diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fcmp.w64.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fcmp.w64.ll --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fcmp.w64.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.fcmp.w64.ll @@ -1,11 +1,11 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -march=amdgcn -mcpu=gfx1100 -mattr="+wavefrontsize64" -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX11 %s -; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI %s -; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9 %s +; RUN: llc -march=amdgcn -mcpu=gfx1100 -mattr="+wavefrontsize64" -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,SDAG-GFX11 %s +; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1100 -mattr="+wavefrontsize64" -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GISEL-GFX11 %s +; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefixes=SDAG-VI %s +; RUN: llc -global-isel -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefixes=GISEL-VI %s +; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,SDAG-GFX9 %s -; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1100 -mattr="+wavefrontsize64" -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX11,GISEL-GFX11 %s -; RUN: llc -global-isel -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,GISEL-VI %s -; RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9,GISEL-GFX9 %s +; RUN: llc -global-isel -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,GISEL-GFX9 %s declare i64 @llvm.amdgcn.fcmp.f32(float, float, i32) #0 declare i64 @llvm.amdgcn.fcmp.f64(double, double, i32) #0 @@ -28,17 +28,18 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_oeq_with_fabs: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_mov_b32_e32 v0, s3 -; GFX9-NEXT: v_cmp_eq_f32_e64 s[2:3], s2, |v0| -; GFX9-NEXT: v_mov_b32_e32 v0, s2 -; GFX9-NEXT: v_mov_b32_e32 v1, s3 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_oeq_with_fabs: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s3 +; SDAG-VI-NEXT: v_cmp_eq_f32_e64 s[2:3], s2, |v0| +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_oeq_with_fabs: ; GISEL-VI: ; %bb.0: @@ -52,18 +53,18 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_oeq_with_fabs: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s3 -; SDAG-VI-NEXT: v_cmp_eq_f32_e64 s[2:3], s2, |v0| -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_oeq_with_fabs: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_mov_b32_e32 v0, s3 +; GFX9-NEXT: v_cmp_eq_f32_e64 s[2:3], s2, |v0| +; GFX9-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] +; GFX9-NEXT: s_endpgm %temp = call float @llvm.fabs.f32(float %a) %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float %temp, i32 1) store i64 %result, ptr addrspace(1) %out @@ -84,17 +85,18 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_oeq_both_operands_with_fabs: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_mov_b32_e32 v0, s3 -; GFX9-NEXT: v_cmp_eq_f32_e64 s[2:3], |s2|, |v0| -; GFX9-NEXT: v_mov_b32_e32 v0, s2 -; GFX9-NEXT: v_mov_b32_e32 v1, s3 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_oeq_both_operands_with_fabs: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s3 +; SDAG-VI-NEXT: v_cmp_eq_f32_e64 s[2:3], |s2|, |v0| +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_oeq_both_operands_with_fabs: ; GISEL-VI: ; %bb.0: @@ -108,18 +110,18 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_oeq_both_operands_with_fabs: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s3 -; SDAG-VI-NEXT: v_cmp_eq_f32_e64 s[2:3], |s2|, |v0| -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_oeq_both_operands_with_fabs: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_mov_b32_e32 v0, s3 +; GFX9-NEXT: v_cmp_eq_f32_e64 s[2:3], |s2|, |v0| +; GFX9-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] +; GFX9-NEXT: s_endpgm %temp = call float @llvm.fabs.f32(float %a) %src_input = call float @llvm.fabs.f32(float %src) %result = call i64 @llvm.amdgcn.fcmp.f32(float %src_input, float %temp, i32 1) @@ -136,14 +138,6 @@ ; SDAG-GFX11: ; %bb.0: ; SDAG-GFX11-NEXT: s_endpgm ; -; SDAG-VI-LABEL: v_fcmp_f32: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_endpgm -; -; SDAG-GFX9-LABEL: v_fcmp_f32: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_endpgm -; ; GISEL-GFX11-LABEL: v_fcmp_f32: ; GISEL-GFX11: ; %bb.0: ; GISEL-GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x24 @@ -153,6 +147,10 @@ ; GISEL-GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GISEL-GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f32: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f32: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 @@ -162,6 +160,10 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[0:1], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f32: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f32: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 @@ -169,13 +171,6 @@ ; GISEL-GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GISEL-GFX9-NEXT: global_store_dwordx2 v0, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; GISEL-GFX-LABEL: v_fcmp_f32: -; GISEL-GFX: ; %bb.0: -; GISEL-GFX-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; GISEL-GFX-NEXT: v_mov_b32_e32 v0, 0 -; GISEL-GFX-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX-NEXT: global_store_dwordx2 v0, v[0:1], s[0:1] -; GISEL-GFX-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 -1) store i64 %result, ptr addrspace(1) %out ret void @@ -197,18 +192,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_oeq: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_eq_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_oeq: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_eq_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_oeq: ; GISEL-VI: ; %bb.0: @@ -223,19 +219,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_oeq: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_eq_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_oeq: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_eq_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 1) store i64 %result, ptr addrspace(1) %out ret void @@ -257,18 +253,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_one: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_neq_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_one: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_neq_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_one: ; GISEL-VI: ; %bb.0: @@ -283,19 +280,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_one: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_neq_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_one: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_neq_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 