2291bd137d
Support for XNACK and SRAMECC is not static on some GPUs. We must be able to differentiate between different scenarios for these dynamic subtarget features. The possible settings are: - Unsupported: The GPU has no support for XNACK/SRAMECC. - Any: Preference is unspecified. Use conservative settings that can run anywhere. - Off: Request support for XNACK/SRAMECC Off - On: Request support for XNACK/SRAMECC On GCNSubtarget will track the four options based on the following criteria. If the subtarget does not support XNACK/SRAMECC we say the setting is "Unsupported". If no subtarget features for XNACK/SRAMECC are requested we must support "Any" mode. If the subtarget features XNACK/SRAMECC exist in the feature string when initializing the subtarget, the settings are "On/Off". The defaults are updated to be conservatively correct, meaning if no setting for XNACK or SRAMECC is explicitly requested, defaults will be used which generate code that can be run anywhere. This corresponds to the "Any" setting. Differential Revision: https://reviews.llvm.org/D85882
2766 lines
108 KiB
LLVM
2766 lines
108 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX7 %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX8 %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX9-NODL %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX9-DL %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx1011 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-DL %s
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx1012 -verify-machineinstrs < %s | FileCheck --check-prefixes=GFX10-DL %s
|
|
|
|
; add(mul(S0.x, S1.y),
|
|
; add (mul (S0.y, S1.y), S3)) -> v_dot2_{I|U}32_{I|U}16(S1, S2, S3)
|
|
|
|
define amdgpu_kernel void @udot2(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: udot2:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX7-NEXT: s_lshr_b32 s7, s5, 16
|
|
; GFX7-NEXT: s_and_b32 s4, s4, s8
|
|
; GFX7-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX7-NEXT: s_load_dword s8, s[0:1], 0x0
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s7, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s5, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: udot2:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s3, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s5, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s6, s3, s2
|
|
; GFX8-NEXT: s_lshr_b32 s3, s3, 16
|
|
; GFX8-NEXT: s_and_b32 s2, s4, s2
|
|
; GFX8-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s4, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s2, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: udot2:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, v1, v2
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: udot2:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s8
|
|
; GFX9-DL-NEXT: v_dot2_u32_u16 v1, s1, v1, v2
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: udot2:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: v_dot2_u32_u16 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul nuw i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul nuw i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
; TODO: Support this pattern
|
|
; add(S3,
|
|
; add (mul (S0.y, S1.y), mul (S0.y, S1.y))) -> v_dot2_{I|U}32_{I|U}16(S1, S2, S3)
|
|
define amdgpu_kernel void @udot2_MulMul(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: udot2_MulMul:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX7-NEXT: s_and_b32 s4, s4, s8
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_load_dword s4, s[0:1], 0x0
|
|
; GFX7-NEXT: s_lshr_b32 s7, s5, 16
|
|
; GFX7-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX7-NEXT: v_mul_u32_u24_e32 v0, s5, v0
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s7, v1, v0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_add_i32_e32 v0, vcc, s4, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: udot2_MulMul:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s3, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s5, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s6, s3, s2
|
|
; GFX8-NEXT: s_and_b32 s2, s4, s2
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX8-NEXT: s_lshr_b32 s3, s3, 16
|
|
; GFX8-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s3
|
|
; GFX8-NEXT: v_mul_u32_u24_e32 v0, s2, v0
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s4, v1, v0
|
|
; GFX8-NEXT: v_add_u32_e32 v2, vcc, s5, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: udot2_MulMul:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mul_u32_u24_e32 v1, s5, v1
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, v2, v1
|
|
; GFX9-NODL-NEXT: v_add_u32_e32 v1, s9, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: udot2_MulMul:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-DL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX9-DL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-DL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-DL-NEXT: v_mul_u32_u24_e32 v1, s5, v1
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s1, v2, v1
|
|
; GFX9-DL-NEXT: v_add_u32_e32 v1, s9, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: udot2_MulMul:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_and_b32 s5, s0, s4
|
|
; GFX10-DL-NEXT: s_and_b32 s4, s1, s4
|
|
; GFX10-DL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX10-DL-NEXT: v_mul_u32_u24_e64 v0, s4, s5
|
|
; GFX10-DL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: v_add_nc_u32_e32 v0, s8, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul nuw i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul nuw i32 %conv4, %conv3
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %mul1
|
|
%add6 = add i32 %add, %s3
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @idot2(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: idot2:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_load_dword s6, s[0:1], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_sext_i32_i16 s7, s4
|
|
; GFX7-NEXT: s_ashr_i32 s4, s4, 16
|
|
; GFX7-NEXT: s_sext_i32_i16 s8, s5
|
|
; GFX7-NEXT: s_ashr_i32 s5, s5, 16
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, s5, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s7
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, s8, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: idot2:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s3, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_sext_i32_i16 s5, s2
|
|
; GFX8-NEXT: s_ashr_i32 s2, s2, 16
|
|
; GFX8-NEXT: s_sext_i32_i16 s6, s3
|
|
; GFX8-NEXT: s_ashr_i32 s3, s3, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX8-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX8-NEXT: v_mad_i32_i24 v0, s3, v1, v0
|
|
; GFX8-NEXT: v_mad_i32_i24 v2, s6, v2, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: idot2:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-NODL-NEXT: s_ashr_i32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s5, s1
|
|
