[RISCV] Expand load extension / truncate store for bf16

[joshua-arch1]

Summary

Currentlt, bf16 operations are automatically supported by promoting to float. This patch adds bf16 support by ensuring that load extension / truncate store operations are properly expanded.

[craig.topper]

llvm/test/CodeGen/RISCV/bf16-promote.ll

2

Use -mattr= and -target-abi=

[craig.topper]

LGTM with the comment addressed.

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

setOperationAction(FPRndMode, MVT::f32,
                        Subtarget.hasStdExtZfa() ? Legal : Custom);
     setCondCodeAction(FPCCToExpand, MVT::f32, Expand);
     setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
     setOperationAction(ISD::SELECT, MVT::f32, Custom);
     setOperationAction(ISD::BR_CC, MVT::f32, Expand);
     setOperationAction(FPOpToExpand, MVT::f32, Expand);
     setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
     setTruncStoreAction(MVT::f32, MVT::f16, Expand);
+    setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::bf16, Expand);
+    setTruncStoreAction(MVT::f32, MVT::bf16, Expand);
     setOperationAction(ISD::IS_FPCLASS, MVT::f32, Custom);
     setOperationAction(ISD::BF16_TO_FP, MVT::f32, Custom);
     setOperationAction(ISD::FP_TO_BF16, MVT::f32,
                        Subtarget.isSoftFPABI() ? LibCall : Custom);
     setOperationAction(ISD::FP_TO_FP16, MVT::f32, Custom);
     setOperationAction(ISD::FP16_TO_FP, MVT::f32, Custom);
 
     if (Subtarget.hasStdExtZfa())
if (Subtarget.hasStdExtDOrZdinx()) {
     setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
     setOperationAction(ISD::SELECT, MVT::f64, Custom);
     setOperationAction(ISD::BR_CC, MVT::f64, Expand);
     setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f32, Expand);
     setTruncStoreAction(MVT::f64, MVT::f32, Expand);
     setOperationAction(FPOpToExpand, MVT::f64, Expand);
     setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Expand);
     setTruncStoreAction(MVT::f64, MVT::f16, Expand);
+    setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::bf16, Expand);
+    setTruncStoreAction(MVT::f64, MVT::bf16, Expand);
     setOperationAction(ISD::IS_FPCLASS, MVT::f64, Custom);
     setOperationAction(ISD::BF16_TO_FP, MVT::f64, Custom);
     setOperationAction(ISD::FP_TO_BF16, MVT::f64,
                        Subtarget.isSoftFPABI() ? LibCall : Custom);
     setOperationAction(ISD::FP_TO_FP16, MVT::f64, Custom);
     setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
   }
 

