[llvm] [AArch64][GlobalISel] SIMD fpcvt codegen for fptoi(_sat) (PR #160831)
David Green via llvm-commits
llvm-commits at lists.llvm.org
Thu Oct 23 02:15:16 PDT 2025
================
@@ -0,0 +1,2039 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple aarch64-unknown-unknown -mattr=+fprcvt,+fullfp16 | FileCheck %s --check-prefixes=CHECK,CHECK-SD
+; RUN: llc < %s -mtriple aarch64-unknown-unknown -global-isel -global-isel-abort=2 -mattr=+fprcvt,+fullfp16 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
+
+; CHECK-GI: warning: Instruction selection used fallback path for fptosi_i32_f16_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptosi_i64_f16_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptosi_i64_f32_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptosi_i32_f64_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptosi_i64_f64_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptosi_i32_f32_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptoui_i32_f16_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptoui_i64_f16_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptoui_i64_f32_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptoui_i32_f64_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptoui_i64_f64_simd
+; CHECK-GI-NEXT: warning: Instruction selection used fallback path for fptoui_i32_f32_simd
+
+;
+; FPTOI
+;
+
+define float @test_fptosi_f16_i32_simd(half %a) {
+; CHECK-LABEL: test_fptosi_f16_i32_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs s0, h0
+; CHECK-NEXT: ret
+ %r = fptosi half %a to i32
+ %bc = bitcast i32 %r to float
+ ret float %bc
+}
+
+define double @test_fptosi_f16_i64_simd(half %a) {
+; CHECK-LABEL: test_fptosi_f16_i64_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs d0, h0
+; CHECK-NEXT: ret
+ %r = fptosi half %a to i64
+ %bc = bitcast i64 %r to double
+ ret double %bc
+}
+
+define float @test_fptosi_f64_i32_simd(double %a) {
+; CHECK-LABEL: test_fptosi_f64_i32_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs s0, d0
+; CHECK-NEXT: ret
+ %r = fptosi double %a to i32
+ %bc = bitcast i32 %r to float
+ ret float %bc
+}
+
+define double @test_fptosi_f32_i64_simd(float %a) {
+; CHECK-LABEL: test_fptosi_f32_i64_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs d0, s0
+; CHECK-NEXT: ret
+ %r = fptosi float %a to i64
+ %bc = bitcast i64 %r to double
+ ret double %bc
+}
+
+define double @test_fptosi_f64_i64_simd(double %a) {
+; CHECK-LABEL: test_fptosi_f64_i64_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs d0, d0
+; CHECK-NEXT: ret
+ %r = fptosi double %a to i64
+ %bc = bitcast i64 %r to double
+ ret double %bc
+}
+
+
+define float @test_fptosi_f32_i32_simd(float %a) {
+; CHECK-LABEL: test_fptosi_f32_i32_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs s0, s0
+; CHECK-NEXT: ret
+ %r = fptosi float %a to i32
+ %bc = bitcast i32 %r to float
+ ret float %bc
+}
+
+define float @test_fptoui_f16_i32_simd(half %a) {
+; CHECK-LABEL: test_fptoui_f16_i32_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu s0, h0
+; CHECK-NEXT: ret
+ %r = fptoui half %a to i32
+ %bc = bitcast i32 %r to float
+ ret float %bc
+}
+
+define double @test_fptoui_f16_i64_simd(half %a) {
+; CHECK-LABEL: test_fptoui_f16_i64_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu d0, h0
+; CHECK-NEXT: ret
+ %r = fptoui half %a to i64
+ %bc = bitcast i64 %r to double
+ ret double %bc
+}
+
+define float @test_fptoui_f64_i32_simd(double %a) {
+; CHECK-LABEL: test_fptoui_f64_i32_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu s0, d0
+; CHECK-NEXT: ret
+ %r = fptoui double %a to i32
+ %bc = bitcast i32 %r to float
+ ret float %bc
+}
+
+define double @test_fptoui_f32_i64_simd(float %a) {
+; CHECK-LABEL: test_fptoui_f32_i64_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu d0, s0
+; CHECK-NEXT: ret
+ %r = fptoui float %a to i64
+ %bc = bitcast i64 %r to double
+ ret double %bc
+}
+
+define double @test_fptoui_f64_i64_simd(double %a) {
+; CHECK-LABEL: test_fptoui_f64_i64_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu d0, d0
+; CHECK-NEXT: ret
+ %r = fptoui double %a to i64
+ %bc = bitcast i64 %r to double
+ ret double %bc
+}
+
+
+define float @test_fptoui_f32_i32_simd(float %a) {
+; CHECK-LABEL: test_fptoui_f32_i32_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu s0, s0
+; CHECK-NEXT: ret
+ %r = fptoui float %a to i32
+ %bc = bitcast i32 %r to float
+ ret float %bc
+}
+
+
+;
+; FPTOI experimental