; GFX9-NODL-NEXT: s_ashr_i32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s1, v2, v1
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: idot2:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s8
|
|
; GFX9-DL-NEXT: v_dot2_i32_i16 v1, s1, v1, v2
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: idot2:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: v_dot2_i32_i16 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = sext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = sext i16 %s2.elt1 to i32
|
|
%mul1 = mul nuw i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = sext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = sext i16 %s2.elt2 to i32
|
|
%mul2 = mul nuw i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @idot2_MixedTypedMul(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: idot2_MixedTypedMul:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_load_dword s6, s[0:1], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_lshr_b32 s7, s4, 16
|
|
; GFX7-NEXT: s_lshr_b32 s8, s5, 16
|
|
; GFX7-NEXT: s_sext_i32_i16 s4, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s7
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s8, v0, v1
|
|
; GFX7-NEXT: s_sext_i32_i16 s5, s5
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, s5, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: idot2_MixedTypedMul:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s3, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_sext_i32_i16 s5, s2
|
|
; GFX8-NEXT: s_lshr_b32 s2, s2, 16
|
|
; GFX8-NEXT: s_sext_i32_i16 s6, s3
|
|
; GFX8-NEXT: s_lshr_b32 s3, s3, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX8-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s3, v1, v0
|
|
; GFX8-NEXT: v_mad_i32_i24 v2, s6, v2, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: idot2_MixedTypedMul:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s5, s1
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, v2, v1
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: idot2_MixedTypedMul:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-DL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-DL-NEXT: s_sext_i32_i16 s5, s1
|
|
; GFX9-DL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s1, v2, v1
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: idot2_MixedTypedMul:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: s_lshr_b32 s4, s0, 16
|
|
; GFX10-DL-NEXT: s_lshr_b32 s5, s1, 16
|
|
; GFX10-DL-NEXT: s_sext_i32_i16 s0, s0
|
|
; GFX10-DL-NEXT: s_sext_i32_i16 s1, s1
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s5, s4, v0
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = sext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = sext i16 %s2.elt1 to i32
|
|
%mul1 = mul nuw i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul nuw i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @udot2_alt_AddOperands(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: udot2_alt_AddOperands:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX7-NEXT: s_lshr_b32 s7, s5, 16
|
|
; GFX7-NEXT: s_and_b32 s4, s4, s8
|
|
; GFX7-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX7-NEXT: s_load_dword s8, s[0:1], 0x0
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s7, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s5, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: udot2_alt_AddOperands:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s3, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s5, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s6, s3, s2
|
|
; GFX8-NEXT: s_lshr_b32 s3, s3, 16
|
|
; GFX8-NEXT: s_and_b32 s2, s4, s2
|
|
; GFX8-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s4, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s2, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: udot2_alt_AddOperands:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, v1, v2
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: udot2_alt_AddOperands:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s8
|
|
; GFX9-DL-NEXT: v_dot2_u32_u16 v1, s1, v1, v2
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: udot2_alt_AddOperands:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: v_dot2_u32_u16 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul nuw i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul nuw i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %s3, %mul2
|
|
%add6 = add i32 %mul1, %add
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @idot2_MixedExt(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: idot2_MixedExt:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_load_dword s6, s[0:1], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_sext_i32_i16 s7, s4
|
|
; GFX7-NEXT: s_ashr_i32 s4, s4, 16
|
|
; GFX7-NEXT: s_and_b32 s8, s5, 0xffff
|
|
; GFX7-NEXT: s_ashr_i32 s5, s5, 16
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, s5, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s7
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, s8, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: idot2_MixedExt:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s3, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_sext_i32_i16 s5, s2
|
|
; GFX8-NEXT: s_ashr_i32 s2, s2, 16
|
|
; GFX8-NEXT: s_and_b32 s6, s3, 0xffff
|
|
; GFX8-NEXT: s_ashr_i32 s3, s3, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX8-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX8-NEXT: v_mad_i32_i24 v0, s3, v1, v0
|
|
; GFX8-NEXT: v_mad_i32_i24 v2, s6, v2, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: idot2_MixedExt:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-NODL-NEXT: s_ashr_i32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s1, 0xffff
|
|
; GFX9-NODL-NEXT: s_ashr_i32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s1, v2, v1
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: idot2_MixedExt:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-DL-NEXT: s_ashr_i32 s0, s0, 16
|
|
; GFX9-DL-NEXT: s_and_b32 s5, s1, 0xffff
|
|
; GFX9-DL-NEXT: s_ashr_i32 s1, s1, 16
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, s1, v2, v1
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: idot2_MixedExt:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: s_ashr_i32 s4, s0, 16
|
|
; GFX10-DL-NEXT: s_ashr_i32 s5, s1, 16
|
|
; GFX10-DL-NEXT: s_sext_i32_i16 s0, s0
|
|
; GFX10-DL-NEXT: s_and_b32 s1, s1, 0xffff
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v0, s5, s4, v0
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = sext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul nuw i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = sext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = sext i16 %s2.elt2 to i32
|
|
%mul2 = mul nuw i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @notudot2_SameVec(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: notudot2_SameVec:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s6, s[6:7], 0x0
|
|
; GFX7-NEXT: s_load_dword s7, s[0:1], 0x0
