| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 |
| ; RUN: llc < %s -mtriple armv8a-none-eabihf -mattr=-bf16 | FileCheck %s --check-prefixes=CHECK,CHECK-CVT |
| ; RUN: llc < %s -mtriple armv8a-none-eabihf -mattr=+bf16 | FileCheck %s --check-prefixes=CHECK,CHECK-BF16 |
| |
| ;define bfloat @test_fadd(bfloat %a, bfloat %b) #0 { |
| ; %r = fadd bfloat %a, %b |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_fadd_fast(bfloat %a, bfloat %b) #0 { |
| ; %r = fadd fast bfloat %a, %b |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_fsub(bfloat %a, bfloat %b) #0 { |
| ; %r = fsub bfloat %a, %b |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_fmul(bfloat %a, bfloat %b) #0 { |
| ; %r = fmul bfloat %a, %b |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_fmadd(bfloat %a, bfloat %b, bfloat %c) #0 { |
| ; %mul = fmul fast bfloat %a, %b |
| ; %r = fadd fast bfloat %mul, %c |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_fdiv(bfloat %a, bfloat %b) #0 { |
| ; %r = fdiv bfloat %a, %b |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_frem(bfloat %a, bfloat %b) #0 { |
| ; %r = frem bfloat %a, %b |
| ; ret bfloat %r |
| ;} |
| |
| define bfloat @test_load(ptr %a) #0 { |
| ; CHECK-LABEL: test_load: |
| ; CHECK: @ %bb.0: |
| ; CHECK-NEXT: ldrh r0, [r0] |
| ; CHECK-NEXT: vmov s0, r0 |
| ; CHECK-NEXT: bx lr |
| %r = load bfloat, ptr %a |
| ret bfloat %r |
| } |
| |
| define void @test_store(bfloat %a, ptr %b) #0 { |
| ; CHECK-LABEL: test_store: |
| ; CHECK: @ %bb.0: |
| ; CHECK-NEXT: vmov r1, s0 |
| ; CHECK-NEXT: strh r1, [r0] |
| ; CHECK-NEXT: bx lr |
| store bfloat %a, ptr %b |
| ret void |
| } |
| |
| declare bfloat @test_callee(bfloat %a, bfloat %b) #0 |
| |
| define bfloat @test_call(bfloat %a, bfloat %b) #0 { |
| ; CHECK-CVT-LABEL: test_call: |
| ; CHECK-CVT: @ %bb.0: |
| ; CHECK-CVT-NEXT: .save {r11, lr} |
| ; CHECK-CVT-NEXT: push {r11, lr} |
| ; CHECK-CVT-NEXT: bl test_callee |
| ; CHECK-CVT-NEXT: pop {r11, pc} |
| ; |
| ; CHECK-BF16-LABEL: test_call: |
| ; CHECK-BF16: @ %bb.0: |
| ; CHECK-BF16-NEXT: .save {r11, lr} |
| ; CHECK-BF16-NEXT: push {r11, lr} |
| ; CHECK-BF16-NEXT: .pad #8 |
| ; CHECK-BF16-NEXT: sub sp, sp, #8 |
| ; CHECK-BF16-NEXT: vmov r0, s1 |
| ; CHECK-BF16-NEXT: vmov r1, s0 |
| ; CHECK-BF16-NEXT: uxth r0, r0 |
| ; CHECK-BF16-NEXT: uxth r1, r1 |
| ; CHECK-BF16-NEXT: vmov s1, r0 |
| ; CHECK-BF16-NEXT: vmov s0, r1 |
| ; CHECK-BF16-NEXT: bl test_callee |
| ; CHECK-BF16-NEXT: vmov r0, s0 |
| ; CHECK-BF16-NEXT: strh r0, [sp, #6] |
| ; CHECK-BF16-NEXT: ldrh r0, [sp, #6] |
| ; CHECK-BF16-NEXT: vmov s0, r0 |
| ; CHECK-BF16-NEXT: add sp, sp, #8 |
| ; CHECK-BF16-NEXT: pop {r11, pc} |
| %r = call bfloat @test_callee(bfloat %a, bfloat %b) |
| ret bfloat %r |
| } |
| |
| define bfloat @test_call_flipped(bfloat %a, bfloat %b) #0 { |
| ; CHECK-CVT-LABEL: test_call_flipped: |
| ; CHECK-CVT: @ %bb.0: |
| ; CHECK-CVT-NEXT: .save {r11, lr} |
| ; CHECK-CVT-NEXT: push {r11, lr} |
| ; CHECK-CVT-NEXT: vmov.f32 s2, s0 |
| ; CHECK-CVT-NEXT: vmov.f32 s0, s1 |
