RISC-V: Add mode switching target hook to insert rounding mode config for fixed-point instructions

Hi, this patch support the new coming fixed-point intrinsics:
https://github.com/riscv-non-isa/rvv-intrinsic-doc/pull/222

Insert fixed-point rounding mode configuration by mode switching target hook.

Mode switching target hook is implemented applying LCM (Lazy code Motion).

So the performance && correctness can be well trusted.

Here is the example:

void f (void * in, void *out, int32_t x, int n, int m)
{
  for (int i = 0; i < n; i++) {
    vint32m1_t v = __riscv_vle32_v_i32m1 (in + i, 4);
    vint32m1_t v2 = __riscv_vle32_v_i32m1_tu (v, in + 100 + i, 4);
    vint32m1_t v3 = __riscv_vaadd_vx_i32m1 (v2, 0, VXRM_RDN, 4);
    v3 = __riscv_vaadd_vx_i32m1 (v3, 3, VXRM_RDN, 4);
    __riscv_vse32_v_i32m1 (out + 100 + i, v3, 4);
  }

  for (int i = 0; i < n; i++) {
    vint32m1_t v = __riscv_vle32_v_i32m1 (in + i + 1000, 4);
    vint32m1_t v2 = __riscv_vle32_v_i32m1_tu (v, in + 100 + i + 1000, 4);
    vint32m1_t v3 = __riscv_vaadd_vx_i32m1 (v2, 0, VXRM_RDN, 4);
    v3 = __riscv_vaadd_vx_i32m1 (v3, 3, VXRM_RDN, 4);
    __riscv_vse32_v_i32m1 (out + 100 + i + 1000, v3, 4);
  }
}

ASM:

...
csrwi   vxrm,2
vsetivli        zero,4,e32,m1,tu,ma
...
Loop 1
...
Loop 2

mode switching can global recognize both Loop 1 and Loop 2 are using RDN
rounding mode and hoist such single "csrwi vxrm,2" to dominate both Loop 1
and Loop 2.

Besides, I have add correctness check sanity tests in this patch too.

Ok for trunk ?

gcc/ChangeLog:

	* config/riscv/riscv-opts.h (enum riscv_entity): New enum.
	* config/riscv/riscv.cc (riscv_emit_mode_set): New function.
	(riscv_mode_needed): Ditto.
	(riscv_mode_after): Ditto.
	(riscv_mode_entry): Ditto.
	(riscv_mode_exit): Ditto.
	(riscv_mode_priority): Ditto.
	(TARGET_MODE_EMIT): New target hook.
	(TARGET_MODE_NEEDED): Ditto.
	(TARGET_MODE_AFTER): Ditto.
	(TARGET_MODE_ENTRY): Ditto.
	(TARGET_MODE_EXIT): Ditto.
	(TARGET_MODE_PRIORITY): Ditto.
	* config/riscv/riscv.h (OPTIMIZE_MODE_SWITCHING): Ditto.
	(NUM_MODES_FOR_MODE_SWITCHING): Ditto.
	* config/riscv/riscv.md: Add csrwvxrm.
	* config/riscv/vector.md (rnu,rne,rdn,rod,none): New attribute.
	(vxrmsi): New pattern.

gcc/testsuite/ChangeLog:

	* gcc.target/riscv/rvv/base/vxrm-10.c: New test.
	* gcc.target/riscv/rvv/base/vxrm-6.c: New test.
	* gcc.target/riscv/rvv/base/vxrm-7.c: New test.
	* gcc.target/riscv/rvv/base/vxrm-8.c: New test.
	* gcc.target/riscv/rvv/base/vxrm-9.c: New test.

gcc/config/riscv/riscv-opts.h

@@ -91,6 +91,14 @@ enum riscv_multilib_select_kind {
   select_by_abi,
 };
 