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_lshr_b32 s5, s6, 16
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s7
|
|
; GFX7-NEXT: s_and_b32 s4, s4, 0xffff
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s5, s5, v0
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s4, s4, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: notudot2_SameVec:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s2, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s3, s[0:1], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_lshr_b32 s2, s2, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s3
|
|
; GFX8-NEXT: s_and_b32 s4, s4, 0xffff
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s2, s2, v0
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s4, s4, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: notudot2_SameVec:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-NODL-NEXT: s_and_b32 s0, s0, 0xffff
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, s1, v1
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s0, s0, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: notudot2_SameVec:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-DL-NEXT: s_and_b32 s0, s0, 0xffff
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s1, s1, v1
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s0, s0, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: notudot2_SameVec:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s1, s1, s8
|
|
; GFX10-DL-NEXT: s_and_b32 s0, s0, 0xffff
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s0, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @udot2_v4i16(<4 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: udot2_v4i16:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_and_b32 s6, s4, s8
|
|
; GFX7-NEXT: s_and_b32 s7, s5, s8
|
|
; GFX7-NEXT: s_load_dword s8, s[0:1], 0x0
|
|
; GFX7-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX7-NEXT: s_lshr_b32 s5, s5, 16
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s5, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s7, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: udot2_v4i16:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s3, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s5, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s6, s3, s2
|
|
; GFX8-NEXT: s_lshr_b32 s3, s3, 16
|
|
; GFX8-NEXT: s_and_b32 s2, s4, s2
|
|
; GFX8-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s4, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s2, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: udot2_v4i16:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, v1, v2
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: udot2_v4i16:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s8
|
|
; GFX9-DL-NEXT: v_dot2_u32_u16 v1, s1, v1, v2
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: udot2_v4i16:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: v_dot2_u32_u16 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<4 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <4 x i16>, <4 x i16> addrspace(1)* %src1
|
|
%vec2 = load <4 x i16>, <4 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <4 x i16> %vec1, i64 0
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <4 x i16> %vec2, i64 0
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <4 x i16> %vec1, i64 1
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <4 x i16> %vec2, i64 1
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @udot2_v4i16_Hi(<4 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: udot2_v4i16_Hi:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x1
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_and_b32 s6, s4, s8
|
|
; GFX7-NEXT: s_and_b32 s7, s5, s8
|
|
; GFX7-NEXT: s_load_dword s8, s[0:1], 0x0
|
|
; GFX7-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX7-NEXT: s_lshr_b32 s5, s5, 16
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s5, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s7, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: udot2_v4i16_Hi:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s3, s[4:5], 0x4
|
|
; GFX8-NEXT: s_load_dword s4, s[6:7], 0x4
|
|
; GFX8-NEXT: s_load_dword s5, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s6, s3, s2
|
|
; GFX8-NEXT: s_lshr_b32 s3, s3, 16
|
|
; GFX8-NEXT: s_and_b32 s2, s4, s2
|
|
; GFX8-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s4, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s2, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: udot2_v4i16_Hi:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x4
|
|
; GFX9-NODL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, v1, v2
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: udot2_v4i16_Hi:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x4
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s8
|
|
; GFX9-DL-NEXT: v_dot2_u32_u16 v1, s1, v1, v2
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: udot2_v4i16_Hi:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x4
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x4
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: v_dot2_u32_u16 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<4 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <4 x i16>, <4 x i16> addrspace(1)* %src1
|
|
%vec2 = load <4 x i16>, <4 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <4 x i16> %vec1, i64 2
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <4 x i16> %vec2, i64 2
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <4 x i16> %vec1, i64 3
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <4 x i16> %vec2, i64 3
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @notudot2_v4i16_Even(<4 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: notudot2_v4i16_Even:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dwordx2 s[6:7], s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX7-NEXT: s_and_b32 s4, s4, s8
|
|
; GFX7-NEXT: s_and_b32 s6, s6, s8
|
|
; GFX7-NEXT: s_and_b32 s7, s7, s8
|
|
; GFX7-NEXT: s_load_dword s8, s[0:1], 0x0
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s7, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s6, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: notudot2_v4i16_Even:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dwordx2 s[2:3], s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dwordx2 s[4:5], s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s6, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s3, s3, s8
|
|
; GFX8-NEXT: s_and_b32 s2, s2, s8
|
|
; GFX8-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s5, v1, v0
|
|
; GFX8-NEXT: s_and_b32 s4, s4, s8
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s4, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: notudot2_v4i16_Even:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s10, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[8:9], s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s11, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s1, s1, s10