| ; CHECK-CVT-NEXT: vmov.f32 s1, s2 |
| ; CHECK-CVT-NEXT: bl test_callee |
| ; CHECK-CVT-NEXT: pop {r11, pc} |
| ; |
| ; CHECK-BF16-LABEL: test_call_flipped: |
| ; CHECK-BF16: @ %bb.0: |
| ; CHECK-BF16-NEXT: .save {r11, lr} |
| ; CHECK-BF16-NEXT: push {r11, lr} |
| ; CHECK-BF16-NEXT: .pad #8 |
| ; CHECK-BF16-NEXT: sub sp, sp, #8 |
| ; CHECK-BF16-NEXT: vmov r0, s0 |
| ; CHECK-BF16-NEXT: vmov r1, s1 |
| ; CHECK-BF16-NEXT: uxth r0, r0 |
| ; CHECK-BF16-NEXT: uxth r1, r1 |
| ; CHECK-BF16-NEXT: vmov s1, r0 |
| ; CHECK-BF16-NEXT: vmov s0, r1 |
| ; CHECK-BF16-NEXT: bl test_callee |
| ; CHECK-BF16-NEXT: vmov r0, s0 |
| ; CHECK-BF16-NEXT: strh r0, [sp, #6] |
| ; CHECK-BF16-NEXT: ldrh r0, [sp, #6] |
| ; CHECK-BF16-NEXT: vmov s0, r0 |
| ; CHECK-BF16-NEXT: add sp, sp, #8 |
| ; CHECK-BF16-NEXT: pop {r11, pc} |
| %r = call bfloat @test_callee(bfloat %b, bfloat %a) |
| ret bfloat %r |
| } |
| |
| define bfloat @test_tailcall_flipped(bfloat %a, bfloat %b) #0 { |
| ; CHECK-CVT-LABEL: test_tailcall_flipped: |
| ; CHECK-CVT: @ %bb.0: |
| ; CHECK-CVT-NEXT: vmov.f32 s2, s0 |
| ; CHECK-CVT-NEXT: vmov.f32 s0, s1 |
| ; CHECK-CVT-NEXT: vmov.f32 s1, s2 |
| ; CHECK-CVT-NEXT: b test_callee |
| ; |
| ; CHECK-BF16-LABEL: test_tailcall_flipped: |
| ; CHECK-BF16: @ %bb.0: |
| ; CHECK-BF16-NEXT: vmov r0, s1 |
| ; CHECK-BF16-NEXT: vmov r1, s0 |
| ; CHECK-BF16-NEXT: uxth r0, r0 |
| ; CHECK-BF16-NEXT: vmov s0, r0 |
| ; CHECK-BF16-NEXT: uxth r0, r1 |
| ; CHECK-BF16-NEXT: vmov s1, r0 |
| ; CHECK-BF16-NEXT: b test_callee |
| %r = tail call bfloat @test_callee(bfloat %b, bfloat %a) |
| ret bfloat %r |
| } |
| |
| ;define bfloat @test_select(bfloat %a, bfloat %b, i1 zeroext %c) #0 { |
| ; %r = select i1 %c, bfloat %a, bfloat %b |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_select_cc(bfloat %a, bfloat %b, bfloat %c, bfloat %d) #0 { |
| ; %cc = fcmp une bfloat %c, %d |
| ; %r = select i1 %cc, bfloat %a, bfloat %b |
| ; ret bfloat %r |
| ;} |
| |
| ;define float @test_select_cc_f32_f16(float %a, float %b, bfloat %c, bfloat %d) #0 { |
| ; %cc = fcmp une bfloat %c, %d |
| ; %r = select i1 %cc, float %a, float %b |
| ; ret float %r |
| ;} |
| |
| ;define bfloat @test_select_cc_f16_f32(bfloat %a, bfloat %b, float %c, float %d) #0 { |
| ; %cc = fcmp une float %c, %d |
| ; %r = select i1 %cc, bfloat %a, bfloat %b |
| ; ret bfloat %r |
| ;} |
| |
| ;define i1 @test_fcmp_une(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp une bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_ueq(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp ueq bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_ugt(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp ugt bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_uge(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp uge bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_ult(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp ult bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_ule(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp ule bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_uno(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp uno bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_one(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp one bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_oeq(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp oeq bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_ogt(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp ogt bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_oge(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp oge bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_olt(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp olt bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_ole(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp ole bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define i1 @test_fcmp_ord(bfloat %a, bfloat %b) #0 { |
| ; %r = fcmp ord bfloat %a, %b |
| ; ret i1 %r |
| ;} |
| |
| ;define void @test_fccmp(bfloat %in, ptr %out) { |
| ; %cmp1 = fcmp ogt bfloat %in, 0xR4800 |
| ; %cmp2 = fcmp olt bfloat %in, 0xR4500 |
| ; %cond = and i1 %cmp1, %cmp2 |
| ; %result = select i1 %cond, bfloat %in, bfloat 0xR4500 |
| ; store bfloat %result, ptr %out |
| ; ret void |
| ;} |
| |
| ;define void @test_br_cc(bfloat %a, bfloat %b, ptr %p1, ptr %p2) #0 { |
| ; %c = fcmp uge bfloat %a, %b |
| ; br i1 %c, label %then, label %else |
| ;then: |
| ; store i32 0, ptr %p1 |
| ; ret void |
| ;else: |
| ; store i32 0, ptr %p2 |
| ; ret void |
| ;} |
| |
| define bfloat @test_phi(ptr %p1) #0 { |
| ; CHECK-LABEL: test_phi: |
| ; CHECK: @ %bb.0: @ %entry |
| ; CHECK-NEXT: .save {r4, lr} |
| ; CHECK-NEXT: push {r4, lr} |
| ; CHECK-NEXT: .vsave {d8, d9} |
| ; CHECK-NEXT: vpush {d8, d9} |
| ; CHECK-NEXT: mov r4, r0 |
| ; CHECK-NEXT: ldrh r0, [r0] |
| ; CHECK-NEXT: vmov s18, r0 |
| ; CHECK-NEXT: .LBB5_1: @ %loop |
| ; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1 |
| ; CHECK-NEXT: ldrh r0, [r4] |
| ; CHECK-NEXT: vmov.f32 s16, s18 |
| ; CHECK-NEXT: vmov s18, r0 |
| ; CHECK-NEXT: mov r0, r4 |
| ; CHECK-NEXT: bl test_dummy |
| ; CHECK-NEXT: tst r0, #1 |
| ; CHECK-NEXT: bne .LBB5_1 |
| ; CHECK-NEXT: @ %bb.2: @ %return |
| ; CHECK-NEXT: vmov.f32 s0, s16 |
| ; CHECK-NEXT: vpop {d8, d9} |
| ; CHECK-NEXT: pop {r4, pc} |
| entry: |
| %a = load bfloat, ptr %p1 |
| br label %loop |
| loop: |
| %r = phi bfloat [%a, %entry], [%b, %loop] |
| %b = load bfloat, ptr %p1 |
| %c = call i1 @test_dummy(ptr %p1) |
| br i1 %c, label %loop, label %return |
| return: |
| ret bfloat %r |
| } |
| |
| declare i1 @test_dummy(ptr %p1) #0 |
| |
| ;define i32 @test_fptosi_i32(bfloat %a) #0 { |
| ; %r = fptosi bfloat %a to i32 |
| ; ret i32 %r |
| ;} |
| |
| ;define i64 @test_fptosi_i64(bfloat %a) #0 { |
| ; %r = fptosi bfloat %a to i64 |
| ; ret i64 %r |
| ;} |
| |
| ;define i32 @test_fptoui_i32(bfloat %a) #0 { |
| ; %r = fptoui bfloat %a to i32 |
| ; ret i32 %r |
| ;} |
| |
| ;define i64 @test_fptoui_i64(bfloat %a) #0 { |
| ; %r = fptoui bfloat %a to i64 |