+;
+
+define float @fptosi_i32_f16_simd(half %x) {
+; CHECK-LABEL: fptosi_i32_f16_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs s0, h0
+; CHECK-NEXT: ret
+ %val = call i32 @llvm.experimental.constrained.fptosi.i32.f16(half %x, metadata !"fpexcept.strict")
+ %sum = bitcast i32 %val to float
+ ret float %sum
+}
+
+define double @fptosi_i64_f16_simd(half %x) {
+; CHECK-LABEL: fptosi_i64_f16_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs d0, h0
+; CHECK-NEXT: ret
+ %val = call i64 @llvm.experimental.constrained.fptosi.i64.f16(half %x, metadata !"fpexcept.strict")
+ %sum = bitcast i64 %val to double
+ ret double %sum
+}
+
+define double @fptosi_i64_f32_simd(float %x) {
+; CHECK-LABEL: fptosi_i64_f32_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs d0, s0
+; CHECK-NEXT: ret
+ %val = call i64 @llvm.experimental.constrained.fptosi.i64.f32(float %x, metadata !"fpexcept.strict")
+ %bc = bitcast i64 %val to double
+ ret double %bc
+}
+
+define float @fptosi_i32_f64_simd(double %x) {
+; CHECK-LABEL: fptosi_i32_f64_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs s0, d0
+; CHECK-NEXT: ret
+ %val = call i32 @llvm.experimental.constrained.fptosi.i32.f64(double %x, metadata !"fpexcept.strict")
+ %bc = bitcast i32 %val to float
+ ret float %bc
+}
+
+define double @fptosi_i64_f64_simd(double %x) {
+; CHECK-LABEL: fptosi_i64_f64_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs d0, d0
+; CHECK-NEXT: ret
+ %val = call i64 @llvm.experimental.constrained.fptosi.i64.f64(double %x, metadata !"fpexcept.strict")
+ %bc = bitcast i64 %val to double
+ ret double %bc
+}
+
+define float @fptosi_i32_f32_simd(float %x) {
+; CHECK-LABEL: fptosi_i32_f32_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzs s0, s0
+; CHECK-NEXT: ret
+ %val = call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %x, metadata !"fpexcept.strict")
+ %bc = bitcast i32 %val to float
+ ret float %bc
+}
+
+
+
+define float @fptoui_i32_f16_simd(half %x) {
+; CHECK-LABEL: fptoui_i32_f16_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu s0, h0
+; CHECK-NEXT: ret
+ %val = call i32 @llvm.experimental.constrained.fptoui.i32.f16(half %x, metadata !"fpexcept.strict")
+ %sum = bitcast i32 %val to float
+ ret float %sum
+}
+
+define double @fptoui_i64_f16_simd(half %x) {
+; CHECK-LABEL: fptoui_i64_f16_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu d0, h0
+; CHECK-NEXT: ret
+ %val = call i64 @llvm.experimental.constrained.fptoui.i64.f16(half %x, metadata !"fpexcept.strict")
+ %sum = bitcast i64 %val to double
+ ret double %sum
+}
+
+define double @fptoui_i64_f32_simd(float %x) {
+; CHECK-LABEL: fptoui_i64_f32_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu d0, s0
+; CHECK-NEXT: ret
+ %val = call i64 @llvm.experimental.constrained.fptoui.i64.f32(float %x, metadata !"fpexcept.strict")
+ %bc = bitcast i64 %val to double
+ ret double %bc
+}
+
+define float @fptoui_i32_f64_simd(double %x) {
+; CHECK-LABEL: fptoui_i32_f64_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu s0, d0
+; CHECK-NEXT: ret
+ %val = call i32 @llvm.experimental.constrained.fptoui.i32.f64(double %x, metadata !"fpexcept.strict")
+ %bc = bitcast i32 %val to float
+ ret float %bc
+}
+
+define double @fptoui_i64_f64_simd(double %x) {
+; CHECK-LABEL: fptoui_i64_f64_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu d0, d0
+; CHECK-NEXT: ret
+ %val = call i64 @llvm.experimental.constrained.fptoui.i64.f64(double %x, metadata !"fpexcept.strict")
+ %bc = bitcast i64 %val to double
+ ret double %bc
+}
+
+define float @fptoui_i32_f32_simd(float %x) {
+; CHECK-LABEL: fptoui_i32_f32_simd:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fcvtzu s0, s0
+; CHECK-NEXT: ret
+ %val = call i32 @llvm.experimental.constrained.fptoui.i32.f32(float %x, metadata !"fpexcept.strict")
+ %bc = bitcast i32 %val to float
+ ret float %bc
+}
+
+;
+; FPTOI rounding
+;
+
+
+define double @fcvtas_ds_round_simd(float %a) {
+; CHECK-SD-LABEL: fcvtas_ds_round_simd:
+; CHECK-SD: // %bb.0:
+; CHECK-SD-NEXT: fcvtas d0, s0
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: fcvtas_ds_round_simd:
----------------
davemgreen wrote:
I believe the frontend produces the llvm.round intrinsics. It could be a missed optimization to not have gisel do the same thing, but gisel needn't do it if it should be done elsewhere, like the mid/front-end.
https://github.com/llvm/llvm-project/pull/160831
More information about the llvm-commits
mailing list