Home
last modified time | relevance | path

Searched refs:FMA (Results 1 – 25 of 203) sorted by relevance

123456789

/external/swiftshader/third_party/llvm-7.0/llvm/test/CodeGen/X86/
Dvector-constrained-fp-intrinsics.ll2 …e=x86_64-pc-linux < %s | FileCheck --check-prefix=COMMON --check-prefix=NO-FMA --check-prefix=FMAC…
3 …linux -mattr=+fma < %s | FileCheck -check-prefix=COMMON --check-prefix=HAS-FMA --check-prefix=FMA6…
6 ; NO-FMA-LABEL: constrained_vector_fdiv_v2f64:
7 ; NO-FMA: # %bb.0: # %entry
8 ; NO-FMA-NEXT: movapd {{.*#+}} xmm0 = [1.000000e+00,2.000000e+00]
9 ; NO-FMA-NEXT: divpd {{.*}}(%rip), %xmm0
10 ; NO-FMA-NEXT: retq
12 ; HAS-FMA-LABEL: constrained_vector_fdiv_v2f64:
13 ; HAS-FMA: # %bb.0: # %entry
14 ; HAS-FMA-NEXT: vmovapd {{.*#+}} xmm0 = [1.000000e+00,2.000000e+00]
[all …]
Dfma-commute-x86.ll2 ; RUN: llc < %s -mtriple=x86_64-pc-win32 -mcpu=core-avx2 | FileCheck %s --check-prefix=FMA
3 ; RUN: llc < %s -mtriple=x86_64-pc-win32 -mattr=+fma | FileCheck %s --check-prefix=FMA
4 ; RUN: llc < %s -mcpu=bdver2 -mtriple=x86_64-pc-win32 -mattr=-fma4 | FileCheck %s --check-prefix=FMA
10 ; FMA-LABEL: test_x86_fmadd_baa_ss:
11 ; FMA: # %bb.0:
12 ; FMA-NEXT: vmovaps (%rdx), %xmm0
13 ; FMA-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
14 ; FMA-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm1
15 ; FMA-NEXT: retq
21 ; FMA-LABEL: test_x86_fmadd_aba_ss:
[all …]
Dfma-intrinsics-x86-upgrade.ll2 …wn -mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
4 …mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-WIN
8 ; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ss:
9 ; CHECK-FMA: # %bb.0:
10 ; CHECK-FMA-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa9,0xc2]
11 ; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
12 ; CHECK-FMA-NEXT: retq # encoding: [0xc3]
20 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ss:
21 ; CHECK-FMA-WIN: # %bb.0:
22 ; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
[all …]
Dfma-intrinsics-x86.ll2 …wn -mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
4 …mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-WIN
8 ; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ss:
9 ; CHECK-FMA: # %bb.0:
10 ; CHECK-FMA-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa9,0xc2]
11 ; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
12 ; CHECK-FMA-NEXT: retq # encoding: [0xc3]
20 ; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ss:
21 ; CHECK-FMA-WIN: # %bb.0:
22 ; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
[all …]
Dfma-intrinsics-canonical.ll2 …wn -mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
4 …mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-WIN
9 ; CHECK-FMA-LABEL: test_mm_fmadd_ps:
10 ; CHECK-FMA: # %bb.0: # %entry
11 ; CHECK-FMA-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
12 ; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
13 ; CHECK-FMA-NEXT: retq # encoding: [0xc3]
21 ; CHECK-FMA-WIN-LABEL: test_mm_fmadd_ps:
22 ; CHECK-FMA-WIN: # %bb.0: # %entry
23 ; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
[all …]
Dfma_patterns.ll2 …+fma -fp-contract=fast | FileCheck %s --check-prefix=ALL --check-prefix=FMA --check-prefix=FMA-INFS
6 …able-no-infs-fp-math | FileCheck %s --check-prefix=ALL --check-prefix=FMA --check-prefix=FMA-NOINFS
16 ; FMA-LABEL: test_f32_fmadd:
17 ; FMA: # %bb.0:
18 ; FMA-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0
19 ; FMA-NEXT: retq
36 ; FMA-LABEL: test_4f32_fmadd:
37 ; FMA: # %bb.0:
38 ; FMA-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0
39 ; FMA-NEXT: retq
[all …]