| ; ret i64 %r |
| ;} |
| |
| ;define bfloat @test_uitofp_i32(i32 %a) #0 { |
| ; %r = uitofp i32 %a to bfloat |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_uitofp_i64(i64 %a) #0 { |
| ; %r = uitofp i64 %a to bfloat |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_sitofp_i32(i32 %a) #0 { |
| ; %r = sitofp i32 %a to bfloat |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_sitofp_i64(i64 %a) #0 { |
| ; %r = sitofp i64 %a to bfloat |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_uitofp_i32_fadd(i32 %a, bfloat %b) #0 { |
| ; %c = uitofp i32 %a to bfloat |
| ; %r = fadd bfloat %b, %c |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_sitofp_i32_fadd(i32 %a, bfloat %b) #0 { |
| ; %c = sitofp i32 %a to bfloat |
| ; %r = fadd bfloat %b, %c |
| ; ret bfloat %r |
| ;} |
| |
| define bfloat @test_fptrunc_float(float %a) #0 { |
| ; CHECK-CVT-LABEL: test_fptrunc_float: |
| ; CHECK-CVT: @ %bb.0: |
| ; CHECK-CVT-NEXT: .save {r11, lr} |
| ; CHECK-CVT-NEXT: push {r11, lr} |
| ; CHECK-CVT-NEXT: bl __truncsfbf2 |
| ; CHECK-CVT-NEXT: pop {r11, pc} |
| ; |
| ; CHECK-BF16-LABEL: test_fptrunc_float: |
| ; CHECK-BF16: @ %bb.0: |
| ; CHECK-BF16-NEXT: vcvtb.bf16.f32 s0, s0 |
| ; CHECK-BF16-NEXT: bx lr |
| %r = fptrunc float %a to bfloat |
| ret bfloat %r |
| } |
| |
| ;define bfloat @test_fptrunc_double(double %a) #0 { |
| ; %r = fptrunc double %a to bfloat |
| ; ret bfloat %r |
| ;} |
| |
| define bfloat @test_fptrunc_float_fast(float %a) #0 { |
| ; CHECK-CVT-LABEL: test_fptrunc_float_fast: |
| ; CHECK-CVT: @ %bb.0: |
| ; CHECK-CVT-NEXT: .save {r11, lr} |
| ; CHECK-CVT-NEXT: push {r11, lr} |
| ; CHECK-CVT-NEXT: bl __truncsfbf2 |
| ; CHECK-CVT-NEXT: pop {r11, pc} |
| ; |
| ; CHECK-BF16-LABEL: test_fptrunc_float_fast: |
| ; CHECK-BF16: @ %bb.0: |
| ; CHECK-BF16-NEXT: vcvtb.bf16.f32 s0, s0 |
| ; CHECK-BF16-NEXT: bx lr |
| %r = fptrunc fast float %a to bfloat |
| ret bfloat %r |
| } |
| |
| ;define bfloat @test_fptrunc_double_fast(double %a) #0 { |
| ; %r = fptrunc fast double %a to bfloat |
| ; ret bfloat %r |
| ;} |
| |
| ;define float @test_fpext_float(bfloat %a) #0 { |
| ; %r = fpext bfloat %a to float |
| ; ret float %r |
| ;} |
| |
| ;define double @test_fpext_double(bfloat %a) #0 { |
| ; %r = fpext bfloat %a to double |
| ; ret double %r |
| ;} |
| |
| ;define float @test_fpext_float_fast(bfloat %a) #0 { |
| ; %r = fpext fast bfloat %a to float |
| ; ret float %r |
| ;} |
| |
| ;define double @test_fpext_double_fast(bfloat %a) #0 { |
| ; %r = fpext fast bfloat %a to double |
| ; ret double %r |
| ;} |
| |
| define i16 @test_bitcast_bfloattoi16(bfloat %a) #0 { |
| ; CHECK-LABEL: test_bitcast_bfloattoi16: |
| ; CHECK: @ %bb.0: |
| ; CHECK-NEXT: vmov r0, s0 |
| ; CHECK-NEXT: bx lr |
| %r = bitcast bfloat %a to i16 |
| ret i16 %r |
| } |
| |
| define bfloat @test_bitcast_i16tobfloat(i16 %a) #0 { |
| ; CHECK-CVT-LABEL: test_bitcast_i16tobfloat: |
| ; CHECK-CVT: @ %bb.0: |
| ; CHECK-CVT-NEXT: vmov s0, r0 |
| ; CHECK-CVT-NEXT: bx lr |
| ; |
| ; CHECK-BF16-LABEL: test_bitcast_i16tobfloat: |
