[llvm] [CodeGen] [AMDGPU] Adds pre-commit test for fmul-select combine (PR #111107)

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Thu Oct 10 00:27:05 PDT 2024


================
@@ -0,0 +1,1191 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+;RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s
+;RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1030 < %s | FileCheck -check-prefix=GFX1030 %s
+;RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefix=GFX1100 %s
+
+define float @fmul_select_f32_test1(float %x, i32 %bool.arg1, i32 %bool.arg2) {
+; GFX9-LABEL: fmul_select_f32_test1:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, 1.0, 2.0, vcc
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-LABEL: fmul_select_f32_test1:
+; GFX1030:       ; %bb.0:
+; GFX1030-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1030-NEXT:    v_cndmask_b32_e64 v1, 1.0, 2.0, vcc_lo
+; GFX1030-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1030-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1100-LABEL: fmul_select_f32_test1:
+; GFX1100:       ; %bb.0:
+; GFX1100-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1100-NEXT:    v_cndmask_b32_e64 v1, 1.0, 2.0, vcc_lo
+; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1100-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1100-NEXT:    s_setpc_b64 s[30:31]
+  %bool = icmp eq i32 %bool.arg1, %bool.arg2
+  %y = select i1 %bool, float 2.000000e+00, float 1.000000e+00
+  %ldexp = fmul float %x, %y
+  ret float %ldexp
+}
+
+define float @fmul_select_f32_test2(float %x, i32 %bool.arg1, i32 %bool.arg2) {
+; GFX9-LABEL: fmul_select_f32_test2:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, 1.0, 0.5, vcc
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-LABEL: fmul_select_f32_test2:
+; GFX1030:       ; %bb.0:
+; GFX1030-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1030-NEXT:    v_cndmask_b32_e64 v1, 1.0, 0.5, vcc_lo
+; GFX1030-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1030-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1100-LABEL: fmul_select_f32_test2:
+; GFX1100:       ; %bb.0:
+; GFX1100-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1100-NEXT:    v_cndmask_b32_e64 v1, 1.0, 0.5, vcc_lo
+; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1100-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1100-NEXT:    s_setpc_b64 s[30:31]
+  %bool = icmp eq i32 %bool.arg1, %bool.arg2
+  %y = select i1 %bool, float 5.000000e-01, float 1.000000e+00
+  %ldexp = fmul float %x, %y
+  ret float %ldexp
+}
+
+define <2 x float> @fmul_select_v2f32_test3(<2 x float> %x, <2 x i32> %bool.arg1, <2 x i32> %bool.arg2) {
+; GFX9-LABEL: fmul_select_v2f32_test3:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v5
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, 1.0, 2.0, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, 1.0, 2.0, vcc
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v2
+; GFX9-NEXT:    v_mul_f32_e32 v1, v1, v3
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-LABEL: fmul_select_v2f32_test3:
+; GFX1030:       ; %bb.0:
+; GFX1030-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v4
+; GFX1030-NEXT:    v_cndmask_b32_e64 v2, 1.0, 2.0, vcc_lo
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v3, v5
+; GFX1030-NEXT:    v_mul_f32_e32 v0, v0, v2
+; GFX1030-NEXT:    v_cndmask_b32_e64 v3, 1.0, 2.0, vcc_lo
+; GFX1030-NEXT:    v_mul_f32_e32 v1, v1, v3
+; GFX1030-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1100-LABEL: fmul_select_v2f32_test3:
+; GFX1100:       ; %bb.0:
+; GFX1100-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v4
+; GFX1100-NEXT:    v_cndmask_b32_e64 v2, 1.0, 2.0, vcc_lo
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v3, v5
+; GFX1100-NEXT:    v_cndmask_b32_e64 v3, 1.0, 2.0, vcc_lo
+; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1100-NEXT:    v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX1100-NEXT:    s_setpc_b64 s[30:31]
+  %bool = icmp eq <2 x i32> %bool.arg1, %bool.arg2
+  %y = select <2 x i1> %bool, <2 x float> <float 2.000000e+00, float 2.000000e+00>, <2 x float> <float 1.000000e+00, float 1.000000e+00>
+  %ldexp = fmul <2 x float> %x, %y
+  ret <2 x float> %ldexp
+}
+