+/* ENTITIES in mode switching.  */
+enum riscv_entity
+{
+  RISCV_VXRM = 0,
+  RISCV_FRM,
+  MAX_RISCV_ENTITIES
+};
+
 #define MASK_ZICSR    (1 << 0)
 #define MASK_ZIFENCEI (1 << 1)
 

gcc/config/riscv/riscv.cc

@@ -7513,6 +7513,95 @@ riscv_vectorize_preferred_vector_alignment (const_tree type)
   return TYPE_ALIGN (type);
 }
 
+/* Implement Mode switching.  */
+
+static void
+riscv_emit_mode_set (int entity, int mode, int prev_mode,
+		     HARD_REG_SET regs_live ATTRIBUTE_UNUSED)
+{
+  switch (entity)
+    {
+    case RISCV_VXRM:
+      if (mode != VXRM_MODE_NONE && mode != prev_mode)
+	emit_insn (gen_vxrmsi (gen_int_mode (mode, SImode)));
+      break;
+    default:
+      gcc_unreachable ();
+    }
+}
+
+/* Return mode that entity must be switched into
+   prior to the execution of insn.  */
+
+static int
+riscv_mode_needed (int entity, rtx_insn *insn)
+{
+  switch (entity)
+    {
+    case RISCV_VXRM:
+      return recog_memoized (insn) >= 0 ? get_attr_vxrm_mode (insn)
+					: VXRM_MODE_NONE;
+    default:
+      gcc_unreachable ();
+    }
+}
+
+/* Return the mode that an insn results in.  */
+
+static int
+riscv_mode_after (int entity, int mode, rtx_insn *insn)
+{
+  switch (entity)
+    {
+    case RISCV_VXRM:
+      if (recog_memoized (insn) >= 0)
+	return reg_mentioned_p (gen_rtx_REG (SImode, VXRM_REGNUM),
+				PATTERN (insn))
+		 ? get_attr_vxrm_mode (insn)
+		 : mode;
+      else
+	return mode;
+    default:
+      gcc_unreachable ();
+    }
+}
+
+/* Return a mode that ENTITY is assumed to be
+   switched to at function entry.  */
+
+static int
+riscv_mode_entry (int entity)
+{
+  switch (entity)
+    {
+    case RISCV_VXRM:
+      return VXRM_MODE_NONE;
+    default:
+      gcc_unreachable ();
+    }
+}
+
+/* Return a mode that ENTITY is assumed to be
+   switched to at function exit.  */
+
+static int
+riscv_mode_exit (int entity)
+{
+  switch (entity)
+    {
+    case RISCV_VXRM:
+      return VXRM_MODE_NONE;
+    default:
+      gcc_unreachable ();
+    }
+}
+
+static int
+riscv_mode_priority (int, int n)
+{
+  return n;
+}
+
 /* Initialize the GCC target structure.  */
 #undef TARGET_ASM_ALIGNED_HI_OP
 #define TARGET_ASM_ALIGNED_HI_OP "\t.half\t"
@@ -7789,6 +7878,21 @@ riscv_vectorize_preferred_vector_alignment (const_tree type)
 #define TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT \
   riscv_vectorize_preferred_vector_alignment
 
+/* Mode switching hooks.  */
+
+#undef TARGET_MODE_EMIT
+#define TARGET_MODE_EMIT riscv_emit_mode_set
+#undef TARGET_MODE_NEEDED
+#define TARGET_MODE_NEEDED riscv_mode_needed
+#undef TARGET_MODE_AFTER
+#define TARGET_MODE_AFTER riscv_mode_after
+#undef TARGET_MODE_ENTRY
+#define TARGET_MODE_ENTRY riscv_mode_entry
+#undef TARGET_MODE_EXIT
+#define TARGET_MODE_EXIT riscv_mode_exit
+#undef TARGET_MODE_PRIORITY
+#define TARGET_MODE_PRIORITY riscv_mode_priority
+
 struct gcc_target targetm = TARGET_INITIALIZER;
 