|
|
; GFX9-NODL-NEXT: s_and_b32 s0, s0, s10
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s9, s10
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s11
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s5, v1, v2
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s8, s10
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s4, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: notudot2_v4i16_Even:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: s_mov_b32 s10, 0xffff
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[8:9], s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s11, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_and_b32 s1, s1, s10
|
|
; GFX9-DL-NEXT: s_and_b32 s0, s0, s10
|
|
; GFX9-DL-NEXT: s_and_b32 s5, s9, s10
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s11
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s5, v1, v2
|
|
; GFX9-DL-NEXT: s_and_b32 s4, s8, s10
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s4, v2, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: notudot2_v4i16_Even:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s10, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[8:9], s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s10
|
|
; GFX10-DL-NEXT: s_and_b32 s1, s1, s4
|
|
; GFX10-DL-NEXT: s_and_b32 s5, s9, s4
|
|
; GFX10-DL-NEXT: s_and_b32 s0, s0, s4
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s5, s1, v0
|
|
; GFX10-DL-NEXT: s_and_b32 s1, s8, s4
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<4 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <4 x i16>, <4 x i16> addrspace(1)* %src1
|
|
%vec2 = load <4 x i16>, <4 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <4 x i16> %vec1, i64 0
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <4 x i16> %vec2, i64 0
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <4 x i16> %vec1, i64 2
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <4 x i16> %vec2, i64 2
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @notudot2_v4i16_Middle(<4 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: notudot2_v4i16_Middle:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dwordx2 s[6:7], s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX7-NEXT: s_and_b32 s7, s7, s8
|
|
; GFX7-NEXT: s_load_dword s8, s[0:1], 0x0
|
|
; GFX7-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX7-NEXT: s_lshr_b32 s6, s6, 16
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s7, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s6, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: notudot2_v4i16_Middle:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dwordx2 s[2:3], s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dwordx2 s[4:5], s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s6, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s3, s3, s8
|
|
; GFX8-NEXT: s_lshr_b32 s2, s2, 16
|
|
; GFX8-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s5, v1, v0
|
|
; GFX8-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s4, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: notudot2_v4i16_Middle:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s10, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[8:9], s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s11, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s1, s1, s10
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s9, s10
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s11
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s5, v1, v2
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s4, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: notudot2_v4i16_Middle:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: s_mov_b32 s10, 0xffff
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[8:9], s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s11, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_and_b32 s1, s1, s10
|
|
; GFX9-DL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-DL-NEXT: s_and_b32 s5, s9, s10
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s11
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s5, v1, v2
|
|
; GFX9-DL-NEXT: s_lshr_b32 s4, s8, 16
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s4, v2, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: notudot2_v4i16_Middle:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s10, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[8:9], s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s10
|
|
; GFX10-DL-NEXT: s_and_b32 s1, s1, s4
|
|
; GFX10-DL-NEXT: s_and_b32 s4, s9, s4
|
|
; GFX10-DL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s4, s1, v0
|
|
; GFX10-DL-NEXT: s_lshr_b32 s1, s8, 16
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<4 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <4 x i16>, <4 x i16> addrspace(1)* %src1
|
|
%vec2 = load <4 x i16>, <4 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <4 x i16> %vec1, i64 1
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <4 x i16> %vec2, i64 1
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <4 x i16> %vec1, i64 2
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <4 x i16> %vec2, i64 2
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @notudot2_DiffIndex(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: notudot2_DiffIndex:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX7-NEXT: s_lshr_b32 s7, s5, 16
|
|
; GFX7-NEXT: s_and_b32 s4, s4, s8
|
|
; GFX7-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX7-NEXT: s_load_dword s8, s[0:1], 0x0
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s5, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s7, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: notudot2_DiffIndex:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s3, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s5, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s6, s3, s2
|
|
; GFX8-NEXT: s_lshr_b32 s3, s3, 16
|
|
; GFX8-NEXT: s_and_b32 s2, s4, s2
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s2, v1, v0
|
|
; GFX8-NEXT: s_lshr_b32 s7, s4, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s7, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: notudot2_DiffIndex:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s5, s1, 16
|
|
; GFX9-NODL-NEXT: s_and_b32 s1, s1, s8
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, v1, v2
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: notudot2_DiffIndex:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-DL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-DL-NEXT: s_lshr_b32 s5, s1, 16
|
|
; GFX9-DL-NEXT: s_and_b32 s1, s1, s8
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s1, v1, v2
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s5, v2, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: notudot2_DiffIndex:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: s_lshr_b32 s5, s0, 16
|