6) store i64 %result, ptr addrspace(1) %out ret void @@ -317,18 +314,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_ogt: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_gt_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_ogt: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_gt_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_ogt: ; GISEL-VI: ; %bb.0: @@ -343,25 +341,25 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_ogt: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_gt_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm - %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 2) - store i64 %result, ptr addrspace(1) %out - ret void -} - -define amdgpu_kernel void @v_fcmp_f32_oge(ptr addrspace(1) %out, float %src) { +; +; GFX9-LABEL: v_fcmp_f32_ogt: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_gt_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm + %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 2) + store i64 %result, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @v_fcmp_f32_oge(ptr addrspace(1) %out, float %src) { ; GFX11-LABEL: v_fcmp_f32_oge: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_clause 0x1 @@ -377,18 +375,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_oge: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_ge_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_oge: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_ge_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_oge: ; GISEL-VI: ; %bb.0: @@ -403,19 +402,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_oge: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_ge_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_oge: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_ge_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 3) store i64 %result, ptr addrspace(1) %out ret void @@ -437,18 +436,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_olt: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_lt_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_olt: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_lt_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_olt: ; GISEL-VI: ; %bb.0: @@ -463,19 +463,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_olt: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_lt_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_olt: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_lt_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 4) store i64 %result, ptr addrspace(1) %out ret void @@ -497,18 +497,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_ole: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_le_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_ole: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_le_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_ole: ; GISEL-VI: ; %bb.0: @@ -523,19 +524,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_ole: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_le_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_ole: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_le_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 5) store i64 %result, ptr addrspace(1) %out ret void @@ -558,18 +559,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_ueq: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_nlg_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_ueq: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nlg_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_ueq: ; GISEL-VI: ; %bb.0: @@ -584,19 +586,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_ueq: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nlg_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_ueq: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_nlg_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 9) store i64 %result, ptr addrspace(1) %out ret void @@ -618,18 +620,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_une: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_neq_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_une: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_neq_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_une: ; GISEL-VI: ; %bb.0: @@ -644,19 +647,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_une: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_neq_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_une: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_neq_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 14) store i64 %result, ptr addrspace(1) %out ret void @@ -672,38 +675,12 @@ ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: v_cmp_nge_f32_e64 s[2:3], 0x42c80000, s2 ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NEXT: v_mov_b32_e32 v0, s2 -; GFX11-NEXT: v_mov_b32_e32 v1, s3 -; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] -; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) -; GFX11-NEXT: s_endpgm -; -; GFX9-LABEL: v_fcmp_f32_ugt: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_nle_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; GFX11-NEXT: v_mov_b32_e32 v0, s2 +; GFX11-NEXT: v_mov_b32_e32 v1, s3 +; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] +; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) +; GFX11-NEXT: s_endpgm ; -; GISEL-VI-LABEL: v_fcmp_f32_ugt: -; GISEL-VI: ; %bb.0: -; GISEL-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; GISEL-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; GISEL-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GISEL-VI-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-VI-NEXT: v_cmp_nle_f32_e64 s[2:3], s2, v0 -; GISEL-VI-NEXT: v_mov_b32_e32 v0, s2 -; GISEL-VI-NEXT: v_mov_b32_e32 v3, s1 -; GISEL-VI-NEXT: v_mov_b32_e32 v1, s3 -; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 -; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] -; GISEL-VI-NEXT: s_endpgm ; SDAG-VI-LABEL: v_fcmp_f32_ugt: ; SDAG-VI: ; %bb.0: ; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c @@ -717,6 +694,33 @@ ; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 ; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] ; SDAG-VI-NEXT: s_endpgm +; +; GISEL-VI-LABEL: v_fcmp_f32_ugt: +; GISEL-VI: ; %bb.0: +; GISEL-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; GISEL-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; GISEL-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GISEL-VI-NEXT: s_waitcnt lgkmcnt(0) +; GISEL-VI-NEXT: v_cmp_nle_f32_e64 s[2:3], s2, v0 +; GISEL-VI-NEXT: v_mov_b32_e32 v0, s2 +; GISEL-VI-NEXT: v_mov_b32_e32 v3, s1 +; GISEL-VI-NEXT: v_mov_b32_e32 v1, s3 +; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 +; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] +; GISEL-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_ugt: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_nle_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 