Home
last modified time | relevance | path

Searched refs:approx (Results 1 – 25 of 36) sorted by relevance

12

/llvm-project-15.0.7/llvm/test/CodeGen/NVPTX/
H A Dsqrt-approx.ll17 ; CHECK: rsqrt.approx.f32
33 ; CHECK: rsqrt.approx.f64
42 ; CHECK: rsqrt.approx.f64
59 ; CHECK: sqrt.approx.f32
90 ; CHECK: rsqrt.approx.f64
106 ; CHECK: rsqrt.approx.f64
119 ; CHECK: rsqrt.approx.f32
134 ; CHECK: rsqrt.approx.f32
141 ; CHECK: rsqrt.approx.f64
156 ; CHECK: rsqrt.approx.f64
[all …]
H A Df16-ex2.ll4 declare half @llvm.nvvm.ex2.approx.f16(half)
5 declare <2 x half> @llvm.nvvm.ex2.approx.f16x2(<2 x half>)
10 ; CHECK: ex2.approx.f16
11 %res = call half @llvm.nvvm.ex2.approx.f16(half %0);
18 ; CHECK: ex2.approx.f16x2
19 %res = call <2 x half> @llvm.nvvm.ex2.approx.f16x2(<2 x half> %0);
H A Drefl1.ll16 declare float @llvm.nvvm.sin.approx.ftz.f(float) #1
19 declare float @llvm.nvvm.cos.approx.ftz.f(float) #1
22 declare float @llvm.nvvm.div.approx.ftz.f(float, float) #1
28 %0 = tail call float @llvm.nvvm.sin.approx.ftz.f(float %a)
29 %1 = tail call float @llvm.nvvm.cos.approx.ftz.f(float %a)
30 %2 = tail call float @llvm.nvvm.div.approx.ftz.f(float %0, float %1)
H A Dfast-math.ll18 ; CHECK: div.approx.f32
26 ; CHECK: sqrt.approx.f32
27 ; CHECK: div.approx.f32
45 ; CHECK: div.approx.ftz.f32
54 ; CHECK: div.approx.ftz.f32
74 ; CHECK: rsqrt.approx.f64
84 ; CHECK-NOT: rsqrt.approx
86 ; CHECK-NOT: rsqrt.approx
96 ; CHECK: rsqrt.approx.f32
198 ; CHECK: rcp.approx.f32
[all …]
H A Dinline-asm.ll6 ; CHECK: ex2.approx.ftz.f32 %f{{[0-9]+}}, %f{{[0-9]+}}
7 %0 = call float asm "ex2.approx.ftz.f32 $0, $1;", "=f,f"(float %x)
H A Ddiv-ri.ll5 ; CHECK: div.approx.f32
/llvm-project-15.0.7/mlir/test/Dialect/LLVMIR/
H A Doptimize-for-nvvm.mlir9 // CHECK-DAG: %[[rcp:.*]] = nvvm.rcp.approx.ftz.f %[[rhs]] : f32
10 // CHECK-DAG: %[[approx:.*]] = llvm.fmul %[[lhs]], %[[rcp]] : f32
12 …// CHECK-DAG: %[[err:.*]] = "llvm.intr.fma"(%[[approx]], %[[neg]], %[[lhs]]) : (f32, f32, f32)…
13 …// CHECK-DAG: %[[refined:.*]] = "llvm.intr.fma"(%[[err]], %[[rcp]], %[[approx]]) : (f32, f32, f32)…
14 // CHECK-DAG: %[[cast:.*]] = llvm.bitcast %[[approx]] : f32 to i32
19 // CHECK-DAG: %[[select:.*]] = llvm.select %[[pred]], %[[approx]], %[[refined]] : i1, f32
H A Dnvvm.mlir34 // CHECK: nvvm.rcp.approx.ftz.f %arg0 : f32
35 %0 = nvvm.rcp.approx.ftz.f %arg0 : f32
/llvm-project-15.0.7/mlir/lib/Dialect/LLVMIR/Transforms/
H A DOptimizeForNVVM.cpp59 Value approx = rewriter.create<LLVM::FMulOp>(loc, lhs, rcp); in matchAndRewrite() local
64 loc, approx, rewriter.create<LLVM::FNegOp>(loc, rhs), lhs); in matchAndRewrite()
65 Value refined = rewriter.create<LLVM::FMAOp>(loc, err, rcp, approx); in matchAndRewrite()
70 Value cast = rewriter.create<LLVM::BitcastOp>(loc, i32Type, approx); in matchAndRewrite()
79 rewriter.create<LLVM::SelectOp>(loc, f32Type, pred, approx, refined); in matchAndRewrite()
/llvm-project-15.0.7/libc/AOR_v20.02/math/tools/