|
; GFX10-DL-NEXT: s_and_b32 s6, s1, s4
|
|
; GFX10-DL-NEXT: s_and_b32 s0, s0, s4
|
|
; GFX10-DL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s6, s5, v0
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @udot2_MultipleUses_add1(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: udot2_MultipleUses_add1:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX7-NEXT: s_lshr_b32 s7, s5, 16
|
|
; GFX7-NEXT: s_and_b32 s4, s4, s8
|
|
; GFX7-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX7-NEXT: s_load_dword s8, s[0:1], 0x0
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s7, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: v_mad_u32_u24 v1, s5, v1, v0
|
|
; GFX7-NEXT: v_add_i32_e32 v0, vcc, v0, v1
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: udot2_MultipleUses_add1:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s3, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s5, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s6, s3, s2
|
|
; GFX8-NEXT: s_lshr_b32 s3, s3, 16
|
|
; GFX8-NEXT: s_and_b32 s2, s4, s2
|
|
; GFX8-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s4, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX8-NEXT: v_mad_u32_u24 v1, s2, v1, v0
|
|
; GFX8-NEXT: v_add_u32_e32 v2, vcc, v0, v1
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: udot2_MultipleUses_add1:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, v1, v2
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v2, s5, v2, v1
|
|
; GFX9-NODL-NEXT: v_add_u32_e32 v1, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: udot2_MultipleUses_add1:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-DL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-DL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-DL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s1, v1, v2
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v2, s5, v2, v1
|
|
; GFX9-DL-NEXT: v_add_u32_e32 v1, v2, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: udot2_MultipleUses_add1:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_mov_b32 s6, 0xffff
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: s_lshr_b32 s4, s0, 16
|
|
; GFX10-DL-NEXT: s_lshr_b32 s5, s1, 16
|
|
; GFX10-DL-NEXT: s_and_b32 s0, s0, s6
|
|
; GFX10-DL-NEXT: s_and_b32 s1, s1, s6
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s5, s4, v0
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v1, s1, s0, v0
|
|
; GFX10-DL-NEXT: v_add_nc_u32_e32 v0, v1, v0
|
|
; GFX10-DL-NEXT: global_store_dword v2, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add1 = add i32 %mul2, %s3
|
|
%add2 = add i32 %add1, %mul1
|
|
|
|
%res = add i32 %add2, %add1
|
|
store i32 %res, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @idot2_MultipleUses_add1(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: idot2_MultipleUses_add1:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_load_dword s6, s[0:1], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_sext_i32_i16 s7, s4
|
|
; GFX7-NEXT: s_ashr_i32 s4, s4, 16
|
|
; GFX7-NEXT: s_sext_i32_i16 s8, s5
|
|
; GFX7-NEXT: s_ashr_i32 s5, s5, 16
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, s5, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s7
|
|
; GFX7-NEXT: v_mad_i32_i24 v1, s8, v1, v0
|
|
; GFX7-NEXT: v_add_i32_e32 v0, vcc, v0, v1
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: idot2_MultipleUses_add1:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s3, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_sext_i32_i16 s5, s2
|
|
; GFX8-NEXT: s_ashr_i32 s2, s2, 16
|
|
; GFX8-NEXT: s_sext_i32_i16 s6, s3
|
|
; GFX8-NEXT: s_ashr_i32 s3, s3, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX8-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX8-NEXT: v_mad_i32_i24 v0, s3, v1, v0
|
|
; GFX8-NEXT: v_mad_i32_i24 v1, s6, v2, v0
|
|
; GFX8-NEXT: v_add_u32_e32 v2, vcc, v0, v1
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: idot2_MultipleUses_add1:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-NODL-NEXT: s_ashr_i32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s5, s1
|
|
; GFX9-NODL-NEXT: s_ashr_i32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s1, v2, v1
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v2, s5, v2, v1
|
|
; GFX9-NODL-NEXT: v_add_u32_e32 v1, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: idot2_MultipleUses_add1:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-DL-NEXT: s_ashr_i32 s0, s0, 16
|
|
; GFX9-DL-NEXT: s_sext_i32_i16 s5, s1
|
|
; GFX9-DL-NEXT: s_ashr_i32 s1, s1, 16
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, s1, v2, v1
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v2, s5, v2, v1
|
|
; GFX9-DL-NEXT: v_add_u32_e32 v1, v2, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: idot2_MultipleUses_add1:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v2, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: s_ashr_i32 s4, s0, 16
|
|
; GFX10-DL-NEXT: s_ashr_i32 s5, s1, 16
|
|
; GFX10-DL-NEXT: s_sext_i32_i16 s0, s0
|
|
; GFX10-DL-NEXT: s_sext_i32_i16 s1, s1
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v0, s5, s4, v0
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v1, s1, s0, v0
|
|
; GFX10-DL-NEXT: v_add_nc_u32_e32 v0, v1, v0
|
|
; GFX10-DL-NEXT: global_store_dword v2, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = sext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = sext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = sext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = sext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add1 = add i32 %mul2, %s3
|
|
%add2 = add i32 %add1, %mul1
|
|
|
|
%res = add i32 %add2, %add1
|
|
store i32 %res, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @udot2_MultipleUses_mul1(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: udot2_MultipleUses_mul1:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX7-NEXT: s_lshr_b32 s7, s5, 16
|
|
; GFX7-NEXT: s_and_b32 s4, s4, s8
|
|
; GFX7-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX7-NEXT: s_load_dword s8, s[0:1], 0x0
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX7-NEXT: v_mad_u32_u24 v1, s5, v0, v1
|
|
; GFX7-NEXT: v_mad_u32_u24 v1, s7, v2, v1
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s5, v0, v1
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: udot2_MultipleUses_mul1:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s3, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s5, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s6, s3, s2
|
|
; GFX8-NEXT: s_and_b32 s2, s4, s2
|
|
; GFX8-NEXT: s_lshr_b32 s3, s3, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX8-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s2, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v2, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s4, v2, v0
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s2, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: udot2_MultipleUses_mul1:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v2, s5, v1, v2
