTX_READ_128 T1.XYZW, T0.X, 16, #1
; EG-NEXT: VTX_READ_128 T0.XYZW, T0.X, 0, #1
; EG-NEXT: ALU clause starting at 10:
; EG-NEXT: MOV * T0.X, KC0[2].Z,
; EG-NEXT: ALU clause starting at 11:
; EG-NEXT: AND_INT * T1.W, T1.Z, literal.x,
; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00)
; EG-NEXT: ASHR T1.Y, T0.W, PV.W,
; EG-NEXT: AND_INT T2.Z, T1.Z, literal.x,
; EG-NEXT: BIT_ALIGN_INT T1.W, T0.W, T0.Z, T1.Z,
; EG-NEXT: AND_INT * T2.W, T1.X, literal.y,
; EG-NEXT: 32(4.484155e-44), 31(4.344025e-44)
; EG-NEXT: ASHR T2.Y, T0.Y, PS,
; EG-NEXT: CNDE_INT T0.Z, PV.Z, PV.W, PV.Y,
; EG-NEXT: BIT_ALIGN_INT T1.W, T0.Y, T0.X, T1.X,
; EG-NEXT: AND_INT * T2.W, T1.X, literal.x,
; EG-NEXT: 32(4.484155e-44), 0(0.000000e+00)
; EG-NEXT: CNDE_INT T0.X, PS, PV.W, PV.Y,
; EG-NEXT: ASHR T0.W, T0.W, literal.x,
; EG-NEXT: ASHR * T1.W, T0.Y, literal.x,
; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00)
; EG-NEXT: CNDE_INT * T0.W, T2.Z, T1.Y, PV.W,
; EG-NEXT: LSHR T1.X, KC0[2].Y, literal.x,
; EG-NEXT: CNDE_INT * T0.Y, T2.W, T2.Y, T1.W,
; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
%b_ptr = getelementptr <2 x i64>, ptr addrspace(1) %in, i64 1
%a = load <2 x i64>, ptr addrspace(1) %in
%b = load <2 x i64>, ptr addrspace(1) %b_ptr
%result = ashr <2 x i64> %a, %b
store <2 x i64> %result, ptr addrspace(1) %out
ret void
}
; FIXME: Broken on r600
define amdgpu_kernel void @ashr_v4i64(ptr addrspace(1) %out, ptr addrspace(1) %in) {
; SI-LABEL: ashr_v4i64:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[4:7], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: s_mov_b32 s10, s2
; SI-NEXT: s_mov_b32 s11, s3
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b32 s8, s6
; SI-NEXT: s_mov_b32 s9, s7
; SI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0 offset:16
; SI-NEXT: buffer_load_dwordx4 v[4:7], off, s[8:11], 0 offset:48
; SI-NEXT: buffer_load_dwordx4 v[7:10], off, s[8:11], 0
; SI-NEXT: buffer_load_dwordx4 v[11:14], off, s[8:11], 0 offset:32
; SI-NEXT: s_mov_b32 s0, s4
; SI-NEXT: s_mov_b32 s1, s5
; SI-NEXT: s_waitcnt vmcnt(2)
; SI-NEXT: v_ashr_i64 v[2:3], v[2:3], v6
; SI-NEXT: v_ashr_i64 v[0:1], v[0:1], v4
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_ashr_i64 v[9:10], v[9:10], v13
; SI-NEXT: v_ashr_i64 v[7:8], v[7:8], v11
; SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:16
; SI-NEXT: buffer_store_dwordx4 v[7:10], off, s[0:3], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: ashr_v4i64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[4:7], s[4:5], 0x24
; VI-NEXT: s_mov_b32 s3, 0xf000
; VI-NEXT: s_mov_b32 s2, -1
; VI-NEXT: s_mov_b32 s10, s2
; VI-NEXT: s_mov_b32 s11, s3
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_mov_b32 s8, s6
; VI-NEXT: s_mov_b32 s9, s7
; VI-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0 offset:16
; VI-NEXT: buffer_load_dwordx4 v[4:7], off, s[8:11], 0 offset:48
; VI-NEXT: buffer_load_dwordx4 v[7:10], off, s[8:11], 0
; VI-NEXT: buffer_load_dwordx4 v[11:14], off, s[8:11], 0 offset:32
; VI-NEXT: s_mov_b32 s0, s4
; VI-NEXT: s_mov_b32 s1, s5
; VI-NEXT: s_waitcnt vmcnt(2)
; VI-NEXT: v_ashrrev_i64 v[2:3], v6, v[2:3]
; VI-NEXT: v_ashrrev_i64 v[0:1], v4, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_ashrrev_i64 v[9:10], v13, v[9:10]
; VI-NEXT: v_ashrrev_i64 v[7:8], v11, v[7:8]
; VI-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:16
; VI-NEXT: buffer_store_dwordx4 v[7:10], off, s[0:3], 0
; VI-NEXT: s_endpgm
;
; EG-LABEL: ashr_v4i64:
; EG: ; %bb.0:
; EG-NEXT: ALU 0, @14, KC0[CB0:0-32], KC1[]
