blob: 7eca56c22aa5d324d9f5ef4dee14d2daae4d6ab6 [file]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -global-isel=0 -mtriple=amdgpu12.00 < %s | FileCheck --check-prefixes=GCN,SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgpu12.00 < %s | FileCheck --check-prefixes=GCN,GISEL %s
; RUN: llc -global-isel=0 -mtriple=amdgpu12.50 < %s | FileCheck --check-prefixes=GFX1250-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgpu12.50 < %s | FileCheck --check-prefixes=GFX1250-GISEL %s
; RUN: llc -global-isel=0 -mtriple=amdgpu13.10 < %s | FileCheck --check-prefixes=GFX1310-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgpu13.10 < %s | FileCheck --check-prefixes=GFX1310-GISEL %s
define amdgpu_ps void @prefetch_data_sgpr_base_sgpr_len(ptr addrspace(4) inreg %ptr, i32 inreg %len) {
; GCN-LABEL: prefetch_data_sgpr_base_sgpr_len:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GCN-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_sgpr_base_sgpr_len:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_sgpr_base_sgpr_len:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_sgpr_base_sgpr_len:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_sgpr_base_sgpr_len:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1310-GISEL-NEXT: s_endpgm
entry:
tail call void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %ptr, i32 %len)
ret void
}
define amdgpu_ps void @prefetch_data_sgpr_imm_base_sgpr_len(ptr addrspace(4) inreg %ptr, i32 inreg %len) {
; GCN-LABEL: prefetch_data_sgpr_imm_base_sgpr_len:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_prefetch_data s[0:1], 0x200, s2, 0
; GCN-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_sgpr_imm_base_sgpr_len:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: s_prefetch_data s[0:1], 0x200, s2, 0
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_sgpr_imm_base_sgpr_len:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: s_prefetch_data s[0:1], 0x200, s2, 0
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_sgpr_imm_base_sgpr_len:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: s_prefetch_data s[0:1], 0x200, s2, 0
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_sgpr_imm_base_sgpr_len:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: s_prefetch_data s[0:1], 0x200, s2, 0
; GFX1310-GISEL-NEXT: s_endpgm
entry:
%gep = getelementptr i32, ptr addrspace(4) %ptr, i32 128
tail call void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %gep, i32 %len)
ret void
}
define amdgpu_ps void @prefetch_data_sgpr_base_imm_len(ptr addrspace(4) inreg %ptr) {
; GCN-LABEL: prefetch_data_sgpr_base_imm_len:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GCN-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_sgpr_base_imm_len:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_sgpr_base_imm_len:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_sgpr_base_imm_len:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_sgpr_base_imm_len:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1310-GISEL-NEXT: s_endpgm
entry:
tail call void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %ptr, i32 31)
ret void
}
define amdgpu_ps void @prefetch_data_sgpr_imm_base_imm_len(ptr addrspace(4) inreg %ptr) {
; GCN-LABEL: prefetch_data_sgpr_imm_base_imm_len:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_prefetch_data s[0:1], 0x200, null, 31
; GCN-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_sgpr_imm_base_imm_len:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: s_prefetch_data s[0:1], 0x200, null, 31
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_sgpr_imm_base_imm_len:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: s_prefetch_data s[0:1], 0x200, null, 31
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_sgpr_imm_base_imm_len:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: s_prefetch_data s[0:1], 0x200, null, 31
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_sgpr_imm_base_imm_len:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: s_prefetch_data s[0:1], 0x200, null, 31
; GFX1310-GISEL-NEXT: s_endpgm