10) store i64 %result, ptr addrspace(1) %out ret void @@ -738,18 +742,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_uge: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_nlt_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_uge: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nlt_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_uge: ; GISEL-VI: ; %bb.0: @@ -764,19 +769,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_uge: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nlt_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_uge: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_nlt_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 11) store i64 %result, ptr addrspace(1) %out ret void @@ -798,18 +803,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_ult: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_nge_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_ult: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nge_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_ult: ; GISEL-VI: ; %bb.0: @@ -824,19 +830,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_ult: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nge_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_ult: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_nge_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 12) store i64 %result, ptr addrspace(1) %out ret void @@ -858,18 +864,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f32_ule: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_ngt_f32_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f32_ule: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_ngt_f32_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f32_ule: ; GISEL-VI: ; %bb.0: @@ -884,19 +891,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f32_ule: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x42c80000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_ngt_f32_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f32_ule: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x42c80000 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_ngt_f32_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f32(float %src, float 100.00, i32 13) store i64 %result, ptr addrspace(1) %out ret void @@ -918,6 +925,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_oeq: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_eq_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_oeq: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -934,6 +955,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_oeq: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_eq_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_oeq: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -948,30 +981,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_oeq: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_eq_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_oeq: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_eq_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 1) store i64 %result, ptr addrspace(1) %out ret void @@ -993,6 +1002,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_one: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_neq_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_one: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1009,6 +1032,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_one: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_neq_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_one: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1023,30 +1058,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_one: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_neq_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_one: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_neq_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 6) store i64 %result, ptr addrspace(1) %out ret void @@ -1068,6 +1079,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_ogt: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_gt_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_ogt: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1084,6 +1109,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_ogt: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_gt_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_ogt: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1098,30 +1135,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_ogt: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_gt_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_ogt: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_gt_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 2) store i64 %result, ptr addrspace(1) %out ret void @@ -1143,6 +1156,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_oge: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_ge_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_oge: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1159,6 +1186,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_oge: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_ge_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_oge: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1173,30 +1212,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_oge: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_ge_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_oge: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_ge_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 3) store i64 %result, ptr addrspace(1) %out ret void @@ -1218,6 +1233,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_olt: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_lt_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_olt: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1234,6 +1263,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_olt: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_lt_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_olt: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1248,30 +1289,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_olt: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_lt_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_olt: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_lt_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 4) store i64 %result, ptr addrspace(1) %out ret void @@ -1293,6 +1310,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_ole: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_le_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_ole: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1309,6 +1340,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_ole: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_le_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_ole: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1323,30 +1366,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_ole: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_le_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_ole: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_le_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 5) store i64 %result, ptr addrspace(1) %out ret void @@ -1368,6 +1387,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_ueq: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nlg_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_ueq: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1384,6 +1417,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_ueq: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_nlg_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_ueq: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1398,30 +1443,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_ueq: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nlg_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_ueq: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_nlg_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 9) store i64 %result, ptr addrspace(1) %out ret void @@ -1443,6 +1464,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_une: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_neq_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_une: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1459,6 +1494,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_une: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_neq_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_une: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1473,30 +1520,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_une: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_neq_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_une: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_neq_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 14) store i64 %result, ptr addrspace(1) %out ret void @@ -1518,6 +1541,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_ugt: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nle_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_ugt: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1534,6 +1571,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_ugt: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_nle_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_ugt: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1548,30 +1597,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_ugt: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nle_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_ugt: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_nle_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 10) store i64 %result, ptr addrspace(1) %out ret void @@ -1593,6 +1618,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_uge: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nlt_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_uge: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1609,6 +1648,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_uge: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_nlt_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_uge: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1623,30 +1674,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_uge: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nlt_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_uge: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_nlt_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 11) store i64 %result, ptr addrspace(1) %out ret void @@ -1668,6 +1695,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_ult: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nge_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_ult: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1684,6 +1725,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_ult: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_nge_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_ult: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1698,30 +1751,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_ult: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nge_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_ult: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_nge_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 12) store i64 %result, ptr addrspace(1) %out ret void @@ -1743,6 +1772,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f64_ule: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_ngt_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f64_ule: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1759,6 +1802,18 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f64_ule: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 +; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-GFX9-NEXT: v_cmp_ngt_f64_e64 s[2:3], s[2:3], v[0:1] +; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 +; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 +; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f64_ule: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 @@ -1773,30 +1828,6 @@ ; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GISEL-GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f64_ule: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_ngt_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm -; SDAG-GFX9-LABEL: v_fcmp_f64_ule: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v0, 0 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, 0x40590000 -; SDAG-GFX9-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-GFX9-NEXT: v_cmp_ngt_f64_e64 s[2:3], s[2:3], v[0:1] -; SDAG-GFX9-NEXT: v_mov_b32_e32 v1, s2 -; SDAG-GFX9-NEXT: v_mov_b32_e32 v2, s3 -; SDAG-GFX9-NEXT: global_store_dwordx2 v0, v[1:2], s[0:1] -; SDAG-GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f64(double %src, double 100.00, i32 13) store i64 %result, ptr addrspace(1) %out ret void @@ -1821,19 +1852,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_oeq_with_fabs: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_lshr_b32 s0, s4, 16 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_cmp_eq_f16_e64 s[0:1], s4, |v0| -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_oeq_with_fabs: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: s_lshr_b32 s3, s2, 16 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s3 +; SDAG-VI-NEXT: v_cmp_eq_f16_e64 