[llvm] goldsteinn/icmp to fcmp (PR #82290)

[via llvm-commits]

================
@@ -1067,19 +1073,21 @@ define <16 x i32> @v16i32(<16 x i32> %x, <16 x i32> %y) nounwind {
 ; AVX1-NEXT:    vxorps %ymm4, %ymm2, %ymm2
 ; AVX1-NEXT:    vblendvps %ymm0, %ymm2, %ymm7, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX1-NEXT:    vpaddd %xmm2, %xmm5, %xmm2
-; AVX1-NEXT:    vpaddd %xmm3, %xmm1, %xmm6
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm6, %ymm7
-; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpcmpgtd %xmm6, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
+; AVX1-NEXT:    vpaddd %xmm2, %xmm6, %xmm2
+; AVX1-NEXT:    vpaddd %xmm3, %xmm1, %xmm7
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm7, %ymm8
+; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm6, %xmm6
+; AVX1-NEXT:    vpcmpgtd %xmm7, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
+; AVX1-NEXT:    vcvtdq2ps %ymm3, %ymm3
+; AVX1-NEXT:    vcmpltps %ymm5, %ymm3, %ymm3
----------------
goldsteinn wrote:

Same as above.

https://github.com/llvm/llvm-project/pull/82290