Dfma_patterns_wide.ll2 …nknown -mattr=+avx,+fma -fp-contract=fast | FileCheck %s --check-prefix=FMA --check-prefix=FMA-INFS
6 …p-contract=fast -enable-no-infs-fp-math | FileCheck %s --check-prefix=FMA --check-prefix=FMA-NOINFS
16 ; FMA-LABEL: test_16f32_fmadd:
17 ; FMA: # %bb.0:
18 ; FMA-NEXT: vfmadd213ps %ymm4, %ymm2, %ymm0
19 ; FMA-NEXT: vfmadd213ps %ymm5, %ymm3, %ymm1
20 ; FMA-NEXT: retq
38 ; FMA-LABEL: test_8f64_fmadd:
39 ; FMA: # %bb.0:
40 ; FMA-NEXT: vfmadd213pd %ymm4, %ymm2, %ymm0
[all …]
Drecip-fastmath2.ll4 …ma -print-schedule | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=FMA-RECIP
8 …le -mattr=-fma | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=HASWELL-NO-FMA
27 ; FMA-RECIP-LABEL: f32_no_step_2:
28 ; FMA-RECIP: # %bb.0:
29 ; FMA-RECIP-NEXT: vrcpss %xmm0, %xmm0, %xmm0
30 ; FMA-RECIP-NEXT: vmulss {{.*}}(%rip), %xmm0, %xmm0
31 ; FMA-RECIP-NEXT: retq
51 ; HASWELL-NO-FMA-LABEL: f32_no_step_2:
52 ; HASWELL-NO-FMA: # %bb.0:
53 ; HASWELL-NO-FMA-NEXT: vrcpss %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
[all …]
Drecip-fastmath.ll4 …wn -mattr=+avx,+fma | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=FMA-RECIP
8 …ll -mattr=-fma | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=HASWELL-NO-FMA
34 ; FMA-RECIP-LABEL: f32_no_estimate:
35 ; FMA-RECIP: # %bb.0:
36 ; FMA-RECIP-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
37 ; FMA-RECIP-NEXT: vdivss %xmm0, %xmm1, %xmm0
38 ; FMA-RECIP-NEXT: retq
58 ; HASWELL-NO-FMA-LABEL: f32_no_estimate:
59 ; HASWELL-NO-FMA: # %bb.0:
60 ; HASWELL-NO-FMA-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
[all …]
Dcomplex-fastmath.ll4 …n-unknown -mattr=+avx2,+fma | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=FMA
5 …n -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=FMA
36 ; FMA-LABEL: complex_square_f32:
37 ; FMA: # %bb.0:
38 ; FMA-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
39 ; FMA-NEXT: vaddss %xmm0, %xmm0, %xmm2
40 ; FMA-NEXT: vmulss %xmm2, %xmm1, %xmm2
41 ; FMA-NEXT: vmulss %xmm1, %xmm1, %xmm1
42 ; FMA-NEXT: vfmsub231ss %xmm0, %xmm0, %xmm1
43 ; FMA-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm2[0],xmm1[2,3]
[all …]
/external/llvm/test/CodeGen/X86/
Dfma_patterns.ll3 …own-unknown -mattr=+avx,+fma -fp-contract=fast | FileCheck %s --check-prefix=ALL --check-prefix=FMA
13 ; FMA-LABEL: test_f32_fmadd:
14 ; FMA: # BB#0:
15 ; FMA-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0
16 ; FMA-NEXT: retq
34 ; FMA-LABEL: test_4f32_fmadd:
35 ; FMA: # BB#0:
36 ; FMA-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0
37 ; FMA-NEXT: retq
54 ; FMA-LABEL: test_8f32_fmadd:
[all …]
Dfma_patterns_wide.ll2 …mtriple=x86_64-unknown-unknown -mattr=+avx,+fma -fp-contract=fast | FileCheck %s --check-prefix=FMA
12 ; FMA-LABEL: test_16f32_fmadd:
13 ; FMA: # BB#0:
14 ; FMA-NEXT: vfmadd213ps %ymm4, %ymm2, %ymm0
15 ; FMA-NEXT: vfmadd213ps %ymm5, %ymm3, %ymm1
16 ; FMA-NEXT: retq
34 ; FMA-LABEL: test_8f64_fmadd:
35 ; FMA: # BB#0:
36 ; FMA-NEXT: vfmadd213pd %ymm4, %ymm2, %ymm0
37 ; FMA-NEXT: vfmadd213pd %ymm5, %ymm3, %ymm1
[all …]
Dfma-intrinsics-x86.ll1 …ch=x86-64 -mcpu=corei7-avx -mattr=+fma | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
2 …6-64 -mcpu=core-avx2 -mattr=+fma,+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
3 … -mcpu=core-avx2 -mattr=+fma,+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-WIN
6 …nown-unknown -mcpu=bdver2 -mattr=-fma4 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
13 ; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
14 ; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
15 ; CHECK-FMA-WIN-NEXT: vfmadd132ss (%rdx), %xmm1, %xmm0
17 ; CHECK-FMA-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0