| ; CHECK-BF16: @ %bb.0: |
| ; CHECK-BF16-NEXT: .pad #4 |
| ; CHECK-BF16-NEXT: sub sp, sp, #4 |
| ; CHECK-BF16-NEXT: strh r0, [sp, #2] |
| ; CHECK-BF16-NEXT: ldrh r0, [sp, #2] |
| ; CHECK-BF16-NEXT: vmov s0, r0 |
| ; CHECK-BF16-NEXT: add sp, sp, #4 |
| ; CHECK-BF16-NEXT: bx lr |
| %r = bitcast i16 %a to bfloat |
| ret bfloat %r |
| } |
| |
| define half @test_bitcast_bfloattof16(bfloat %a) #0 { |
| ; CHECK-LABEL: test_bitcast_bfloattof16: |
| ; CHECK: @ %bb.0: |
| ; CHECK-NEXT: bx lr |
| %r = bitcast bfloat %a to half |
| ret half %r |
| } |
| |
| define bfloat @test_bitcast_f16tobfloat(half %a) #0 { |
| ; CHECK-CVT-LABEL: test_bitcast_f16tobfloat: |
| ; CHECK-CVT: @ %bb.0: |
| ; CHECK-CVT-NEXT: bx lr |
| ; |
| ; CHECK-BF16-LABEL: test_bitcast_f16tobfloat: |
| ; CHECK-BF16: @ %bb.0: |
| ; CHECK-BF16-NEXT: .pad #4 |
| ; CHECK-BF16-NEXT: sub sp, sp, #4 |
| ; CHECK-BF16-NEXT: vmov r0, s0 |
| ; CHECK-BF16-NEXT: strh r0, [sp, #2] |
| ; CHECK-BF16-NEXT: ldrh r0, [sp, #2] |
| ; CHECK-BF16-NEXT: vmov s0, r0 |
| ; CHECK-BF16-NEXT: add sp, sp, #4 |
| ; CHECK-BF16-NEXT: bx lr |
| %r = bitcast half %a to bfloat |
| ret bfloat %r |
| } |
| |
| ;define bfloat @test_sqrt(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.sqrt.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_powi(bfloat %a, i32 %b) #0 { |
| ; %r = call bfloat @llvm.powi.bf16.i32(bfloat %a, i32 %b) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_sin(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.sin.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_cos(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.cos.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_tan(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.tan.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_acos(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.acos.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_asin(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.asin.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_atan(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.atan.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_atan2(bfloat %a, bfloat %b) #0 { |
| ; %r = call bfloat @llvm.atan2.bf16(bfloat %a, bfloat %b) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_cosh(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.cosh.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_sinh(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.sinh.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_tanh(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.tanh.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_pow(bfloat %a, bfloat %b) #0 { |