 #include "gt-riscv.h"

gcc/config/riscv/riscv.h

@@ -845,7 +845,7 @@ typedef struct {
   "fs0", "fs1", "fa0", "fa1", "fa2", "fa3", "fa4", "fa5",	\
   "fa6", "fa7", "fs2", "fs3", "fs4", "fs5", "fs6", "fs7",	\
   "fs8", "fs9", "fs10","fs11","ft8", "ft9", "ft10","ft11",	\
-  "arg", "frame", "vl", "vtype", "N/A", "N/A", "N/A", "N/A",    \
+  "arg", "frame", "vl", "vtype", "vxrm", "N/A", "N/A", "N/A",   \
   "N/A", "N/A", "N/A", "N/A", "N/A", "N/A", "N/A", "N/A",	\
   "N/A", "N/A", "N/A", "N/A", "N/A", "N/A", "N/A", "N/A",	\
   "N/A", "N/A", "N/A", "N/A", "N/A", "N/A", "N/A", "N/A",	\
@@ -1107,4 +1107,8 @@ extern void riscv_remove_unneeded_save_restore_calls (void);
    CONST1_RTX for the simplification.  */
 #define VECTOR_STORE_FLAG_VALUE(MODE) CONSTM1_RTX (GET_MODE_INNER (MODE))
 
+/* Mode switching (Lazy code motion) for RVV rounding mode instructions.  */
+#define OPTIMIZE_MODE_SWITCHING(ENTITY) (TARGET_VECTOR)
+#define NUM_MODES_FOR_MODE_SWITCHING {VXRM_MODE_NONE}
+
 #endif /* ! GCC_RISCV_H */

gcc/config/riscv/riscv.md

@@ -284,6 +284,7 @@
 ;; Classification of RVV instructions which will be added to each RVV .md pattern and used by scheduler.
 ;; rdvlenb     vector byte length vlenb csrr read
 ;; rdvl        vector length vl csrr read
+;; wrvxrm      vector fixed-point rounding mode write
 ;; vsetvl      vector configuration-setting instrucions
 ;; 7. Vector Loads and Stores
 ;; vlde        vector unit-stride load instructions
@@ -387,7 +388,7 @@
    mtc,mfc,const,arith,logical,shift,slt,imul,idiv,move,fmove,fadd,fmul,
    fmadd,fdiv,fcmp,fcvt,fsqrt,multi,auipc,sfb_alu,nop,ghost,bitmanip,rotate,
    clmul,min,max,minu,maxu,clz,ctz,cpop,
-   atomic,condmove,crypto,rdvlenb,rdvl,vsetvl,vlde,vste,vldm,vstm,vlds,vsts,
+   atomic,condmove,crypto,rdvlenb,rdvl,wrvxrm,vsetvl,vlde,vste,vldm,vstm,vlds,vsts,
    vldux,vldox,vstux,vstox,vldff,vldr,vstr,
    vlsegde,vssegte,vlsegds,vssegts,vlsegdux,vlsegdox,vssegtux,vssegtox,vlsegdff,
    vialu,viwalu,vext,vicalu,vshift,vnshift,vicmp,viminmax,

gcc/config/riscv/vector.md

@@ -407,6 +407,26 @@
 	   (symbol_ref "INTVAL (operands[4])")]
 	(const_int INVALID_ATTRIBUTE)))
 
+;; Defines rounding mode of an fixed-point operation.
+
+(define_attr "vxrm_mode" "rnu,rne,rdn,rod,none"
+  (cond [(and (eq_attr "type" "vsalu,vaalu,vsmul,vsshift,vnclip")
+	      (match_test "INTVAL(operands[9]) == riscv_vector::VXRM_RNU"))
+	 (const_string "rnu")
+
+	 (and (eq_attr "type" "vsalu,vaalu,vsmul,vsshift,vnclip")
+	      (match_test "INTVAL(operands[9]) == riscv_vector::VXRM_RNE"))
+	 (const_string "rne")
+
+	 (and (eq_attr "type" "vsalu,vaalu,vsmul,vsshift,vnclip")
+	      (match_test "INTVAL(operands[9]) == riscv_vector::VXRM_RDN"))
+	 (const_string "rdn")
+
+	 (and (eq_attr "type" "vsalu,vaalu,vsmul,vsshift,vnclip")
+	      (match_test "INTVAL(operands[9]) == riscv_vector::VXRM_ROD"))
+	 (const_string "rod")]
+	 (const_string "none")))
+
 ;; -----------------------------------------------------------------
 ;; ---- Miscellaneous Operations
 ;; -----------------------------------------------------------------
@@ -594,6 +614,15 @@
   "TARGET_VECTOR"
   "")
 
