Skip to content

Commit 25bf58b

Browse files
Ana MihajlovicAna Mihajlovic
authored andcommitted
rebase
1 parent 51c61e7 commit 25bf58b

File tree

2 files changed

+7
-31
lines changed

2 files changed

+7
-31
lines changed

llvm/test/CodeGen/AMDGPU/dag-divergence.ll

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -77,10 +77,10 @@ define <2 x i128> @wide_carry_divergence_error(i128 %arg) {
7777
; GCN-NEXT: v_add_u32_e32 v0, vcc, 64, v0
7878
; GCN-NEXT: v_ffbh_u32_e32 v5, v3
7979
; GCN-NEXT: v_addc_u32_e64 v1, s[4:5], 0, 0, vcc
80-
; GCN-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[2:3]
80+
; GCN-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3]
8181
; GCN-NEXT: v_min_u32_e32 v4, v4, v5
82-
; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v4, vcc
83-
; GCN-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc
82+
; GCN-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc
83+
; GCN-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
8484
; GCN-NEXT: v_sub_u32_e32 v0, vcc, 0, v0
8585
; GCN-NEXT: v_mov_b32_e32 v3, 0
8686
; GCN-NEXT: v_subb_u32_e32 v1, vcc, 0, v1, vcc

llvm/test/CodeGen/AMDGPU/dagcombine-fmul-sel.ll

Lines changed: 4 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -2219,7 +2219,6 @@ define half @fmul_select_f16_test6(half %x, i32 %bool.arg1, i32 %bool.arg2) {
22192219
; GFX11-SDAG-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
22202220
; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
22212221
;
2222-
<<<<<<< HEAD
22232222
; GFX11-SDAG-FAKE16-LABEL: fmul_select_f16_test6:
22242223
; GFX11-SDAG-FAKE16: ; %bb.0:
22252224
; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -2244,22 +2243,11 @@ define half @fmul_select_f16_test6(half %x, i32 %bool.arg1, i32 %bool.arg2) {
22442243
; GFX11-GISEL-FAKE16: ; %bb.0:
22452244
; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
22462245
; GFX11-GISEL-FAKE16-NEXT: v_mov_b32_e32 v3, 0x4200
2247-
; GFX11-GISEL-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2
2246+
; GFX11-GISEL-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, v1, v2
22482247
; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
2249-
; GFX11-GISEL-FAKE16-NEXT: v_cndmask_b32_e64 v1, v3, 0xc800, vcc_lo
2248+
; GFX11-GISEL-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xc800, v3, vcc_lo
22502249
; GFX11-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
22512250
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
2252-
=======
2253-
; GFX11-GISEL-LABEL: fmul_select_f16_test6:
2254-
; GFX11-GISEL: ; %bb.0:
2255-
; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
2256-
; GFX11-GISEL-NEXT: v_mov_b32_e32 v3, 0x4200
2257-
; GFX11-GISEL-NEXT: v_cmp_ne_u32_e32 vcc_lo, v1, v2
2258-
; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
2259-
; GFX11-GISEL-NEXT: v_cndmask_b32_e32 v1, 0xc800, v3, vcc_lo
2260-
; GFX11-GISEL-NEXT: v_mul_f16_e32 v0, v0, v1
2261-
; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
2262-
>>>>>>> e85470abb9d1 ([AMDGPU] Switch V_CNDMASK operands to shrink it into VOP2)
22632251
%bool = icmp eq i32 %bool.arg1, %bool.arg2
22642252
%y = select i1 %bool, half -8.000000e+00, half 3.000000e+00
22652253
%ldexp = fmul half %x, %y
@@ -2339,7 +2327,6 @@ define half @fmul_select_f16_test7(half %x, i32 %bool.arg1, i32 %bool.arg2) {
23392327
; GFX11-SDAG-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
23402328
; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
23412329
;
2342-
<<<<<<< HEAD
23432330
; GFX11-SDAG-FAKE16-LABEL: fmul_select_f16_test7:
23442331
; GFX11-SDAG-FAKE16: ; %bb.0:
23452332
; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -2364,22 +2351,11 @@ define half @fmul_select_f16_test7(half %x, i32 %bool.arg1, i32 %bool.arg2) {
23642351
; GFX11-GISEL-FAKE16: ; %bb.0:
23652352
; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
23662353
; GFX11-GISEL-FAKE16-NEXT: v_mov_b32_e32 v3, 0xc400
2367-
; GFX11-GISEL-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2
2354+
; GFX11-GISEL-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, v1, v2
23682355
; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
2369-
; GFX11-GISEL-FAKE16-NEXT: v_cndmask_b32_e64 v1, v3, 0x4800, vcc_lo
2356+
; GFX11-GISEL-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4800, v3, vcc_lo
23702357
; GFX11-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
23712358
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
2372-
=======
2373-
; GFX11-GISEL-LABEL: fmul_select_f16_test7:
2374-
; GFX11-GISEL: ; %bb.0:
2375-
; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
2376-
; GFX11-GISEL-NEXT: v_mov_b32_e32 v3, 0xc400
2377-
; GFX11-GISEL-NEXT: v_cmp_ne_u32_e32 vcc_lo, v1, v2
2378-
; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
2379-
; GFX11-GISEL-NEXT: v_cndmask_b32_e32 v1, 0x4800, v3, vcc_lo
2380-
; GFX11-GISEL-NEXT: v_mul_f16_e32 v0, v0, v1
2381-
; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
2382-
>>>>>>> e85470abb9d1 ([AMDGPU] Switch V_CNDMASK operands to shrink it into VOP2)
23832359
%bool = icmp eq i32 %bool.arg1, %bool.arg2
23842360
%y = select i1 %bool, half 8.000000e+00, half -4.000000e+00
23852361
%ldexp = fmul half %x, %y

0 commit comments

Comments
 (0)