| ; %r = call bfloat @llvm.pow.bf16(bfloat %a, bfloat %b) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_exp(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.exp.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_exp2(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.exp2.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_log(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.log.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_log10(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.log10.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_log2(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.log2.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_fma(bfloat %a, bfloat %b, bfloat %c) #0 { |
| ; %r = call bfloat @llvm.fma.bf16(bfloat %a, bfloat %b, bfloat %c) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_fabs(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.fabs.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_fneg(bfloat %a) #0 { |
| ; %r = fneg bfloat %a |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_minnum(bfloat %a, bfloat %b) #0 { |
| ; %r = call bfloat @llvm.minnum.bf16(bfloat %a, bfloat %b) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_maxnum(bfloat %a, bfloat %b) #0 { |
| ; %r = call bfloat @llvm.maxnum.bf16(bfloat %a, bfloat %b) |
| ; ret bfloat %r |
| ;} |
| |
| define bfloat @test_copysign(bfloat %a, bfloat %b) #0 { |
| ; CHECK-CVT-LABEL: test_copysign: |
| ; CHECK-CVT: @ %bb.0: |
| ; CHECK-CVT-NEXT: vmov r0, s1 |
| ; CHECK-CVT-NEXT: vmov r1, s0 |
| ; CHECK-CVT-NEXT: lsr r0, r0, #15 |
| ; CHECK-CVT-NEXT: bfi r1, r0, #15, #17 |
| ; CHECK-CVT-NEXT: vmov s0, r1 |
| ; CHECK-CVT-NEXT: bx lr |
| ; |
| ; CHECK-BF16-LABEL: test_copysign: |
| ; CHECK-BF16: @ %bb.0: |
| ; CHECK-BF16-NEXT: .pad #12 |
| ; CHECK-BF16-NEXT: sub sp, sp, #12 |
| ; CHECK-BF16-NEXT: vmov r0, s0 |
| ; CHECK-BF16-NEXT: vmov r1, s1 |
| ; CHECK-BF16-NEXT: strh r0, [sp, #10] |
| ; CHECK-BF16-NEXT: ldrh r0, [sp, #10] |
| ; CHECK-BF16-NEXT: strh r1, [sp, #8] |
| ; CHECK-BF16-NEXT: ldrh r1, [sp, #8] |
| ; CHECK-BF16-NEXT: strh r1, [sp, #4] |
| ; CHECK-BF16-NEXT: ldrb r1, [sp, #5] |
| ; CHECK-BF16-NEXT: strh r0, [sp] |
| ; CHECK-BF16-NEXT: ldrb r0, [sp, #1] |
| ; CHECK-BF16-NEXT: and r1, r1, #128 |
| ; CHECK-BF16-NEXT: and r0, r0, #127 |
| ; CHECK-BF16-NEXT: orr r0, r0, r1 |
| ; CHECK-BF16-NEXT: strb r0, [sp, #1] |
| ; CHECK-BF16-NEXT: ldrh r0, [sp] |
| ; CHECK-BF16-NEXT: vmov s0, r0 |
| ; CHECK-BF16-NEXT: add sp, sp, #12 |
| ; CHECK-BF16-NEXT: bx lr |
| %r = call bfloat @llvm.copysign.bf16(bfloat %a, bfloat %b) |
| ret bfloat %r |
| } |
| |
| define bfloat @test_copysign_f32(bfloat %a, float %b) #0 { |
| ; CHECK-CVT-LABEL: test_copysign_f32: |
| ; CHECK-CVT: @ %bb.0: |
| ; CHECK-CVT-NEXT: vmov r0, s0 |
| ; CHECK-CVT-NEXT: vmov r1, s1 |
| ; CHECK-CVT-NEXT: and r1, r1, #-2147483648 |
| ; CHECK-CVT-NEXT: bfc r0, #15, #17 |
| ; CHECK-CVT-NEXT: orr r0, r0, r1, lsr #16 |
| ; CHECK-CVT-NEXT: vmov s0, r0 |
| ; CHECK-CVT-NEXT: bx lr |
| ; |
| ; CHECK-BF16-LABEL: test_copysign_f32: |
| ; CHECK-BF16: @ %bb.0: |
| ; CHECK-BF16-NEXT: .pad #8 |
| ; CHECK-BF16-NEXT: sub sp, sp, #8 |
| ; CHECK-BF16-NEXT: vmov r0, s0 |
| ; CHECK-BF16-NEXT: vmov r1, s1 |
| ; CHECK-BF16-NEXT: strh r0, [sp, #6] |
| ; CHECK-BF16-NEXT: ldrh r0, [sp, #6] |
| ; CHECK-BF16-NEXT: strh r0, [sp] |
| ; CHECK-BF16-NEXT: and r1, r1, #-2147483648 |