+define <2 x float> @fmul_select_v2f32_test4(<2 x float> %x, <2 x i32> %bool.arg1, <2 x i32> %bool.arg2) {
+; GFX9-LABEL: fmul_select_v2f32_test4:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v5
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, 1.0, 0.5, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, 1.0, 0.5, vcc
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v2
+; GFX9-NEXT:    v_mul_f32_e32 v1, v1, v3
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-LABEL: fmul_select_v2f32_test4:
+; GFX1030:       ; %bb.0:
+; GFX1030-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v4
+; GFX1030-NEXT:    v_cndmask_b32_e64 v2, 1.0, 0.5, vcc_lo
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v3, v5
+; GFX1030-NEXT:    v_mul_f32_e32 v0, v0, v2
+; GFX1030-NEXT:    v_cndmask_b32_e64 v3, 1.0, 0.5, vcc_lo
+; GFX1030-NEXT:    v_mul_f32_e32 v1, v1, v3
+; GFX1030-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1100-LABEL: fmul_select_v2f32_test4:
+; GFX1100:       ; %bb.0:
+; GFX1100-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v2, v4
+; GFX1100-NEXT:    v_cndmask_b32_e64 v2, 1.0, 0.5, vcc_lo
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v3, v5
+; GFX1100-NEXT:    v_cndmask_b32_e64 v3, 1.0, 0.5, vcc_lo
+; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1100-NEXT:    v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX1100-NEXT:    s_setpc_b64 s[30:31]
+  %bool = icmp eq <2 x i32> %bool.arg1, %bool.arg2
+  %y = select <2 x i1> %bool, <2 x float> <float 5.000000e-01, float 5.000000e-01>, <2 x float> <float 1.000000e+00, float 1.000000e+00>
+  %ldexp = fmul <2 x float> %x, %y
+  ret <2 x float> %ldexp
+}
+
+define float @fmul_select_f32_test5(float %x, i32 %bool.arg1, i32 %bool.arg2) {
+; GFX9-LABEL: fmul_select_f32_test5:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, -1.0, -2.0, vcc
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-LABEL: fmul_select_f32_test5:
+; GFX1030:       ; %bb.0:
+; GFX1030-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1030-NEXT:    v_cndmask_b32_e64 v1, -1.0, -2.0, vcc_lo
+; GFX1030-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1030-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1100-LABEL: fmul_select_f32_test5:
+; GFX1100:       ; %bb.0:
+; GFX1100-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1100-NEXT:    v_cndmask_b32_e64 v1, -1.0, -2.0, vcc_lo
+; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1100-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1100-NEXT:    s_setpc_b64 s[30:31]
+  %bool = icmp eq i32 %bool.arg1, %bool.arg2
+  %y = select i1 %bool, float -2.000000e+00, float -1.000000e+00
+  %ldexp = fmul float %x, %y
+  ret float %ldexp
+}
+
+define float @fmul_select_f32_test6(float %x, i32 %bool.arg1, i32 %bool.arg2) {
+; GFX9-LABEL: fmul_select_f32_test6:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0x41000000
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0xc0400000
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-LABEL: fmul_select_f32_test6:
+; GFX1030:       ; %bb.0:
+; GFX1030-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-NEXT:    v_mov_b32_e32 v3, 0xc0400000
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1030-NEXT:    v_cndmask_b32_e32 v1, 0x41000000, v3, vcc_lo
+; GFX1030-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1030-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1100-LABEL: fmul_select_f32_test6:
+; GFX1100:       ; %bb.0:
+; GFX1100-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT:    v_mov_b32_e32 v3, 0xc0400000
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1100-NEXT:    v_cndmask_b32_e32 v1, 0x41000000, v3, vcc_lo
+; GFX1100-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1100-NEXT:    s_setpc_b64 s[30:31]
+  %bool = icmp eq i32 %bool.arg1, %bool.arg2
+  %y = select i1 %bool, float -3.000000e+00, float 8.000000e+00
+  %ldexp = fmul float %x, %y
+  ret float %ldexp
+}
+
+define float @fmul_select_f32_test7(float %x, i32 %bool.arg1, i32 %bool.arg2) {
+; GFX9-LABEL: fmul_select_f32_test7:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0x41000000
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v3, 4.0, vcc
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-LABEL: fmul_select_f32_test7:
+; GFX1030:       ; %bb.0:
+; GFX1030-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1030-NEXT:    v_cndmask_b32_e64 v1, 0x41000000, 4.0, vcc_lo
+; GFX1030-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1030-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1100-LABEL: fmul_select_f32_test7:
+; GFX1100:       ; %bb.0:
+; GFX1100-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1100-NEXT:    v_cndmask_b32_e64 v1, 0x41000000, 4.0, vcc_lo
+; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1100-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1100-NEXT:    s_setpc_b64 s[30:31]
+  %bool = icmp eq i32 %bool.arg1, %bool.arg2
+  %y = select i1 %bool, float 4.000000e+00, float 8.000000e+00
+  %ldexp = fmul float %x, %y
+  ret float %ldexp
+}
+
+define float @fmul_select_f32_test8(float %x, i32 %bool.arg1, i32 %bool.arg2) {
+; GFX9-LABEL: fmul_select_f32_test8:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0xc1000000
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0x41800000
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-LABEL: fmul_select_f32_test8:
+; GFX1030:       ; %bb.0:
+; GFX1030-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-NEXT:    v_mov_b32_e32 v3, 0x41800000
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1030-NEXT:    v_cndmask_b32_e32 v1, 0xc1000000, v3, vcc_lo
+; GFX1030-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1030-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1100-LABEL: fmul_select_f32_test8:
+; GFX1100:       ; %bb.0:
+; GFX1100-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT:    v_mov_b32_e32 v3, 0x41800000
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1100-NEXT:    v_cndmask_b32_e32 v1, 0xc1000000, v3, vcc_lo
+; GFX1100-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1100-NEXT:    s_setpc_b64 s[30:31]
+  %bool = icmp eq i32 %bool.arg1, %bool.arg2
+  %y = select i1 %bool, float 1.600000e+01, float -8.000000e+00
+  %ldexp = fmul float %x, %y
+  ret float %ldexp
+}
+
+define float @fmul_select_f32_test9(float %x, i32 %bool.arg1, i32 %bool.arg2) {
+; GFX9-LABEL: fmul_select_f32_test9:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, 2.0, 0, vcc
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-LABEL: fmul_select_f32_test9:
+; GFX1030:       ; %bb.0:
+; GFX1030-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1030-NEXT:    v_cndmask_b32_e64 v1, 2.0, 0, vcc_lo
+; GFX1030-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1030-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1100-LABEL: fmul_select_f32_test9:
+; GFX1100:       ; %bb.0:
+; GFX1100-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1100-NEXT:    v_cndmask_b32_e64 v1, 2.0, 0, vcc_lo
+; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1100-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1100-NEXT:    s_setpc_b64 s[30:31]
+  %bool = icmp eq i32 %bool.arg1, %bool.arg2
+  %y = select i1 %bool, float 0.000000e+00, float 2.000000e+00
+  %ldexp = fmul float %x, %y
+  ret float %ldexp
+}
+
+define float @fmul_select_f32_test10(float %x, i32 %bool.arg1, i32 %bool.arg2) {
+; GFX9-LABEL: fmul_select_f32_test10:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, 0, v3, vcc
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX9-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-LABEL: fmul_select_f32_test10:
+; GFX1030:       ; %bb.0:
+; GFX1030-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1030-NEXT:    v_cndmask_b32_e64 v1, 0, 0x80000000, vcc_lo
+; GFX1030-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1030-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1100-LABEL: fmul_select_f32_test10:
+; GFX1100:       ; %bb.0:
+; GFX1100-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT:    v_cmp_eq_u32_e32 vcc_lo, v1, v2
+; GFX1100-NEXT:    v_cndmask_b32_e64 v1, 0, 0x80000000, vcc_lo
+; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1100-NEXT:    v_mul_f32_e32 v0, v0, v1
+; GFX1100-NEXT:    s_setpc_b64 s[30:31]
+  %bool = icmp eq i32 %bool.arg1, %bool.arg2
+  %y = select i1 %bool, float -0.000000e+00, float 0.000000e+00
+  %ldexp = fmul float %x, %y
+  ret float %ldexp
+}
+
+
+
----------------
arsenm wrote:

```suggestion
```

https://github.com/llvm/llvm-project/pull/111107


More information about the llvm-commits mailing list