+;; Set VXRM
+(define_insn "vxrmsi"
+  [(set (reg:SI VXRM_REGNUM)
+	(match_operand 0 "const_int_operand" "i"))]
+  "TARGET_VECTOR"
+  "csrwi\tvxrm,%0"
+  [(set_attr "type" "wrvxrm")
+   (set_attr "mode" "SI")])
+
 ;; -----------------------------------------------------------------
 ;; ---- Moves Operations
 ;; -----------------------------------------------------------------

gcc/testsuite/gcc.target/riscv/rvv/base/vxrm-10.c

@@ -0,0 +1,26 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64d -O3 -fno-schedule-insns -fno-schedule-insns2" } */
+
+#include "riscv_vector.h"
+
+void f (void * in, void *out, int32_t x, int n, int m)
+{
+  for (int i = 0; i < n; i++) {
+    vint32m1_t v = __riscv_vle32_v_i32m1 (in + i, 4);
+    vint32m1_t v2 = __riscv_vle32_v_i32m1_tu (v, in + 100 + i, 4);
+    vint32m1_t v3 = __riscv_vaadd_vx_i32m1 (v2, 0, VXRM_RDN, 4);
+    v3 = __riscv_vaadd_vx_i32m1 (v3, 3, VXRM_RDN, 4);
+    __riscv_vse32_v_i32m1 (out + 100 + i, v3, 4);
+  }
+  
+  for (int i = 0; i < n; i++) {
+    vint32m1_t v = __riscv_vle32_v_i32m1 (in + i + 1000, 4);
+    vint32m1_t v2 = __riscv_vle32_v_i32m1_tu (v, in + 100 + i + 1000, 4);
+    vint32m1_t v3 = __riscv_vaadd_vx_i32m1 (v2, 0, VXRM_RNE, 4);
+    v3 = __riscv_vaadd_vx_i32m1 (v3, 3, VXRM_RNE, 4);
+    __riscv_vse32_v_i32m1 (out + 100 + i + 1000, v3, 4);
+  }
+}
+
+/* { dg-final { scan-assembler-times {csrwi\s+vxrm,\s*1} 1 } } */
+/* { dg-final { scan-assembler-times {csrwi\s+vxrm,\s*2} 1 } } */

gcc/testsuite/gcc.target/riscv/rvv/base/vxrm-6.c

@@ -0,0 +1,15 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64d -O3" } */
+
+#include "riscv_vector.h"
+
+void f (void * in, void *out, int32_t x, int n, int m)
+{
+  vint32m1_t v = __riscv_vle32_v_i32m1 (in, 4);
+  vint32m1_t v2 = __riscv_vle32_v_i32m1_tu (v, in + 100, 4);
+  vint32m1_t v3 = __riscv_vaadd_vx_i32m1 (v2, 0, VXRM_RDN, 4);
+  v3 = __riscv_vaadd_vx_i32m1 (v3, 3, VXRM_RDN, 4);
+  __riscv_vse32_v_i32m1 (out + 100, v3, 4);
+}
+
+/* { dg-final { scan-assembler-times {csrwi\s+vxrm,\s*2} 1 } } */

