88 lines · plain
1; RUN: llc -mtriple=amdgcn < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s2; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s3; RUN: llc -mtriple=r600 -mcpu=redwood < %s | FileCheck -check-prefix=R600 -check-prefix=FUNC %s4 5; FUNC-LABEL: {{^}}fmul_f32:6; GCN: v_mul_f327 8; R600: MUL_IEEE {{\** *}}{{T[0-9]+\.[XYZW]}}, KC0[2].Z, KC0[2].W9define amdgpu_kernel void @fmul_f32(ptr addrspace(1) %out, float %a, float %b) {10entry:11 %0 = fmul float %a, %b12 store float %0, ptr addrspace(1) %out13 ret void14}15 16; FUNC-LABEL: {{^}}fmul_v2f32:17; GCN: v_mul_f3218; GCN: v_mul_f3219 20; R600: MUL_IEEE {{\** *}}T{{[0-9]+\.[XYZW]}}21; R600: MUL_IEEE {{\** *}}T{{[0-9]+\.[XYZW]}}22define amdgpu_kernel void @fmul_v2f32(ptr addrspace(1) %out, <2 x float> %a, <2 x float> %b) {23entry:24 %0 = fmul <2 x float> %a, %b25 store <2 x float> %0, ptr addrspace(1) %out26 ret void27}28 29; FUNC-LABEL: {{^}}fmul_v4f32:30; GCN: v_mul_f3231; GCN: v_mul_f3232; GCN: v_mul_f3233; GCN: v_mul_f3234 35; R600: MUL_IEEE {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}36; R600: MUL_IEEE {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}37; R600: MUL_IEEE {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}38; R600: MUL_IEEE {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}39define amdgpu_kernel void @fmul_v4f32(ptr addrspace(1) %out, ptr addrspace(1) %in) {40 %b_ptr = getelementptr <4 x float>, ptr addrspace(1) %in, i32 141 %a = load <4 x float>, ptr addrspace(1) %in42 %b = load <4 x float>, ptr addrspace(1) %b_ptr43 %result = fmul <4 x float> %a, %b44 store <4 x float> %result, ptr addrspace(1) %out45 ret void46}47 48; FUNC-LABEL: {{^}}test_mul_2_k:49; GCN: v_mul_f3250; GCN-NOT: v_mul_f3251; GCN: s_endpgm52define amdgpu_kernel void @test_mul_2_k(ptr addrspace(1) %out, float %x) #0 {53 %y = fmul float %x, 2.054 %z = fmul float %y, 3.055 store float %z, ptr addrspace(1) %out56 ret void57}58 59; FUNC-LABEL: {{^}}test_mul_2_k_inv:60; GCN: v_mul_f3261; GCN-NOT: v_mul_f3262; GCN-NOT: v_mad_f3263; GCN: s_endpgm64define amdgpu_kernel void @test_mul_2_k_inv(ptr addrspace(1) %out, float %x) #0 {65 %y = fmul float %x, 3.066 %z = fmul float %y, 2.067 store float %z, ptr addrspace(1) %out68 ret void69}70 71; There should be three multiplies here; %a should be used twice (once72; negated), not duplicated into mul x, 5.0 and mul x, -5.0.73; FUNC-LABEL: {{^}}test_mul_twouse:74; GCN: v_mul_f3275; GCN: v_mul_f3276; GCN: v_mul_f3277; GCN-NOT: v_mul_f3278define amdgpu_kernel void @test_mul_twouse(ptr addrspace(1) %out, float %x, float %y) #0 {79 %a = fmul float %x, 5.080 %b = fsub float -0.0, %a81 %c = fmul float %b, %y82 %d = fmul float %c, %a83 store float %d, ptr addrspace(1) %out84 ret void85}86 87attributes #0 = { nounwind }88