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v3, s0
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v2, s1, v3, v2
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s5, v1, v2
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: udot2_MultipleUses_mul1:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-DL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-DL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-DL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v2, s5, v1, v2
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v3, s0
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v2, s1, v3, v2
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s5, v1, v2
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: udot2_MultipleUses_mul1:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_mov_b32 s4, 0xffff
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: s_and_b32 s5, s0, s4
|
|
; GFX10-DL-NEXT: s_and_b32 s4, s1, s4
|
|
; GFX10-DL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX10-DL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s4, s5, v0
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s4, s5, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add0 = add i32 %mul1, %s3
|
|
|
|
%add1 = add i32 %mul2, %add0
|
|
%add2 = add i32 %add1, %mul1
|
|
|
|
store i32 %add2, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @idot2_MultipleUses_mul1(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: idot2_MultipleUses_mul1:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_load_dword s6, s[0:1], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_sext_i32_i16 s7, s4
|
|
; GFX7-NEXT: s_sext_i32_i16 s8, s5
|
|
; GFX7-NEXT: s_ashr_i32 s4, s4, 16
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s7
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: s_ashr_i32 s5, s5, 16
|
|
; GFX7-NEXT: v_mad_i32_i24 v1, s8, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX7-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, s8, v0, v1
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: idot2_MultipleUses_mul1:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s3, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_sext_i32_i16 s5, s2
|
|
; GFX8-NEXT: s_sext_i32_i16 s6, s3
|
|
; GFX8-NEXT: s_ashr_i32 s2, s2, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX8-NEXT: s_ashr_i32 s3, s3, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v2, s2
|
|
; GFX8-NEXT: v_mad_i32_i24 v0, s6, v1, v0
|
|
; GFX8-NEXT: v_mad_i32_i24 v0, s3, v2, v0
|
|
; GFX8-NEXT: v_mad_i32_i24 v2, s6, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: idot2_MultipleUses_mul1:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s5, s1
|
|
; GFX9-NODL-NEXT: s_ashr_i32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: s_ashr_i32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v3, s0
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s1, v3, v1
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: idot2_MultipleUses_mul1:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-DL-NEXT: s_sext_i32_i16 s5, s1
|
|
; GFX9-DL-NEXT: s_ashr_i32 s0, s0, 16
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-DL-NEXT: s_ashr_i32 s1, s1, 16
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v3, s0
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, s1, v3, v1
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: idot2_MultipleUses_mul1:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX10-DL-NEXT: s_sext_i32_i16 s5, s1
|
|
; GFX10-DL-NEXT: s_ashr_i32 s0, s0, 16
|
|
; GFX10-DL-NEXT: s_ashr_i32 s1, s1, 16
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v0, s5, s4, v0
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v0, s5, s4, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = sext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = sext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = sext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = sext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add0 = add i32 %mul1, %s3
|
|
|
|
%add1 = add i32 %mul2, %add0
|
|
%add2 = add i32 %add1, %mul1
|
|
|
|
store i32 %add2, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @udot2_MultipleUses_mul2(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: udot2_MultipleUses_mul2:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX7-NEXT: s_lshr_b32 s7, s5, 16
|
|
; GFX7-NEXT: s_and_b32 s4, s4, s8
|
|
; GFX7-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX7-NEXT: s_load_dword s8, s[0:1], 0x0
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s6
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX7-NEXT: v_mad_u32_u24 v1, s7, v0, v1
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s7, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s4
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s5, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: udot2_MultipleUses_mul2:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_mov_b32 s2, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s3, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s5, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s6, s3, s2
|
|
; GFX8-NEXT: s_lshr_b32 s3, s3, 16
|
|
; GFX8-NEXT: s_and_b32 s2, s4, s2
|
|
; GFX8-NEXT: s_lshr_b32 s4, s4, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s5
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s4, v1, v0
|
|
; GFX8-NEXT: v_mad_u32_u24 v0, s4, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s2, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: udot2_MultipleUses_mul2:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v2, s1, v1, v2
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, v1, v2
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: udot2_MultipleUses_mul2:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s9, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_and_b32 s4, s0, s8
|
|
; GFX9-DL-NEXT: s_lshr_b32 s0, s0, 16
|
|
; GFX9-DL-NEXT: s_and_b32 s5, s1, s8
|
|
; GFX9-DL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s0
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s9
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v2, s1, v1, v2
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s1, v1, v2
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-DL-NEXT: v_mad_u32_u24 v1, s5, v2, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: udot2_MultipleUses_mul2:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_mov_b32 s6, 0xffff
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: s_lshr_b32 s4, s0, 16
|
|
; GFX10-DL-NEXT: s_lshr_b32 s5, s1, 16
|
|
; GFX10-DL-NEXT: s_and_b32 s0, s0, s6
|
|
; GFX10-DL-NEXT: s_and_b32 s1, s1, s6
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s5, s4, v0
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s5, s4, v0
|
|
; GFX10-DL-NEXT: v_mad_u32_u24 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = zext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = zext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = zext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = zext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add0 = add i32 %mul2, %s3