gcc/testsuite/gcc.target/riscv/rvv/base/vxrm-7.c

@@ -0,0 +1,16 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64d -O3" } */
+
+#include "riscv_vector.h"
+
+void f (void * in, void *out, int32_t x, int n, int m)
+{
+  vint32m1_t v = __riscv_vle32_v_i32m1 (in, 4);
+  vint32m1_t v2 = __riscv_vle32_v_i32m1_tu (v, in + 100, 4);
+  vint32m1_t v3 = __riscv_vaadd_vx_i32m1 (v2, 0, VXRM_RNE, 4);
+  v3 = __riscv_vaadd_vx_i32m1 (v3, 3, VXRM_RDN, 4);
+  __riscv_vse32_v_i32m1 (out + 100, v3, 4);
+}
+
+/* { dg-final { scan-assembler-times {csrwi\s+vxrm,\s*1} 1 } } */
+/* { dg-final { scan-assembler-times {csrwi\s+vxrm,\s*2} 1 } } */

gcc/testsuite/gcc.target/riscv/rvv/base/vxrm-8.c

@@ -0,0 +1,18 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64d -O3 -fno-schedule-insns -fno-schedule-insns2" } */
+
+#include "riscv_vector.h"
+
+void f (void * in, void *out, int32_t x, int n, int m)
+{
+  for (int i = 0; i < n; i++) {
+    vint32m1_t v = __riscv_vle32_v_i32m1 (in + i, 4);
+    vint32m1_t v2 = __riscv_vle32_v_i32m1_tu (v, in + 100 + i, 4);
+    vint32m1_t v3 = __riscv_vaadd_vx_i32m1 (v2, 0, VXRM_RDN, 4);
+    v3 = __riscv_vaadd_vx_i32m1 (v3, 3, VXRM_RDN, 4);
+    __riscv_vse32_v_i32m1 (out + 100 + i, v3, 4);
+  }
+}
+
+/* { dg-final { scan-assembler-times {csrwi\s+vxrm,\s*2\s+vsetivli\s+zero,\s*4,\s*e32,\s*m1,\s*tu,\s*m[au]} 1 } } */
+/* { dg-final { scan-assembler-times {csrwi\s+vxrm,\s*2} 1 } } */

gcc/testsuite/gcc.target/riscv/rvv/base/vxrm-9.c

@@ -0,0 +1,26 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64d -O3 -fno-schedule-insns -fno-schedule-insns2" } */
+
+#include "riscv_vector.h"
+
+void f (void * in, void *out, int32_t x, int n, int m)
+{
+  for (int i = 0; i < n; i++) {
+    vint32m1_t v = __riscv_vle32_v_i32m1 (in + i, 4);
+    vint32m1_t v2 = __riscv_vle32_v_i32m1_tu (v, in + 100 + i, 4);
+    vint32m1_t v3 = __riscv_vaadd_vx_i32m1 (v2, 0, VXRM_RDN, 4);
+    v3 = __riscv_vaadd_vx_i32m1 (v3, 3, VXRM_RDN, 4);
+    __riscv_vse32_v_i32m1 (out + 100 + i, v3, 4);
+  }
+  
+  for (int i = 0; i < n; i++) {
+    vint32m1_t v = __riscv_vle32_v_i32m1 (in + i + 1000, 4);
+    vint32m1_t v2 = __riscv_vle32_v_i32m1_tu (v, in + 100 + i + 1000, 4);
+    vint32m1_t v3 = __riscv_vaadd_vx_i32m1 (v2, 0, VXRM_RDN, 4);
+    v3 = __riscv_vaadd_vx_i32m1 (v3, 3, VXRM_RDN, 4);
+    __riscv_vse32_v_i32m1 (out + 100 + i + 1000, v3, 4);
+  }
+}
+
+/* { dg-final { scan-assembler-times {csrwi\s+vxrm,\s*2\s+vsetivli\s+zero,\s*4,\s*e32,\s*m1,\s*tu,\s*m[au]} 1 } } */
+/* { dg-final { scan-assembler-times {csrwi\s+vxrm,\s*2} 1 } } */