llvm/test/CodeGen/RISCV/bf16-promote.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv64 -mattr +d -target-abi lp64d < %s | FileCheck --check-prefixes=CHECK,RV64 %s
+; RUN: llc -mtriple=riscv32 -mattr +d -target-abi ilp32d < %s | FileCheck --check-prefixes=CHECK,RV32 %s
+
+define void @test_load_store(ptr %p, ptr %q) nounwind {
+; CHECK-LABEL: test_load_store:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lh a0, 0(a0)
+; CHECK-NEXT:    sh a0, 0(a1)
+; CHECK-NEXT:    ret
+  %a = load bfloat, ptr %p
+  store bfloat %a, ptr %q
+  ret void
+}
+
+define float @test_fpextend_float(ptr %p) nounwind {
+; CHECK-LABEL: test_fpextend_float:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lhu a0, 0(a0)
+; CHECK-NEXT:    slli a0, a0, 16
+; CHECK-NEXT:    fmv.w.x fa0, a0
+; CHECK-NEXT:    ret
+  %a = load bfloat, ptr %p
+  %r = fpext bfloat %a to float
+  ret float %r
+}
+
+define double @test_fpextend_double(ptr %p) nounwind {
+; CHECK-LABEL: test_fpextend_double:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lhu a0, 0(a0)
+; CHECK-NEXT:    slli a0, a0, 16
+; CHECK-NEXT:    fmv.w.x fa5, a0
+; CHECK-NEXT:    fcvt.d.s fa0, fa5
+; CHECK-NEXT:    ret
+  %a = load bfloat, ptr %p
+  %r = fpext bfloat %a to double
+  ret double %r
+}
+
+define void @test_fptrunc_float(float %f, ptr %p) nounwind {
+; RV64-LABEL: test_fptrunc_float:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi sp, sp, -16
+; RV64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
+; RV64-NEXT:    mv s0, a0
+; RV64-NEXT:    call __truncsfbf2@plt
+; RV64-NEXT:    fmv.x.w a0, fa0
+; RV64-NEXT:    sh a0, 0(s0)
+; RV64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
+; RV64-NEXT:    addi sp, sp, 16
+; RV64-NEXT:    ret
+;
+; RV32-LABEL: test_fptrunc_float:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi sp, sp, -16
+; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
+; RV32-NEXT:    mv s0, a0
+; RV32-NEXT:    call __truncsfbf2@plt
+; RV32-NEXT:    fmv.x.w a0, fa0
+; RV32-NEXT:    sh a0, 0(s0)
+; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
+; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    ret
+  %a = fptrunc float %f to bfloat
+  store bfloat %a, ptr %p
+  ret void
+}
+
+define void @test_fptrunc_double(double %d, ptr %p) nounwind {
+; RV64-LABEL: test_fptrunc_double:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi sp, sp, -16
+; RV64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
+; RV64-NEXT:    mv s0, a0
+; RV64-NEXT:    call __truncdfbf2@plt
+; RV64-NEXT:    fmv.x.w a0, fa0
+; RV64-NEXT:    sh a0, 0(s0)
+; RV64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
+; RV64-NEXT:    addi sp, sp, 16
+; RV64-NEXT:    ret
+;
+; RV32-LABEL: test_fptrunc_double:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi sp, sp, -16
+; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
+; RV32-NEXT:    mv s0, a0
+; RV32-NEXT:    call __truncdfbf2@plt
+; RV32-NEXT:    fmv.x.w a0, fa0
+; RV32-NEXT:    sh a0, 0(s0)
+; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
+; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    ret
+  %a = fptrunc double %d to bfloat
+  store bfloat %a, ptr %p
+  ret void
+}
+
+define void @test_fadd(ptr %p, ptr %q) nounwind {
+; RV64-LABEL: test_fadd:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi sp, sp, -16
+; RV64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
+; RV64-NEXT:    lhu a1, 0(a1)
+; RV64-NEXT:    mv s0, a0
+; RV64-NEXT:    lhu a0, 0(a0)
+; RV64-NEXT:    slli a1, a1, 16
+; RV64-NEXT:    fmv.w.x fa5, a1
+; RV64-NEXT:    slli a0, a0, 16
+; RV64-NEXT:    fmv.w.x fa4, a0
+; RV64-NEXT:    fadd.s fa0, fa4, fa5
+; RV64-NEXT:    call __truncsfbf2@plt
+; RV64-NEXT:    fmv.x.w a0, fa0
+; RV64-NEXT:    sh a0, 0(s0)
+; RV64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
+; RV64-NEXT:    addi sp, sp, 16
+; RV64-NEXT:    ret
+;
+; RV32-LABEL: test_fadd:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi sp, sp, -16
+; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
+; RV32-NEXT:    lhu a1, 0(a1)
+; RV32-NEXT:    mv s0, a0
+; RV32-NEXT:    lhu a0, 0(a0)
+; RV32-NEXT:    slli a1, a1, 16
+; RV32-NEXT:    fmv.w.x fa5, a1
+; RV32-NEXT:    slli a0, a0, 16
+; RV32-NEXT:    fmv.w.x fa4, a0
+; RV32-NEXT:    fadd.s fa0, fa4, fa5
+; RV32-NEXT:    call __truncsfbf2@plt
+; RV32-NEXT:    fmv.x.w a0, fa0
+; RV32-NEXT:    sh a0, 0(s0)
+; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
+; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    ret
+  %a = load bfloat, ptr %p
+  %b = load bfloat, ptr %q
+  %r = fadd bfloat %a, %b
+  store bfloat %r, ptr %p
+  ret void
+}
 No newline at end of file