s[2:3], s2, |v0| +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_oeq_with_fabs: ; GISEL-VI: ; %bb.0: @@ -1849,20 +1881,20 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_oeq_with_fabs: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: s_lshr_b32 s3, s2, 16 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s3 -; SDAG-VI-NEXT: v_cmp_eq_f16_e64 s[2:3], s2, |v0| -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_oeq_with_fabs: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_lshr_b32 s0, s4, 16 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_cmp_eq_f16_e64 s[0:1], s4, |v0| +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %temp = call half @llvm.fabs.f16(half %a) %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half %temp, i32 1) store i64 %result, ptr addrspace(1) %out @@ -1888,19 +1920,20 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_oeq_both_operands_with_fabs: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_lshr_b32 s0, s4, 16 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_cmp_eq_f16_e64 s[0:1], |s4|, |v0| -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_oeq_both_operands_with_fabs: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: s_lshr_b32 s3, s2, 16 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s3 +; SDAG-VI-NEXT: v_cmp_eq_f16_e64 s[2:3], |s2|, |v0| +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_oeq_both_operands_with_fabs: ; GISEL-VI: ; %bb.0: @@ -1916,20 +1949,20 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_oeq_both_operands_with_fabs: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: s_lshr_b32 s3, s2, 16 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s3 -; SDAG-VI-NEXT: v_cmp_eq_f16_e64 s[2:3], |s2|, |v0| -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_oeq_both_operands_with_fabs: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_lshr_b32 s0, s4, 16 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_cmp_eq_f16_e64 s[0:1], |s4|, |v0| +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %temp = call half @llvm.fabs.f16(half %a) %src_input = call half @llvm.fabs.f16(half %src) %result = call i64 @llvm.amdgcn.fcmp.f16(half %src_input, half %temp, i32 1) @@ -1946,14 +1979,6 @@ ; SDAG-GFX11: ; %bb.0: ; SDAG-GFX11-NEXT: s_endpgm ; -; SDAG-VI-LABEL: v_fcmp_f16: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_endpgm -; -; SDAG-GFX9-LABEL: v_fcmp_f16: -; SDAG-GFX9: ; %bb.0: -; SDAG-GFX9-NEXT: s_endpgm -; ; GISEL-GFX11-LABEL: v_fcmp_f16: ; GISEL-GFX11: ; %bb.0: ; GISEL-GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x24 @@ -1963,6 +1988,10 @@ ; GISEL-GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GISEL-GFX11-NEXT: s_endpgm ; +; SDAG-VI-LABEL: v_fcmp_f16: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_endpgm +; ; GISEL-VI-LABEL: v_fcmp_f16: ; GISEL-VI: ; %bb.0: ; GISEL-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 @@ -1972,6 +2001,10 @@ ; GISEL-VI-NEXT: flat_store_dwordx2 v[0:1], v[0:1] ; GISEL-VI-NEXT: s_endpgm ; +; SDAG-GFX9-LABEL: v_fcmp_f16: +; SDAG-GFX9: ; %bb.0: +; SDAG-GFX9-NEXT: s_endpgm +; ; GISEL-GFX9-LABEL: v_fcmp_f16: ; GISEL-GFX9: ; %bb.0: ; GISEL-GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 @@ -1979,13 +2012,6 @@ ; GISEL-GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GISEL-GFX9-NEXT: global_store_dwordx2 v0, v[0:1], s[0:1] ; GISEL-GFX9-NEXT: s_endpgm -; GISEL-GFX-LABEL: v_fcmp_f16: -; GISEL-GFX: ; %bb.0: -; GISEL-GFX-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; GISEL-GFX-NEXT: v_mov_b32_e32 v0, 0 -; GISEL-GFX-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX-NEXT: global_store_dwordx2 v0, v[0:1], s[0:1] -; GISEL-GFX-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 -1) store i64 %result, ptr addrspace(1) %out ret void @@ -2008,18 +2034,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_oeq: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_eq_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_oeq: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_eq_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_oeq: ; GISEL-VI: ; %bb.0: @@ -2034,19 +2061,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_oeq: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_eq_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_oeq: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_eq_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 1) store i64 %result, ptr addrspace(1) %out ret void @@ -2069,18 +2096,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_one: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_neq_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_one: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_neq_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_one: ; GISEL-VI: ; %bb.0: @@ -2095,19 +2123,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_one: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_neq_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_one: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_neq_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 6) store i64 %result, ptr addrspace(1) %out ret void @@ -2130,18 +2158,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_ogt: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_gt_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_ogt: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_gt_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_ogt: ; GISEL-VI: ; %bb.0: @@ -2156,19 +2185,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_ogt: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_gt_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_ogt: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_gt_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 2) store i64 %result, ptr addrspace(1) %out ret void @@ -2191,18 +2220,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_oge: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_ge_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_oge: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_ge_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_oge: ; GISEL-VI: ; %bb.0: @@ -2217,19 +2247,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_oge: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_ge_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_oge: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_ge_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 3) store i64 %result, ptr addrspace(1) %out ret void @@ -2252,18 +2282,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_olt: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_lt_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_olt: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_lt_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_olt: ; GISEL-VI: ; %bb.0: @@ -2278,19 +2309,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_olt: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_lt_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_olt: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_lt_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 4) store i64 %result, ptr addrspace(1) %out ret void @@ -2313,18 +2344,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_ole: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_le_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_ole: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_le_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_ole: ; GISEL-VI: ; %bb.0: @@ -2339,19 +2371,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_ole: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_le_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_ole: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_le_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 5) store i64 %result, ptr addrspace(1) %out ret void @@ -2374,18 +2406,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_ueq: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_nlg_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_ueq: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nlg_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_ueq: ; GISEL-VI: ; %bb.0: @@ -2400,19 +2433,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_ueq: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nlg_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_ueq: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_nlg_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 9) store i64 %result, ptr addrspace(1) %out ret void @@ -2435,18 +2468,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_une: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_neq_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_une: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_neq_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_une: ; GISEL-VI: ; %bb.0: @@ -2461,19 +2495,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_une: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_neq_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_une: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_neq_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 14) store i64 %result, ptr addrspace(1) %out ret void @@ -2496,18 +2530,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_ugt: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_nle_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_ugt: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nle_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_ugt: ; GISEL-VI: ; %bb.0: @@ -2522,19 +2557,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_ugt: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nle_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_ugt: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_nle_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 10) store i64 %result, ptr addrspace(1) %out ret void @@ -2557,18 +2592,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_uge: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_nlt_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_uge: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nlt_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_uge: ; GISEL-VI: ; %bb.0: @@ -2583,19 +2619,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_uge: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nlt_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_uge: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_nlt_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 11) store i64 %result, ptr addrspace(1) %out ret void @@ -2618,18 +2654,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_ult: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_nge_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_ult: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_nge_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_ult: ; GISEL-VI: ; %bb.0: @@ -2644,19 +2681,19 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_ult: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_nge_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_ult: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_nge_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 12) store i64 %result, ptr addrspace(1) %out ret void @@ -2679,18 +2716,19 @@ ; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS) ; GFX11-NEXT: s_endpgm ; -; GFX9-LABEL: v_fcmp_f16_ule: -; GFX9: ; %bb.0: -; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c -; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_cmp_ngt_f16_e64 s[0:1], s4, v0 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: v_mov_b32_e32 v1, s1 -; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] -; GFX9-NEXT: s_endpgm +; SDAG-VI-LABEL: v_fcmp_f16_ule: +; SDAG-VI: ; %bb.0: +; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c +; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 +; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) +; SDAG-VI-NEXT: v_cmp_ngt_f16_e64 s[2:3], s2, v0 +; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 +; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 +; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 +; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 +; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; SDAG-VI-NEXT: s_endpgm ; ; GISEL-VI-LABEL: v_fcmp_f16_ule: ; GISEL-VI: ; %bb.0: @@ -2705,25 +2743,22 @@ ; GISEL-VI-NEXT: v_mov_b32_e32 v2, s0 ; GISEL-VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GISEL-VI-NEXT: s_endpgm -; SDAG-VI-LABEL: v_fcmp_f16_ule: -; SDAG-VI: ; %bb.0: -; SDAG-VI-NEXT: s_load_dword s2, s[0:1], 0x2c -; SDAG-VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, 0x5640 -; SDAG-VI-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-VI-NEXT: v_cmp_ngt_f16_e64 s[2:3], s2, v0 -; SDAG-VI-NEXT: v_mov_b32_e32 v0, s0 -; SDAG-VI-NEXT: v_mov_b32_e32 v2, s2 -; SDAG-VI-NEXT: v_mov_b32_e32 v1, s1 -; SDAG-VI-NEXT: v_mov_b32_e32 v3, s3 -; SDAG-VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] -; SDAG-VI-NEXT: s_endpgm +; +; GFX9-LABEL: v_fcmp_f16_ule: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x5640 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_cmp_ngt_f16_e64 s[0:1], s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v0, s0 +; GFX9-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[2:3] +; GFX9-NEXT: s_endpgm %result = call i64 @llvm.amdgcn.fcmp.f16(half %src, half 100.00, i32 13) store i64 %result, ptr addrspace(1) %out ret void } attributes #0 = { nounwind readnone convergent } -;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: -; GCN: {{.*}} -; VI: {{.*}}