entry:
%gep = getelementptr i32, ptr addrspace(4) %ptr, i32 128
tail call void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %gep, i32 31)
ret void
}
define amdgpu_ps void @prefetch_data_vgpr_base_sgpr_len(ptr addrspace(4) %ptr, i32 inreg %len) {
; GCN-LABEL: prefetch_data_vgpr_base_sgpr_len:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: v_readfirstlane_b32 s2, v0
; GCN-NEXT: v_readfirstlane_b32 s3, v1
; GCN-NEXT: s_prefetch_data s[2:3], 0x0, s0, 0
; GCN-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_vgpr_base_sgpr_len:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s2, v0
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s3, v1
; GFX1250-SDAG-NEXT: s_prefetch_data s[2:3], 0x0, s0, 0
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_vgpr_base_sgpr_len:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s2, v0
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s3, v1
; GFX1250-GISEL-NEXT: s_prefetch_data s[2:3], 0x0, s0, 0
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_vgpr_base_sgpr_len:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: v_readfirstlane_b32 s2, v0
; GFX1310-SDAG-NEXT: v_readfirstlane_b32 s3, v1
; GFX1310-SDAG-NEXT: s_prefetch_data s[2:3], 0x0, s0, 0
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_vgpr_base_sgpr_len:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: v_readfirstlane_b32 s2, v0
; GFX1310-GISEL-NEXT: v_readfirstlane_b32 s3, v1
; GFX1310-GISEL-NEXT: s_prefetch_data s[2:3], 0x0, s0, 0
; GFX1310-GISEL-NEXT: s_endpgm
entry:
tail call void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %ptr, i32 %len)
ret void
}
define amdgpu_ps void @prefetch_data_vgpr_imm_base_sgpr_len(ptr addrspace(4) %ptr, i32 inreg %len) {
; SDAG-LABEL: prefetch_data_vgpr_imm_base_sgpr_len:
; SDAG: ; %bb.0: ; %entry
; SDAG-NEXT: v_readfirstlane_b32 s2, v0
; SDAG-NEXT: v_readfirstlane_b32 s3, v1
; SDAG-NEXT: s_prefetch_data s[2:3], 0x200, s0, 0
; SDAG-NEXT: s_endpgm
;
; GISEL-LABEL: prefetch_data_vgpr_imm_base_sgpr_len:
; GISEL: ; %bb.0: ; %entry
; GISEL-NEXT: v_add_co_u32 v0, vcc_lo, 0x200, v0
; GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GISEL-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GISEL-NEXT: v_readfirstlane_b32 s2, v0
; GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GISEL-NEXT: v_readfirstlane_b32 s3, v1
; GISEL-NEXT: s_prefetch_data s[2:3], 0x0, s0, 0
; GISEL-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_vgpr_imm_base_sgpr_len:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s2, v0
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s3, v1
; GFX1250-SDAG-NEXT: s_prefetch_data s[2:3], 0x200, s0, 0
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_vgpr_imm_base_sgpr_len:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: v_add_co_u32 v0, vcc_lo, 0x200, v0
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX1250-GISEL-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s2, v0
; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s3, v1
; GFX1250-GISEL-NEXT: s_prefetch_data s[2:3], 0x0, s0, 0
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_vgpr_imm_base_sgpr_len:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: v_readfirstlane_b32 s2, v0
; GFX1310-SDAG-NEXT: v_readfirstlane_b32 s3, v1
; GFX1310-SDAG-NEXT: s_prefetch_data s[2:3], 0x200, s0, 0
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_vgpr_imm_base_sgpr_len:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: v_add_co_u32 v0, vcc_lo, 0x200, v0
; GFX1310-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX1310-GISEL-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX1310-GISEL-NEXT: v_readfirstlane_b32 s2, v0
; GFX1310-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1310-GISEL-NEXT: v_readfirstlane_b32 s3, v1
; GFX1310-GISEL-NEXT: s_prefetch_data s[2:3], 0x0, s0, 0
; GFX1310-GISEL-NEXT: s_endpgm
entry:
%gep = getelementptr i32, ptr addrspace(4) %ptr, i32 128
tail call void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %gep, i32 %len)
ret void
}
define amdgpu_ps void @prefetch_data_sgpr_base_vgpr_len(ptr addrspace(4) inreg %ptr, i32 %len) {