; EG-NEXT: TEX 3 @6
; EG-NEXT: ALU 39, @15, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T2.XYZW, T3.X, 0
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: Fetch clause starting at 6:
; EG-NEXT: VTX_READ_128 T1.XYZW, T0.X, 32, #1
; EG-NEXT: VTX_READ_128 T2.XYZW, T0.X, 48, #1
; EG-NEXT: VTX_READ_128 T3.XYZW, T0.X, 0, #1
; EG-NEXT: VTX_READ_128 T0.XYZW, T0.X, 16, #1
; EG-NEXT: ALU clause starting at 14:
; EG-NEXT: MOV * T0.X, KC0[2].Z,
; EG-NEXT: ALU clause starting at 15:
; EG-NEXT: AND_INT * T1.W, T1.Z, literal.x,
; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00)
; EG-NEXT: ASHR T1.Y, T0.W, literal.x,
; EG-NEXT: ASHR T4.Z, T3.W, PV.W, BS:VEC_120/SCL_212
; EG-NEXT: AND_INT T1.W, T1.Z, literal.y,
; EG-NEXT: AND_INT * T2.W, T2.Z, literal.x,
; EG-NEXT: 31(4.344025e-44), 32(4.484155e-44)
; EG-NEXT: BIT_ALIGN_INT T4.X, T3.W, T3.Z, T1.Z,
; EG-NEXT: ASHR T2.Y, T0.W, PS, BS:VEC_120/SCL_212
; EG-NEXT: AND_INT * T1.Z, T2.Z, literal.x,
; EG-NEXT: 32(4.484155e-44), 0(0.000000e+00)
; EG-NEXT: BIT_ALIGN_INT T0.W, T0.W, T0.Z, T2.Z,
; EG-NEXT: AND_INT * T2.W, T2.X, literal.x,
; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00)
; EG-NEXT: AND_INT T5.X, T1.X, literal.x,
; EG-NEXT: ASHR T4.Y, T0.Y, PS,
; EG-NEXT: CNDE_INT T0.Z, T1.Z, PV.W, T2.Y,
; EG-NEXT: BIT_ALIGN_INT T0.W, T0.Y, T0.X, T2.X,
; EG-NEXT: AND_INT * T2.W, T2.X, literal.y,
; EG-NEXT: 31(4.344025e-44), 32(4.484155e-44)
; EG-NEXT: CNDE_INT T0.X, PS, PV.W, PV.Y,
; EG-NEXT: ASHR T5.Y, T3.Y, PV.X,
; EG-NEXT: CNDE_INT T2.Z, T1.W, T4.X, T4.Z,
; EG-NEXT: BIT_ALIGN_INT T0.W, T3.Y, T3.X, T1.X, BS:VEC_102/SCL_221
; EG-NEXT: AND_INT * T4.W, T1.X, literal.x,
; EG-NEXT: 32(4.484155e-44), 0(0.000000e+00)
; EG-NEXT: CNDE_INT T2.X, PS, PV.W, PV.Y,
; EG-NEXT: ASHR T6.Y, T3.W, literal.x,
; EG-NEXT: ASHR T3.Z, T0.Y, literal.x, BS:VEC_201
; EG-NEXT: ADD_INT T3.W, KC0[2].Y, literal.y,
; EG-NEXT: CNDE_INT * T0.W, T1.Z, T2.Y, T1.Y,
; EG-NEXT: 31(4.344025e-44), 16(2.242078e-44)
; EG-NEXT: LSHR T1.X, PV.W, literal.x,
; EG-NEXT: CNDE_INT T0.Y, T2.W, T4.Y, PV.Z,
; EG-NEXT: ASHR T3.W, T3.Y, literal.y,
; EG-NEXT: CNDE_INT * T2.W, T1.W, T4.Z, PV.Y,
; EG-NEXT: 2(2.802597e-45), 31(4.344025e-44)
; EG-NEXT: LSHR T3.X, KC0[2].Y, literal.x,
; EG-NEXT: CNDE_INT * T2.Y, T4.W, T5.Y, PV.W,
; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
%b_ptr = getelementptr <4 x i64>, ptr addrspace(1) %in, i64 1
%a = load <4 x i64>, ptr addrspace(1) %in
%b = load <4 x i64>, ptr addrspace(1) %b_ptr
%result = ashr <4 x i64> %a, %b
store <4 x i64> %result, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @s_ashr_32_i64(ptr addrspace(1) %out, [8 x i32], i64 %a, [8 x i32], i64 %b) {
; SI-LABEL: s_ashr_32_i64:
; SI: ; %bb.0:
; SI-NEXT: s_load_dword s8, s[4:5], 0x14
; SI-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x1d
; SI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_ashr_i32 s5, s8, 31
; SI-NEXT: s_add_u32 s4, s8, s6
; SI-NEXT: s_addc_u32 s5, s5, s7
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: s_ashr_32_i64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dword s8, s[4:5], 0x50
; VI-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x74
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-NEXT: s_mov_b32 s3, 0xf000
; VI-NEXT: s_mov_b32 s2, -1
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_ashr_i32 s5, s8, 31
; VI-NEXT: s_add_u32 s4, s8, s6
; VI-NEXT: s_addc_u32 s5, s5, s7
; VI-NEXT: v_mov_b32_e32 v0, s4
; VI-NEXT: v_mov_b32_e32 v1, s5
; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT: s_endpgm
;
; EG-LABEL: s_ashr_32_i64:
; EG: ; %bb.0:
; EG-NEXT: ALU 7, @4, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: ALU clause starting at 4:
; EG-NEXT: ASHR * T0.W, KC0[5].X, literal.x,
; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00)
; EG-NEXT: ADD_INT * T0.W, PV.W, KC0[7].Z,
; EG-NEXT: ADDC_UINT * T1.W, KC0[5].X, KC0[7].Y,
; EG-NEXT: ADD_INT * T0.Y, T0.W, PV.W,
; EG-NEXT: ADD_INT * T0.X, KC0[5].X, KC0[7].Y,
; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
%result = ashr i64 %a, 32
%add = add i64 %result, %b
store i64 %add, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_ashr_32_i64(ptr addrspace(1) %out, ptr addrspace(1) %in) {
; SI-LABEL: v_ashr_32_i64:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, 0
; SI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT: v_mov_b32_e32 v1, 0
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b64 s[8:9], s[2:3]
; SI-NEXT: s_mov_b64 s[10:11], s[6:7]
; SI-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 offset:4
; SI-NEXT: s_mov_b64 s[4:5], s[0:1]
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_ashrrev_i32_e32 v3, 31, v2
; SI-NEXT: buffer_store_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
; SI-NEXT: s_endpgm
;
; VI-LABEL: v_ashr_32_i64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT: v_lshlrev_b32_e32 v2, 3, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s3
; VI-NEXT: v_add_u32_e32 v1, vcc, s2, v2
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v0, vcc
; VI-NEXT: v_add_u32_e32 v0, vcc, 4, v1
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc
; VI-NEXT: flat_load_dword v0, v[0:1]
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: v_add_u32_e32 v2, vcc, s0, v2
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v1, vcc
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_ashrrev_i32_e32 v1, 31, v0
; VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT: s_endpgm
;
; EG-LABEL: v_ashr_32_i64:
; EG: ; %bb.0:
; EG-NEXT: ALU 2, @8, KC0[CB0:0-32], KC1[]
; EG-NEXT: TEX 0 @6
; EG-NEXT: ALU 3, @11, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: Fetch clause starting at 6:
; EG-NEXT: VTX_READ_32 T0.X, T0.X, 4, #1
; EG-NEXT: ALU clause starting at 8:
; EG-NEXT: LSHL * T0.W, T0.X, literal.x,
; EG-NEXT: 3(4.203895e-45), 0(0.000000e+00)
; EG-NEXT: ADD_INT * T0.X, KC0[2].Z, PV.W,
; EG-NEXT: ALU clause starting at 11:
; EG-NEXT: ADD_INT * T0.W, KC0[2].Y, T0.W,
; EG-NEXT: LSHR T1.X, PV.W, literal.x,
; EG-NEXT: ASHR * T0.Y, T0.X, literal.y,
; EG-NEXT: 2(2.802597e-45), 31(4.344025e-44)
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
%gep.in = getelementptr i64, ptr addrspace(1) %in, i32 %tid
%gep.out = getelementptr i64, ptr addrspace(1) %out, i32 %tid
%a = load i64, ptr addrspace(1) %gep.in
%result = ashr i64 %a, 32
store i64 %result, ptr addrspace(1) %gep.out
ret void
}
define amdgpu_kernel void @s_ashr_63_i64(ptr addrspace(1) %out, [8 x i32], i64 %a, [8 x i32], i64 %b) {
; SI-LABEL: s_ashr_63_i64:
; SI: ; %bb.0:
; SI-NEXT: s_load_dword s8, s[4:5], 0x14
; SI-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x1d
; SI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_ashr_i32 s5, s8, 31
; SI-NEXT: s_add_u32 s4, s5, s6
; SI-NEXT: s_addc_u32 s5, s5, s7
; SI-NEXT: v_mov_b32_e32 v0, s4
; SI-NEXT: v_mov_b32_e32 v1, s5
; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT: s_endpgm
;
; VI-LABEL: s_ashr_63_i64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dword s8, s[4:5], 0x50