H A Dcos.sollya16 approx = proc(poly,d) {
23 p = roundcoefficients(approx(poly,2*i), [|D ...|]);
H A Dv_exp.sollya15 approx = proc(poly,d) {
22 p = roundcoefficients(approx(poly,i), [|D ...|]);
H A Dv_sin.sollya22 approx = proc(poly,d) {
29 p = roundcoefficients(approx(poly,2*i), [|D ...|]);
H A Dlog.sollya21 approx = proc(poly,d) {
28 p = roundcoefficients(approx(poly,i), [|D ...|]);
H A Dsin.sollya22 approx = proc(poly,d) {
29 p = roundcoefficients(approx(poly,2*i), [|D ...|]);
H A Dv_log.sollya20 approx = proc(poly,d) {
27 p = roundcoefficients(approx(poly,i), [|D ...|]);
H A Dlog_abs.sollya16 approx = proc(poly,d) {
23 p = roundcoefficients(approx(poly,i), [|D ...|]);
H A Dexp.sollya16 approx = proc(poly,d) {
23 p = roundcoefficients(approx(poly,i), [|D ...|]);
H A Dlog2_abs.sollya20 approx = proc(poly,d) {
27 p = roundcoefficients(approx(poly,i), [|D ...|]);
H A Dlog2.sollya26 approx = proc(poly,d) {
33 p = roundcoefficients(approx(poly,i), [|D ...|]);
H A Dexp2.sollya24 approx = proc(poly,d) {
35 p = roundcoefficients(approx(poly,i), [|D ...|]);
/llvm-project-15.0.7/llvm/lib/Target/NVPTX/
H A DNVPTXInstrInfo.td949 "rcp.approx.ftz.f32 \t$dst, $b;",
955 "rcp.approx.f32 \t$dst, $b;",
964 "div.approx.ftz.f32 \t$dst, $a, $b;",
970 "div.approx.ftz.f32 \t$dst, $a, $b;",
976 "div.approx.f32 \t$dst, $a, $b;",
982 "div.approx.f32 \t$dst, $a, $b;",
988 // rcp.approx gives the same result as div.full(1.0f, a) and is faster.
993 "rcp.approx.ftz.f32 \t$dst, $b;",
999 "rcp.approx.f32 \t$dst, $b;",
1112 "sin.approx.f32 \t$dst, $src;",
[all …]
H A DNVPTXIntrinsics.td784 : F_MATH_2<"div.approx.ftz.f32 \t$dst, $src0, $src1;", Float32Regs,
906 def INT_NVVM_EX2_APPROX_F : F_MATH_1<"ex2.approx.f32 \t$dst, $src0;",
908 def INT_NVVM_EX2_APPROX_D : F_MATH_1<"ex2.approx.f64 \t$dst, $src0;",
910 def INT_NVVM_EX2_APPROX_F16 : F_MATH_1<"ex2.approx.f16 \t$dst, $src0;",
917 def INT_NVVM_LG2_APPROX_F : F_MATH_1<"lg2.approx.f32 \t$dst, $src0;",
919 def INT_NVVM_LG2_APPROX_D : F_MATH_1<"lg2.approx.f64 \t$dst, $src0;",
928 def INT_NVVM_SIN_APPROX_F : F_MATH_1<"sin.approx.f32 \t$dst, $src0;",
933 def INT_NVVM_COS_APPROX_F : F_MATH_1<"cos.approx.f32 \t$dst, $src0;",
1064 def INT_NVVM_SQRT_APPROX_F : F_MATH_1<"sqrt.approx.f32 \t$dst, $src0;",
1093 def INT_NVVM_RSQRT_APPROX_F : F_MATH_1<"rsqrt.approx.f32 \t$dst, $src0;",
[all …]
/llvm-project-15.0.7/mlir/test/Target/LLVMIR/
H A Dnvvmir.mlir38 // CHECK: call float @llvm.nvvm.rcp.approx.ftz.f
39 %1 = nvvm.rcp.approx.ftz.f %0 : f32
/llvm-project-15.0.7/llvm/test/CodeGen/PowerPC/
H A Dpow-025-075-intrinsic-scalar-mass-fast.ll312 …-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "approx-func-fp-math"="tru…
/llvm-project-15.0.7/llvm/docs/
H A DCompileCudaWithLLVM.rst129 * ``-fcuda-approx-transcendentals`` (default: off) When this is enabled, the
132 example, this flag allows clang to emit the ptx ``sin.approx.f32``

12