|
|
|
|
%add1 = add i32 %mul2, %add0
|
|
%add2 = add i32 %add1, %mul1
|
|
|
|
store i32 %add2, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @idot2_MultipleUses_mul2(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: idot2_MultipleUses_mul2:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_load_dword s6, s[0:1], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_sext_i32_i16 s7, s4
|
|
; GFX7-NEXT: s_ashr_i32 s4, s4, 16
|
|
; GFX7-NEXT: s_sext_i32_i16 s8, s5
|
|
; GFX7-NEXT: s_ashr_i32 s5, s5, 16
|
|
; GFX7-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s6
|
|
; GFX7-NEXT: v_mad_i32_i24 v1, s5, v0, v1
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, s5, v0, v1
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s7
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, s8, v1, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: idot2_MultipleUses_mul2:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_load_dword s2, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s3, s[6:7], 0x0
|
|
; GFX8-NEXT: s_load_dword s4, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_sext_i32_i16 s5, s2
|
|
; GFX8-NEXT: s_ashr_i32 s2, s2, 16
|
|
; GFX8-NEXT: s_sext_i32_i16 s6, s3
|
|
; GFX8-NEXT: s_ashr_i32 s3, s3, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s2
|
|
; GFX8-NEXT: v_mad_i32_i24 v0, s3, v1, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX8-NEXT: v_mad_i32_i24 v0, s3, v1, v0
|
|
; GFX8-NEXT: v_mad_i32_i24 v2, s6, v2, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: idot2_MultipleUses_mul2:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-NODL-NEXT: s_ashr_i32 s0, s0, 16
|
|
; GFX9-NODL-NEXT: s_sext_i32_i16 s5, s1
|
|
; GFX9-NODL-NEXT: s_ashr_i32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s1, v2, v1
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s1, v2, v1
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: idot2_MultipleUses_mul2:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_sext_i32_i16 s4, s0
|
|
; GFX9-DL-NEXT: s_ashr_i32 s0, s0, 16
|
|
; GFX9-DL-NEXT: s_sext_i32_i16 s5, s1
|
|
; GFX9-DL-NEXT: s_ashr_i32 s1, s1, 16
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v1, s8
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s0
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, s1, v2, v1
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, s1, v2, v1
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, s5, v2, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: idot2_MultipleUses_mul2:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v1, 0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_load_dword s8, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, s8
|
|
; GFX10-DL-NEXT: s_ashr_i32 s4, s0, 16
|
|
; GFX10-DL-NEXT: s_ashr_i32 s5, s1, 16
|
|
; GFX10-DL-NEXT: s_sext_i32_i16 s0, s0
|
|
; GFX10-DL-NEXT: s_sext_i32_i16 s1, s1
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v0, s5, s4, v0
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v0, s5, s4, v0
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v0, s1, s0, v0
|
|
; GFX10-DL-NEXT: global_store_dword v1, v0, s[2:3]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%vec2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i16> %vec1, i64 0
|
|
%conv = sext i16 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i16> %vec2, i64 0
|
|
%conv2 = sext i16 %s2.elt1 to i32
|
|
%mul1 = mul i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i16> %vec1, i64 1
|
|
%conv3 = sext i16 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i16> %vec2, i64 1
|
|
%conv4 = sext i16 %s2.elt2 to i32
|
|
%mul2 = mul i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add0 = add i32 %mul2, %s3
|
|
|
|
%add1 = add i32 %mul2, %add0
|
|
%add2 = add i32 %add1, %mul1
|
|
|
|
store i32 %add2, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @udot2_acc16(<2 x i16> addrspace(1)* %src1,
|
|
; GFX7-LABEL: udot2_acc16:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_mov_b32 s8, 0xffff
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_load_dword s4, s[4:5], 0x0
|
|
; GFX7-NEXT: s_load_dword s5, s[6:7], 0x0
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_lshr_b32 s6, s4, 16
|
|
; GFX7-NEXT: s_lshr_b32 s7, s5, 16
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s7
|
|
; GFX7-NEXT: s_and_b32 s5, s5, s8
|
|
; GFX7-NEXT: s_and_b32 s4, s4, s8
|
|
; GFX7-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s6, v1, v0
|
|
; GFX7-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX7-NEXT: v_mad_u32_u24 v0, s4, v1, v0
|
|
; GFX7-NEXT: buffer_store_short v0, off, s[0:3], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: udot2_acc16:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_load_ushort v2, v[0:1]
|
|
; GFX8-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX8-NEXT: s_load_dword s2, s[6:7], 0x0
|
|
; GFX8-NEXT: s_mov_b32 s0, 0xffff
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: s_and_b32 s3, s2, s0
|
|
; GFX8-NEXT: s_lshr_b32 s2, s2, 16
|
|
; GFX8-NEXT: s_and_b32 s0, s1, s0
|
|
; GFX8-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX8-NEXT: v_mov_b32_e32 v3, s2
|
|
; GFX8-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s1, v3, v2
|
|
; GFX8-NEXT: v_mov_b32_e32 v3, s3
|
|
; GFX8-NEXT: v_mad_u32_u24 v2, s0, v3, v2
|
|
; GFX8-NEXT: flat_store_short v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: udot2_acc16:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_mov_b32 s0, 0xffff
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: global_load_ushort v1, v0, s[2:3]
|
|
; GFX9-NODL-NEXT: s_load_dword s1, s[4:5], 0x0
|
|
; GFX9-NODL-NEXT: s_load_dword s8, s[6:7], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s5, s8, 16
|
|
; GFX9-NODL-NEXT: s_and_b32 s4, s8, s0
|
|
; GFX9-NODL-NEXT: s_and_b32 s0, s1, s0
|
|
; GFX9-NODL-NEXT: s_lshr_b32 s1, s1, 16
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s5
|
|
; GFX9-NODL-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s1, v2, v1
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v2, s4
|
|
; GFX9-NODL-NEXT: v_mad_u32_u24 v1, s0, v2, v1
|
|
; GFX9-NODL-NEXT: global_store_short v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: udot2_acc16:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[4:5], 0x0
|
|
; GFX9-DL-NEXT: s_load_dword s1, s[6:7], 0x0
|
|
; GFX9-DL-NEXT: global_load_ushort v1, v0, s[2:3]
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v2, s1
|
|
; GFX9-DL-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-DL-NEXT: v_dot2_u32_u16 v1, s0, v2, v1
|
|
; GFX9-DL-NEXT: global_store_short v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: udot2_acc16:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: global_load_ushort v1, v0, s[4:5]
|
|
; GFX10-DL-NEXT: s_load_dword s6, s[0:1], 0x0
|
|
; GFX10-DL-NEXT: s_load_dword s7, s[2:3], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_dot2_u32_u16 v1, s6, s7, v1