; VI-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x74
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-NEXT: s_mov_b32 s3, 0xf000
; VI-NEXT: s_mov_b32 s2, -1
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_ashr_i32 s5, s8, 31
; VI-NEXT: s_add_u32 s4, s5, s6
; VI-NEXT: s_addc_u32 s5, s5, s7
; VI-NEXT: v_mov_b32_e32 v0, s4
; VI-NEXT: v_mov_b32_e32 v1, s5
; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT: s_endpgm
;
; EG-LABEL: s_ashr_63_i64:
; EG: ; %bb.0:
; EG-NEXT: ALU 7, @4, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: ALU clause starting at 4:
; EG-NEXT: ASHR * T0.W, KC0[5].X, literal.x,
; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00)
; EG-NEXT: ADD_INT T1.W, PV.W, KC0[7].Z,
; EG-NEXT: ADDC_UINT * T2.W, PV.W, KC0[7].Y,
; EG-NEXT: ADD_INT * T0.Y, PV.W, PS,
; EG-NEXT: ADD_INT T0.X, T0.W, KC0[7].Y,
; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
%result = ashr i64 %a, 63
%add = add i64 %result, %b
store i64 %add, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_ashr_63_i64(ptr addrspace(1) %out, ptr addrspace(1) %in) {
; SI-LABEL: v_ashr_63_i64:
; SI: ; %bb.0:
; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT: s_mov_b32 s7, 0xf000
; SI-NEXT: s_mov_b32 s6, 0
; SI-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT: v_mov_b32_e32 v1, 0
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_mov_b64 s[8:9], s[2:3]
; SI-NEXT: s_mov_b64 s[10:11], s[6:7]
; SI-NEXT: buffer_load_dword v2, v[0:1], s[8:11], 0 addr64 offset:4
; SI-NEXT: s_mov_b64 s[4:5], s[0:1]
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: v_ashrrev_i32_e32 v2, 31, v2
; SI-NEXT: v_mov_b32_e32 v3, v2
; SI-NEXT: buffer_store_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
; SI-NEXT: s_endpgm
;
; VI-LABEL: v_ashr_63_i64:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT: v_lshlrev_b32_e32 v2, 3, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s3
; VI-NEXT: v_add_u32_e32 v1, vcc, s2, v2
; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v0, vcc
; VI-NEXT: v_add_u32_e32 v0, vcc, 4, v1
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc
; VI-NEXT: flat_load_dword v3, v[0:1]
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: v_add_u32_e32 v0, vcc, s0, v2
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_ashrrev_i32_e32 v2, 31, v3
; VI-NEXT: v_mov_b32_e32 v3, v2
; VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3]
; VI-NEXT: s_endpgm
;
; EG-LABEL: v_ashr_63_i64:
; EG: ; %bb.0:
; EG-NEXT: ALU 2, @8, KC0[CB0:0-32], KC1[]
; EG-NEXT: TEX 0 @6
; EG-NEXT: ALU 5, @11, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
; EG-NEXT: Fetch clause starting at 6:
; EG-NEXT: VTX_READ_32 T0.X, T0.X, 4, #1
; EG-NEXT: ALU clause starting at 8:
; EG-NEXT: LSHL * T0.W, T0.X, literal.x,
; EG-NEXT: 3(4.203895e-45), 0(0.000000e+00)
; EG-NEXT: ADD_INT * T0.X, KC0[2].Z, PV.W,
; EG-NEXT: ALU clause starting at 11:
; EG-NEXT: ASHR T0.X, T0.X, literal.x,
; EG-NEXT: ADD_INT * T0.W, KC0[2].Y, T0.W,
; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00)
; EG-NEXT: LSHR T1.X, PV.W, literal.x,
; EG-NEXT: MOV * T0.Y, PV.X,
; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
%gep.in = getelementptr i64, ptr addrspace(1) %in, i32 %tid
%gep.out = getelementptr i64, ptr addrspace(1) %out, i32 %tid
%a = load i64, ptr addrspace(1) %gep.in
%result = ashr i64 %a, 63
store i64 %result, ptr addrspace(1) %gep.out
ret void
}
attributes #0 = { nounwind readnone }
Reference in New Issue View Git Blame Copy Permalink