30 ; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
31 ; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
[all …]
Dfma.ll1 …pple-darwin10 -mattr=+fma,-fma4 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-INST
2 …pple-darwin10 -mattr=-fma,-fma4 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-CALL
3 …-apple-darwin10 -mattr=+fma,-fma4 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-INST
4 …apple-darwin10 -mattr=-fma,-fma4 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-CALL
5 …win10 -mattr=+avx512f,-fma,-fma4 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-INST
6 …ch=x86 -mcpu=bdver2 -mattr=-fma4 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-INST
7 …86 -mcpu=bdver2 -mattr=-fma,-fma4 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-CALL
10 ; CHECK-FMA-INST: vfmadd213ss
11 ; CHECK-FMA-CALL: fmaf
19 ; CHECK-FMA-INST: vfmadd213sd
[all …]
/external/llvm/test/Transforms/SLPVectorizer/X86/
Dfma.ll2 …: opt < %s -mtriple=x86_64-unknown -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=NO-FMA
3 …le=x86_64-unknown -mcpu=corei7-avx -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=NO-FMA
4 …own -mcpu=bdver1 -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=FMA --check-prefix=FMA2…
5 … -mcpu=core-avx2 -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=FMA --check-prefix=FMA2…
6 …u=skylake-avx512 -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=FMA --check-prefix=FMA5…
23 ; FMA
27 ; NO-FMA-LABEL: @fma_2f64(
28 ; NO-FMA-NEXT: [[A0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x doub…
29 ; NO-FMA-NEXT: [[A1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x doub…
30 ; NO-FMA-NEXT: [[B0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x doub…
[all …]
/external/swiftshader/third_party/llvm-7.0/llvm/test/Transforms/SLPVectorizer/X86/
Dfma.ll2 …: opt < %s -mtriple=x86_64-unknown -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=NO-FMA
3 …le=x86_64-unknown -mcpu=corei7-avx -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=NO-FMA
4 …own -mcpu=bdver1 -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=FMA --check-prefix=FMA2…
5 … -mcpu=core-avx2 -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=FMA --check-prefix=FMA2…
6 …u=skylake-avx512 -basicaa -slp-vectorizer -S | FileCheck %s --check-prefix=FMA --check-prefix=FMA5…
23 ; FMA
27 ; NO-FMA-LABEL: @fma_2f64(
28 ; NO-FMA-NEXT: [[A0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x doub…
29 ; NO-FMA-NEXT: [[A1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x doub…
30 ; NO-FMA-NEXT: [[B0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x doub…
[all …]
/external/swiftshader/third_party/llvm-7.0/llvm/test/Transforms/InstCombine/
Dfma.ll13 ; CHECK-NEXT: [[FMA:%.*]] = call float @llvm.fma.f32(float [[X:%.*]], float [[Y:%.*]], float [[Z…
14 ; CHECK-NEXT: ret float [[FMA]]
24 ; CHECK-NEXT: [[FMA:%.*]] = call <2 x float> @llvm.fma.v2f32(<2 x float> [[X:%.*]], <2 x float> …
25 ; CHECK-NEXT: ret <2 x float> [[FMA]]
35 ; CHECK-NEXT: [[FMA:%.*]] = call <2 x float> @llvm.fma.v2f32(<2 x float> [[X:%.*]], <2 x float> …
36 ; CHECK-NEXT: ret <2 x float> [[FMA]]
46 ; CHECK-NEXT: [[FMA:%.*]] = call fast float @llvm.fma.f32(float [[X:%.*]], float [[Y:%.*]], floa…
47 ; CHECK-NEXT: ret float [[FMA]]
57 ; CHECK-NEXT: [[FMA:%.*]] = call float @llvm.fma.f32(float [[Y:%.*]], float bitcast (i32 ptrtoin…
58 ; CHECK-NEXT: ret float [[FMA]]
[all …]
/external/llvm/lib/Target/X86/
DX86InstrFMA.td1 //===-- X86InstrFMA.td - FMA Instruction Set ---------------*- tablegen -*-===//
10 // This file describes FMA (Fused Multiply-Add) instructions.