|
|
; GFX10-DL-NEXT: global_store_short v0, v1, s[4:5]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i16> addrspace(1)* %src2,
|
|
i16 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%v1 = load <2 x i16>, <2 x i16> addrspace(1)* %src1
|
|
%v2 = load <2 x i16>, <2 x i16> addrspace(1)* %src2
|
|
|
|
%v1e1 = extractelement <2 x i16> %v1, i64 0
|
|
%v2e1 = extractelement <2 x i16> %v2, i64 0
|
|
%mul1 = mul i16 %v1e1, %v2e1
|
|
|
|
%v1e2 = extractelement <2 x i16> %v1, i64 1
|
|
%v2e2 = extractelement <2 x i16> %v2, i64 1
|
|
%mul2 = mul i16 %v1e2, %v2e2
|
|
|
|
%s2 = load i16, i16 addrspace(1)* %dst, align 2
|
|
%add1 = add i16 %mul2, %s2
|
|
%add2 = add i16 %add1, %mul1
|
|
store i16 %add2, i16 addrspace(1)* %dst, align 2
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @notsdot2_sext8(<2 x i8> addrspace(1)* %src1,
|
|
; GFX7-LABEL: notsdot2_sext8:
|
|
; GFX7: ; %bb.0: ; %entry
|
|
; GFX7-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
|
|
; GFX7-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0xd
|
|
; GFX7-NEXT: s_mov_b32 s3, 0xf000
|
|
; GFX7-NEXT: s_mov_b32 s2, -1
|
|
; GFX7-NEXT: s_mov_b32 s10, s2
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: s_mov_b32 s0, s4
|
|
; GFX7-NEXT: s_mov_b32 s1, s5
|
|
; GFX7-NEXT: s_mov_b32 s4, s6
|
|
; GFX7-NEXT: s_mov_b32 s5, s7
|
|
; GFX7-NEXT: s_mov_b32 s6, s2
|
|
; GFX7-NEXT: s_mov_b32 s7, s3
|
|
; GFX7-NEXT: buffer_load_ushort v0, off, s[0:3], 0
|
|
; GFX7-NEXT: buffer_load_ushort v1, off, s[4:7], 0
|
|
; GFX7-NEXT: s_load_dword s0, s[8:9], 0x0
|
|
; GFX7-NEXT: s_mov_b32 s11, s3
|
|
; GFX7-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX7-NEXT: v_bfe_i32 v2, v0, 0, 8
|
|
; GFX7-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX7-NEXT: v_bfe_i32 v3, v1, 0, 8
|
|
; GFX7-NEXT: v_bfe_i32 v0, v0, 8, 8
|
|
; GFX7-NEXT: v_bfe_i32 v1, v1, 8, 8
|
|
; GFX7-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, v1, v0, s0
|
|
; GFX7-NEXT: v_mad_i32_i24 v0, v3, v2, v0
|
|
; GFX7-NEXT: buffer_store_dword v0, off, s[8:11], 0
|
|
; GFX7-NEXT: s_endpgm
|
|
;
|
|
; GFX8-LABEL: notsdot2_sext8:
|
|
; GFX8: ; %bb.0: ; %entry
|
|
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s4
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s5
|
|
; GFX8-NEXT: v_mov_b32_e32 v2, s6
|
|
; GFX8-NEXT: v_mov_b32_e32 v3, s7
|
|
; GFX8-NEXT: flat_load_ushort v0, v[0:1]
|
|
; GFX8-NEXT: flat_load_ushort v1, v[2:3]
|
|
; GFX8-NEXT: s_load_dword s2, s[0:1], 0x0
|
|
; GFX8-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX8-NEXT: v_bfe_i32 v2, v0, 0, 8
|
|
; GFX8-NEXT: v_lshrrev_b16_e32 v0, 8, v0
|
|
; GFX8-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX8-NEXT: v_bfe_i32 v3, v1, 0, 8
|
|
; GFX8-NEXT: v_lshrrev_b16_e32 v1, 8, v1
|
|
; GFX8-NEXT: v_bfe_i32 v0, v0, 0, 8
|
|
; GFX8-NEXT: v_bfe_i32 v1, v1, 0, 8
|
|
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX8-NEXT: v_mad_i32_i24 v0, v1, v0, s2
|
|
; GFX8-NEXT: v_mad_i32_i24 v2, v3, v2, v0
|
|
; GFX8-NEXT: v_mov_b32_e32 v0, s0
|
|
; GFX8-NEXT: v_mov_b32_e32 v1, s1
|
|
; GFX8-NEXT: flat_store_dword v[0:1], v2
|
|
; GFX8-NEXT: s_endpgm
|
|
;
|
|
; GFX9-NODL-LABEL: notsdot2_sext8:
|
|
; GFX9-NODL: ; %bb.0: ; %entry
|
|
; GFX9-NODL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-NODL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-NODL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: global_load_ushort v1, v0, s[4:5]
|
|
; GFX9-NODL-NEXT: global_load_ushort v2, v0, s[6:7]
|
|
; GFX9-NODL-NEXT: s_load_dword s0, s[2:3], 0x0
|
|
; GFX9-NODL-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX9-NODL-NEXT: v_bfe_i32 v3, v1, 0, 8
|
|
; GFX9-NODL-NEXT: v_lshrrev_b16_e32 v1, 8, v1
|
|
; GFX9-NODL-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-NODL-NEXT: v_bfe_i32 v4, v2, 0, 8
|
|
; GFX9-NODL-NEXT: v_lshrrev_b16_e32 v2, 8, v2
|
|
; GFX9-NODL-NEXT: v_bfe_i32 v1, v1, 0, 8
|
|
; GFX9-NODL-NEXT: v_bfe_i32 v2, v2, 0, 8
|
|
; GFX9-NODL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, v2, v1, s0
|
|
; GFX9-NODL-NEXT: v_mad_i32_i24 v1, v4, v3, v1
|
|
; GFX9-NODL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-NODL-NEXT: s_endpgm
|
|
;
|
|
; GFX9-DL-LABEL: notsdot2_sext8:
|
|
; GFX9-DL: ; %bb.0: ; %entry
|
|
; GFX9-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX9-DL-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x34
|
|
; GFX9-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: global_load_ushort v1, v0, s[4:5]
|
|
; GFX9-DL-NEXT: global_load_ushort v2, v0, s[6:7]
|
|
; GFX9-DL-NEXT: s_load_dword s0, s[2:3], 0x0
|
|
; GFX9-DL-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX9-DL-NEXT: v_bfe_i32 v3, v1, 0, 8
|
|
; GFX9-DL-NEXT: v_lshrrev_b16_e32 v1, 8, v1
|
|
; GFX9-DL-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX9-DL-NEXT: v_bfe_i32 v4, v2, 0, 8
|
|
; GFX9-DL-NEXT: v_lshrrev_b16_e32 v2, 8, v2
|
|
; GFX9-DL-NEXT: v_bfe_i32 v1, v1, 0, 8
|
|
; GFX9-DL-NEXT: v_bfe_i32 v2, v2, 0, 8
|
|
; GFX9-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, v2, v1, s0
|
|
; GFX9-DL-NEXT: v_mad_i32_i24 v1, v4, v3, v1
|
|
; GFX9-DL-NEXT: global_store_dword v0, v1, s[2:3]
|
|
; GFX9-DL-NEXT: s_endpgm
|
|
;
|
|
; GFX10-DL-LABEL: notsdot2_sext8:
|
|
; GFX10-DL: ; %bb.0: ; %entry
|
|
; GFX10-DL-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
|
|
; GFX10-DL-NEXT: v_mov_b32_e32 v0, 0
|
|
; GFX10-DL-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: s_clause 0x1
|
|
; GFX10-DL-NEXT: global_load_ushort v1, v0, s[4:5]
|
|
; GFX10-DL-NEXT: global_load_ushort v2, v0, s[6:7]
|
|
; GFX10-DL-NEXT: s_load_dword s2, s[0:1], 0x0
|
|
; GFX10-DL-NEXT: s_waitcnt vmcnt(1)
|
|
; GFX10-DL-NEXT: v_lshrrev_b16_e64 v3, 8, v1
|
|
; GFX10-DL-NEXT: s_waitcnt vmcnt(0)
|
|
; GFX10-DL-NEXT: v_lshrrev_b16_e64 v4, 8, v2
|
|
; GFX10-DL-NEXT: v_bfe_i32 v1, v1, 0, 8
|
|
; GFX10-DL-NEXT: v_bfe_i32 v2, v2, 0, 8
|
|
; GFX10-DL-NEXT: v_bfe_i32 v3, v3, 0, 8
|
|
; GFX10-DL-NEXT: v_bfe_i32 v4, v4, 0, 8
|
|
; GFX10-DL-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v3, v4, v3, s2
|
|
; GFX10-DL-NEXT: v_mad_i32_i24 v1, v2, v1, v3
|
|
; GFX10-DL-NEXT: global_store_dword v0, v1, s[0:1]
|
|
; GFX10-DL-NEXT: s_endpgm
|
|
<2 x i8> addrspace(1)* %src2,
|
|
i32 addrspace(1)* nocapture %dst) {
|
|
entry:
|
|
%vec1 = load <2 x i8>, <2 x i8> addrspace(1)* %src1
|
|
%vec2 = load <2 x i8>, <2 x i8> addrspace(1)* %src2
|
|
|
|
%s1.elt1 = extractelement <2 x i8> %vec1, i64 0
|
|
%conv = sext i8 %s1.elt1 to i32
|
|
%s2.elt1 = extractelement <2 x i8> %vec2, i64 0
|
|
%conv2 = sext i8 %s2.elt1 to i32
|
|
%mul1 = mul nuw i32 %conv2, %conv
|
|
|
|
%s1.elt2 = extractelement <2 x i8> %vec1, i64 1
|
|
%conv3 = sext i8 %s1.elt2 to i32
|
|
%s2.elt2 = extractelement <2 x i8> %vec2, i64 1
|
|
%conv4 = sext i8 %s2.elt2 to i32
|
|
%mul2 = mul nuw i32 %conv4, %conv3
|
|
|
|
%s3 = load i32, i32 addrspace(1)* %dst, align 4
|
|
%add = add i32 %mul2, %s3
|
|
%add6 = add i32 %add, %mul1
|
|
store i32 %add6, i32 addrspace(1)* %dst, align 4
|
|
ret void
|
|
}
|