| ; CHECK-BF16-NEXT: ldrb r0, [sp, #1] |
| ; CHECK-BF16-NEXT: and r0, r0, #127 |
| ; CHECK-BF16-NEXT: orr r0, r0, r1, lsr #24 |
| ; CHECK-BF16-NEXT: strb r0, [sp, #1] |
| ; CHECK-BF16-NEXT: ldrh r0, [sp] |
| ; CHECK-BF16-NEXT: vmov s0, r0 |
| ; CHECK-BF16-NEXT: add sp, sp, #8 |
| ; CHECK-BF16-NEXT: bx lr |
| %tb = fptrunc float %b to bfloat |
| %r = call bfloat @llvm.copysign.bf16(bfloat %a, bfloat %tb) |
| ret bfloat %r |
| } |
| |
| define bfloat @test_copysign_f64(bfloat %a, double %b) #0 { |
| ; CHECK-CVT-LABEL: test_copysign_f64: |
| ; CHECK-CVT: @ %bb.0: |
| ; CHECK-CVT-NEXT: vmov r0, s0 |
| ; CHECK-CVT-NEXT: vmov r1, r2, d1 |
| ; CHECK-CVT-NEXT: and r1, r2, #-2147483648 |
| ; CHECK-CVT-NEXT: bfc r0, #15, #17 |
| ; CHECK-CVT-NEXT: orr r0, r0, r1, lsr #16 |
| ; CHECK-CVT-NEXT: vmov s0, r0 |
| ; CHECK-CVT-NEXT: bx lr |
| ; |
| ; CHECK-BF16-LABEL: test_copysign_f64: |
| ; CHECK-BF16: @ %bb.0: |
| ; CHECK-BF16-NEXT: .pad #24 |
| ; CHECK-BF16-NEXT: sub sp, sp, #24 |
| ; CHECK-BF16-NEXT: vmov r0, s0 |
| ; CHECK-BF16-NEXT: strh r0, [sp, #22] |
| ; CHECK-BF16-NEXT: ldrh r0, [sp, #22] |
| ; CHECK-BF16-NEXT: vstr d1, [sp, #8] |
| ; CHECK-BF16-NEXT: ldrb r1, [sp, #15] |
| ; CHECK-BF16-NEXT: strh r0, [sp, #4] |
| ; CHECK-BF16-NEXT: ldrb r0, [sp, #5] |
| ; CHECK-BF16-NEXT: and r1, r1, #128 |
| ; CHECK-BF16-NEXT: and r0, r0, #127 |
| ; CHECK-BF16-NEXT: orr r0, r0, r1 |
| ; CHECK-BF16-NEXT: strb r0, [sp, #5] |
| ; CHECK-BF16-NEXT: ldrh r0, [sp, #4] |
| ; CHECK-BF16-NEXT: vmov s0, r0 |
| ; CHECK-BF16-NEXT: add sp, sp, #24 |
| ; CHECK-BF16-NEXT: bx lr |
| %tb = fptrunc double %b to bfloat |
| %r = call bfloat @llvm.copysign.bf16(bfloat %a, bfloat %tb) |
| ret bfloat %r |
| } |
| |
| ; away the (fpext (fp_round <result>)) here. |
| |
| ;define float @test_copysign_extended(bfloat %a, bfloat %b) #0 { |
| ; %r = call bfloat @llvm.copysign.bf16(bfloat %a, bfloat %b) |
| ; %xr = fpext bfloat %r to float |
| ; ret float %xr |
| ;} |
| |
| ;define bfloat @test_floor(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.floor.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_ceil(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.ceil.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_trunc(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.trunc.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_rint(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.rint.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_nearbyint(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.nearbyint.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_round(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.round.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_roundeven(bfloat %a) #0 { |
| ; %r = call bfloat @llvm.roundeven.bf16(bfloat %a) |
| ; ret bfloat %r |
| ;} |
| |
| ;define bfloat @test_fmuladd(bfloat %a, bfloat %b, bfloat %c) #0 { |
| ; %r = call bfloat @llvm.fmuladd.bf16(bfloat %a, bfloat %b, bfloat %c) |
| ; ret bfloat %r |
| ;} |
| |
| attributes #0 = { nounwind } |