18 // For all FMA opcodes declared in fma3p_rm and fma3s_rm milticlasses defined
24 // FMA*213*:
28 // FMA*132*:
32 // FMA*231*:
133 // All source register operands of FMA opcodes defined in fma3s_rm multiclass
135 // adjustment, for example, commuting the operands 1 and 2 in FMA*132 form
136 // would require an opcode change to FMA*231:
137 // FMA*132* reg1, reg2, reg3; // reg1 * reg3 + reg2;
[all …]
/external/llvm/test/CodeGen/NVPTX/
Dfma-disable.ll1 ; RUN: llc < %s -march=nvptx -mcpu=sm_20 -nvptx-fma-level=1 | FileCheck %s -check-prefix=FMA
3 ; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 -nvptx-fma-level=1 | FileCheck %s -check-prefix=FMA
8 ; FMA: fma.rn.f32
18 ; FMA: fma.rn.f64
/external/swiftshader/third_party/llvm-7.0/llvm/test/CodeGen/NVPTX/
Dfma-disable.ll1 ; RUN: llc < %s -march=nvptx -mcpu=sm_20 -nvptx-fma-level=1 | FileCheck %s -check-prefix=FMA
3 ; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 -nvptx-fma-level=1 | FileCheck %s -check-prefix=FMA
8 ; FMA: fma.rn.f32
18 ; FMA: fma.rn.f64
/external/llvm/test/CodeGen/AMDGPU/
Dfma.ll14 ; EG: FMA {{\*? *}}[[RES]]
30 ; EG-DAG: FMA {{\*? *}}[[RES]].[[CHLO]]
31 ; EG-DAG: FMA {{\*? *}}[[RES]].[[CHHI]]
49 ; EG-DAG: FMA {{\*? *}}[[RES]].X
50 ; EG-DAG: FMA {{\*? *}}[[RES]].Y
51 ; EG-DAG: FMA {{\*? *}}[[RES]].Z
52 ; EG-DAG: FMA {{\*? *}}[[RES]].W
/external/tensorflow/tensorflow/core/kernels/
Dsparse_matmul_op.cc283 #define FMA(a, b, c, d) d = Eigen::internal::pmadd<Packet>(a, b, c); macro
418 FMA(a, b_0, c1, c1); in MulAdd()
419 FMA(a, b_1, c2, c2); in MulAdd()
446 FMA(a1, b1_0, c1, c1); in MulAdd3Way()
447 FMA(a1, b1_1, c2, c2); in MulAdd3Way()
448 FMA(a2, b2_0, c1, c1); in MulAdd3Way()
449 FMA(a2, b2_1, c2, c2); in MulAdd3Way()
450 FMA(a3, b3_0, c1, c1); in MulAdd3Way()
451 FMA(a3, b3_1, c2, c2); in MulAdd3Way()
490 FMA(a1, b1_0, c1, c1); in TwoMulAdd3Way()
[all …]
/external/swiftshader/third_party/llvm-7.0/llvm/lib/Target/X86/
DX86InstrFMA.td1 //===-- X86InstrFMA.td - FMA Instruction Set ---------------*- tablegen -*-===//
10 // This file describes FMA (Fused Multiply-Add) instructions.
18 // For all FMA opcodes declared in fma3p_rm_* and fma3s_rm_* multiclasses
24 // FMA*213*:
28 // FMA*132*:
32 // FMA*231*:
168 // All source register operands of FMA opcodes defined in fma3s_rm multiclass
170 // adjustment, for example, commuting the operands 1 and 2 in FMA*132 form
171 // would require an opcode change to FMA*231:
172 // FMA*132* reg1, reg2, reg3; // reg1 * reg3 + reg2;
[all …]
/external/swiftshader/third_party/llvm-7.0/llvm/test/CodeGen/AMDGPU/
Dfma.ll23 ; EG: FMA {{\*? *}}[[RES]]
49 ; EG-DAG: FMA {{\*? *}}[[RES]].[[CHLO]]
50 ; EG-DAG: FMA {{\*? *}}[[RES]].[[CHHI]]
72 ; EG-DAG: FMA {{\*? *}}[[RES]].X
73 ; EG-DAG: FMA {{\*? *}}[[RES]].Y
74 ; EG-DAG: FMA {{\*? *}}[[RES]].Z
75 ; EG-DAG: FMA {{\*? *}}[[RES]].W
/external/skia/src/core/
DSkCpu.h23 FMA = 1 << 8, enumerator
28 HSW = AVX2 | BMI1 | BMI2 | F16C | FMA,

123456789