[clang] [llvm] [X86][AVX512FP16] implements f64 naive expansion for lower f16 frem (PR #231164)

via cfe-commits cfe-commits at lists.llvm.org
Sun Nov 8 02:42:19 PST 2026


================
@@ -18,87 +24,17 @@ define half @frem(half %x, half %y) nounwind {
 define <2 x half> @frem_vec2(<2 x half> %x, <2 x half> %y) nounwind {
 ; CHECK-LABEL: frem_vec2:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    subq $88, %rsp
-; CHECK-NEXT:    vmovapd %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; CHECK-NEXT:    vmovapd %xmm0, (%rsp) # 16-byte Spill
-; CHECK-NEXT:    vpsrldq {{.*#+}} xmm2 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; CHECK-NEXT:    vcvtsh2ss %xmm2, %xmm2, %xmm0
-; CHECK-NEXT:    vpsrldq {{.*#+}} xmm2 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; CHECK-NEXT:    vcvtsh2ss %xmm2, %xmm2, %xmm1
-; CHECK-NEXT:    callq fmodf at PLT
-; CHECK-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; CHECK-NEXT:    vpermilps $255, (%rsp), %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm0 = mem[3,3,3,3]
-; CHECK-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm1 = mem[3,3,3,3]
-; CHECK-NEXT:    vcvtsh2ss %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    callq fmodf at PLT
-; CHECK-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; CHECK-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; CHECK-NEXT:    vpsrldq $10, (%rsp), %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm0 = mem[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; CHECK-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vpsrldq $10, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm1 = mem[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; CHECK-NEXT:    vcvtsh2ss %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    callq fmodf at PLT
-; CHECK-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; CHECK-NEXT:    vpermilpd $1, (%rsp), %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm0 = mem[1,0]
-; CHECK-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm1 = mem[1,0]
-; CHECK-NEXT:    vcvtsh2ss %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    callq fmodf at PLT
-; CHECK-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; CHECK-NEXT:    vpunpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; CHECK-NEXT:    vpsrlq $48, (%rsp), %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vpsrlq $48, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; CHECK-NEXT:    vcvtsh2ss %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    callq fmodf at PLT
-; CHECK-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; CHECK-NEXT:    vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm0 = mem[1,1,3,3]
-; CHECK-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vmovshdup {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm1 = mem[1,1,3,3]
-; CHECK-NEXT:    vcvtsh2ss %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    callq fmodf at PLT
-; CHECK-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; CHECK-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; CHECK-NEXT:    vmovaps (%rsp), %xmm0 # 16-byte Reload
-; CHECK-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; CHECK-NEXT:    vcvtsh2ss %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    callq fmodf at PLT
-; CHECK-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; CHECK-NEXT:    vpsrld $16, (%rsp), %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vpsrld $16, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; CHECK-NEXT:    vcvtsh2ss %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    callq fmodf at PLT
-; CHECK-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; CHECK-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; CHECK-NEXT:    vpunpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
-; CHECK-NEXT:    addq $88, %rsp
+; CHECK-NEXT:    vcvtph2pd %xmm1, %xmm1
+; CHECK-NEXT:    vcvtph2pd %xmm0, %xmm0
+; CHECK-NEXT:    vdivpd %xmm1, %xmm0, %xmm2
+; CHECK-NEXT:    vroundpd $11, %xmm2, %xmm2
+; CHECK-NEXT:    vfnmadd213pd {{.*#+}} xmm2 = -(xmm1 * xmm2) + xmm0
+; CHECK-NEXT:    vpternlogq {{.*#+}} xmm2 = xmm0 ^ (m64bcst & (xmm2 ^ xmm0))
+; CHECK-NEXT:    vfpclasspd $153, %xmm0, %k0 # k0 = isQuietNaN(xmm0) | isPositiveInfinity(xmm0) | isNegativeInfinity(xmm0) | isSignalingNaN(xmm0)
----------------
Andarwinux wrote:

Why this is vfpclasspd instead of vfpclassph?

https://github.com/llvm/llvm-project/pull/231164


More information about the cfe-commits mailing list