; GCN-LABEL: prefetch_data_sgpr_base_vgpr_len:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: v_readfirstlane_b32 s2, v0
; GCN-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GCN-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_sgpr_base_vgpr_len:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s2, v0
; GFX1250-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_sgpr_base_vgpr_len:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s2, v0
; GFX1250-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_sgpr_base_vgpr_len:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: v_readfirstlane_b32 s2, v0
; GFX1310-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_sgpr_base_vgpr_len:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: v_readfirstlane_b32 s2, v0
; GFX1310-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1310-GISEL-NEXT: s_endpgm
entry:
tail call void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %ptr, i32 %len)
ret void
}
define amdgpu_ps void @prefetch_data_sgpr_base_imm_len_global(ptr addrspace(1) inreg %ptr) {
; GCN-LABEL: prefetch_data_sgpr_base_imm_len_global:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GCN-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_sgpr_base_imm_len_global:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_sgpr_base_imm_len_global:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_sgpr_base_imm_len_global:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_sgpr_base_imm_len_global:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1310-GISEL-NEXT: s_endpgm
entry:
tail call void @llvm.amdgcn.s.prefetch.data.p1(ptr addrspace(1) %ptr, i32 31)
ret void
}
define amdgpu_ps void @prefetch_data_sgpr_base_imm_len_flat(ptr inreg %ptr) {
; GCN-LABEL: prefetch_data_sgpr_base_imm_len_flat:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GCN-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_sgpr_base_imm_len_flat:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_sgpr_base_imm_len_flat:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_sgpr_base_imm_len_flat:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_sgpr_base_imm_len_flat:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, null, 31
; GFX1310-GISEL-NEXT: s_endpgm
entry:
tail call void @llvm.amdgcn.s.prefetch.data.p0(ptr %ptr, i32 31)
ret void
}
define amdgpu_ps void @prefetch_data_vgpr_base_imm_len(ptr addrspace(4) %ptr) {
; GCN-LABEL: prefetch_data_vgpr_base_imm_len:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: v_readfirstlane_b32 s0, v0
; GCN-NEXT: v_readfirstlane_b32 s1, v1
; GCN-NEXT: s_prefetch_data s[0:1], 0x0, null, 0
; GCN-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_vgpr_base_imm_len:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s1, v1
; GFX1250-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, null, 0
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_vgpr_base_imm_len:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s1, v1
; GFX1250-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, null, 0
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_vgpr_base_imm_len:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX1310-SDAG-NEXT: v_readfirstlane_b32 s1, v1
; GFX1310-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, null, 0
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_vgpr_base_imm_len:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1310-GISEL-NEXT: v_readfirstlane_b32 s1, v1
; GFX1310-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, null, 0
; GFX1310-GISEL-NEXT: s_endpgm
entry:
tail call void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %ptr, i32 0)
ret void
}
define amdgpu_ps void @prefetch_data_vgpr_base_vgpr_len(ptr addrspace(4) %ptr, i32 %len) {
; SDAG-LABEL: prefetch_data_vgpr_base_vgpr_len:
; SDAG: ; %bb.0: ; %entry
; SDAG-NEXT: v_readfirstlane_b32 s2, v2
; SDAG-NEXT: v_readfirstlane_b32 s0, v0
; SDAG-NEXT: v_readfirstlane_b32 s1, v1
; SDAG-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; SDAG-NEXT: s_endpgm
;
; GISEL-LABEL: prefetch_data_vgpr_base_vgpr_len:
; GISEL: ; %bb.0: ; %entry
; GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GISEL-NEXT: v_readfirstlane_b32 s1, v1
; GISEL-NEXT: v_readfirstlane_b32 s2, v2
; GISEL-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GISEL-NEXT: s_endpgm
;
; GFX1250-SDAG-LABEL: prefetch_data_vgpr_base_vgpr_len:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s2, v2
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-SDAG-NEXT: v_readfirstlane_b32 s1, v1
; GFX1250-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1250-SDAG-NEXT: s_endpgm
;
; GFX1250-GISEL-LABEL: prefetch_data_vgpr_base_vgpr_len:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s1, v1
; GFX1250-GISEL-NEXT: v_readfirstlane_b32 s2, v2
; GFX1250-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1250-GISEL-NEXT: s_endpgm
;
; GFX1310-SDAG-LABEL: prefetch_data_vgpr_base_vgpr_len:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: v_readfirstlane_b32 s2, v2
; GFX1310-SDAG-NEXT: v_readfirstlane_b32 s0, v0
; GFX1310-SDAG-NEXT: v_readfirstlane_b32 s1, v1
; GFX1310-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1310-SDAG-NEXT: s_endpgm
;
; GFX1310-GISEL-LABEL: prefetch_data_vgpr_base_vgpr_len:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: v_readfirstlane_b32 s0, v0
; GFX1310-GISEL-NEXT: v_readfirstlane_b32 s1, v1
; GFX1310-GISEL-NEXT: v_readfirstlane_b32 s2, v2
; GFX1310-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, s2, 0
; GFX1310-GISEL-NEXT: s_endpgm
entry:
tail call void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %ptr, i32 %len)
ret void
}
define amdgpu_ps float @prefetch_and_load_b32(ptr addrspace(4) align 4 inreg %p) {
; GCN-LABEL: prefetch_and_load_b32:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_prefetch_data s[0:1], 0x0, null, 0
; GCN-NEXT: s_load_b32 s0, s[0:1], 0x0
; GCN-NEXT: s_wait_kmcnt 0x0
; GCN-NEXT: v_mov_b32_e32 v0, s0
; GCN-NEXT: ; return to shader part epilog
;
; GFX1250-SDAG-LABEL: prefetch_and_load_b32:
; GFX1250-SDAG: ; %bb.0: ; %entry
; GFX1250-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-SDAG-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-SDAG-NEXT: v_nop
; GFX1250-SDAG-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, null, 0
; GFX1250-SDAG-NEXT: s_wait_xcnt 0x0
; GFX1250-SDAG-NEXT: s_load_b32 s0, s[0:1], 0x0 nv
; GFX1250-SDAG-NEXT: s_wait_kmcnt 0x0
; GFX1250-SDAG-NEXT: v_mov_b32_e32 v0, s0
; GFX1250-SDAG-NEXT: ; return to shader part epilog
;
; GFX1250-GISEL-LABEL: prefetch_and_load_b32:
; GFX1250-GISEL: ; %bb.0: ; %entry
; GFX1250-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
; GFX1250-GISEL-NEXT: s_mov_b64 s[64:65], 0
; GFX1250-GISEL-NEXT: v_nop
; GFX1250-GISEL-NEXT: global_prefetch_b8 v0, s[64:65] scope:SCOPE_SE
; GFX1250-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, null, 0
; GFX1250-GISEL-NEXT: s_wait_xcnt 0x0
; GFX1250-GISEL-NEXT: s_load_b32 s0, s[0:1], 0x0 nv
; GFX1250-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX1250-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX1250-GISEL-NEXT: ; return to shader part epilog
;
; GFX1310-SDAG-LABEL: prefetch_and_load_b32:
; GFX1310-SDAG: ; %bb.0: ; %entry
; GFX1310-SDAG-NEXT: s_prefetch_data s[0:1], 0x0, null, 0
; GFX1310-SDAG-NEXT: s_load_b32 s0, s[0:1], 0x0 nv
; GFX1310-SDAG-NEXT: s_wait_kmcnt 0x0
; GFX1310-SDAG-NEXT: v_mov_b32_e32 v0, s0
; GFX1310-SDAG-NEXT: ; return to shader part epilog
;
; GFX1310-GISEL-LABEL: prefetch_and_load_b32:
; GFX1310-GISEL: ; %bb.0: ; %entry
; GFX1310-GISEL-NEXT: s_prefetch_data s[0:1], 0x0, null, 0
; GFX1310-GISEL-NEXT: s_load_b32 s0, s[0:1], 0x0 nv
; GFX1310-GISEL-NEXT: s_wait_kmcnt 0x0
; GFX1310-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX1310-GISEL-NEXT: ; return to shader part epilog
entry:
tail call void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %p, i32 0)
%ret = load float, ptr addrspace(4) %p, align 4
ret float %ret
}
declare void @llvm.amdgcn.s.prefetch.data.p4(ptr addrspace(4) %ptr, i32 %len)
declare void @llvm.amdgcn.s.prefetch.data.p1(ptr addrspace(1) %ptr, i32 %len)
declare void @llvm.amdgcn.s.prefetch.data.p0(ptr %ptr, i32 %len)