target/arm/vec_helper: Add gvec fp indexed multiply-and-add operations

Add gvec helpers for doing Neon-style indexed non-fused fp
multiply-and-accumulate operations.

Backports commit c50d8d144098a8261233ca31b47e3bc487e112fe

qemu/aarch64.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_aarch64
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_aarch64
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_aarch64
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_aarch64
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_aarch64
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_aarch64
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_aarch64
 #define helper_power_down helper_power_down_aarch64
 #define helper_pre_hvc helper_pre_hvc_aarch64
 #define helper_pre_smc helper_pre_smc_aarch64

qemu/aarch64eb.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_aarch64eb
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_aarch64eb
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_aarch64eb
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_aarch64eb
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_aarch64eb
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_aarch64eb
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_aarch64eb
 #define helper_power_down helper_power_down_aarch64eb
 #define helper_pre_hvc helper_pre_hvc_aarch64eb
 #define helper_pre_smc helper_pre_smc_aarch64eb

qemu/arm.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_arm
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_arm
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_arm
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_arm
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_arm
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_arm
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_arm
 #define helper_power_down helper_power_down_arm
 #define helper_pre_hvc helper_pre_hvc_arm
 #define helper_pre_smc helper_pre_smc_arm

qemu/armeb.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_armeb
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_armeb
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_armeb
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_armeb
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_armeb
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_armeb
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_armeb
 #define helper_power_down helper_power_down_armeb
 #define helper_pre_hvc helper_pre_hvc_armeb
 #define helper_pre_smc helper_pre_smc_armeb

qemu/header_gen.py

@@ -1858,6 +1858,10 @@ symbols = (
     'helper_gvec_vrint_rm_s',
     'helper_gvec_vrintx_h',
     'helper_gvec_vrintx_s',
+    'helper_gvec_fmla_nf_idx_h',
+    'helper_gvec_fmla_nf_idx_s',
+    'helper_gvec_fmls_nf_idx_h',
+    'helper_gvec_fmls_nf_idx_s',
     'helper_power_down',
     'helper_pre_hvc',
     'helper_pre_smc',

qemu/m68k.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_m68k
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_m68k
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_m68k
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_m68k
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_m68k
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_m68k
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_m68k
 #define helper_power_down helper_power_down_m68k
 #define helper_pre_hvc helper_pre_hvc_m68k
 #define helper_pre_smc helper_pre_smc_m68k

qemu/mips.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_mips
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_mips
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_mips
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_mips
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_mips
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_mips
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_mips
 #define helper_power_down helper_power_down_mips
 #define helper_pre_hvc helper_pre_hvc_mips
 #define helper_pre_smc helper_pre_smc_mips

qemu/mips64.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_mips64
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_mips64
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_mips64
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_mips64
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_mips64
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_mips64
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_mips64
 #define helper_power_down helper_power_down_mips64
 #define helper_pre_hvc helper_pre_hvc_mips64
 #define helper_pre_smc helper_pre_smc_mips64

qemu/mips64el.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_mips64el
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_mips64el
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_mips64el
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_mips64el
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_mips64el
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_mips64el
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_mips64el
 #define helper_power_down helper_power_down_mips64el
 #define helper_pre_hvc helper_pre_hvc_mips64el
 #define helper_pre_smc helper_pre_smc_mips64el

qemu/mipsel.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_mipsel
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_mipsel
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_mipsel
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_mipsel
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_mipsel
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_mipsel
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_mipsel
 #define helper_power_down helper_power_down_mipsel
 #define helper_pre_hvc helper_pre_hvc_mipsel
 #define helper_pre_smc helper_pre_smc_mipsel

qemu/powerpc.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_powerpc
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_powerpc
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_powerpc
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_powerpc
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_powerpc
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_powerpc
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_powerpc
 #define helper_power_down helper_power_down_powerpc
 #define helper_pre_hvc helper_pre_hvc_powerpc
 #define helper_pre_smc helper_pre_smc_powerpc

qemu/riscv32.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_riscv32
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_riscv32
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_riscv32
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_riscv32
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_riscv32
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_riscv32
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_riscv32
 #define helper_power_down helper_power_down_riscv32
 #define helper_pre_hvc helper_pre_hvc_riscv32
 #define helper_pre_smc helper_pre_smc_riscv32

qemu/riscv64.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_riscv64
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_riscv64
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_riscv64
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_riscv64
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_riscv64
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_riscv64
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_riscv64
 #define helper_power_down helper_power_down_riscv64
 #define helper_pre_hvc helper_pre_hvc_riscv64
 #define helper_pre_smc helper_pre_smc_riscv64

qemu/sparc.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_sparc
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_sparc
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_sparc
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_sparc
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_sparc
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_sparc
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_sparc
 #define helper_power_down helper_power_down_sparc
 #define helper_pre_hvc helper_pre_hvc_sparc
 #define helper_pre_smc helper_pre_smc_sparc

qemu/sparc64.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_sparc64
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_sparc64
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_sparc64
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_sparc64
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_sparc64
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_sparc64
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_sparc64
 #define helper_power_down helper_power_down_sparc64
 #define helper_pre_hvc helper_pre_hvc_sparc64
 #define helper_pre_smc helper_pre_smc_sparc64

qemu/target/arm/helper.h

@@ -738,6 +738,16 @@ DEF_HELPER_FLAGS_5(gvec_fmul_idx_s, TCG_CALL_NO_RWG,
 DEF_HELPER_FLAGS_5(gvec_fmul_idx_d, TCG_CALL_NO_RWG,
                    void, ptr, ptr, ptr, ptr, i32)
 
+DEF_HELPER_FLAGS_5(gvec_fmla_nf_idx_h, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_5(gvec_fmla_nf_idx_s, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_5(gvec_fmls_nf_idx_h, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_5(gvec_fmls_nf_idx_s, TCG_CALL_NO_RWG,
+                   void, ptr, ptr, ptr, ptr, i32)
+
 DEF_HELPER_FLAGS_6(gvec_fmla_idx_h, TCG_CALL_NO_RWG,
                    void, ptr, ptr, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_6(gvec_fmla_idx_s, TCG_CALL_NO_RWG,

qemu/target/arm/vec_helper.c

@@ -1086,7 +1086,7 @@ DO_MLA_IDX(gvec_mls_idx_d, uint64_t, -,   )
 
 #undef DO_MLA_IDX
 
-#define DO_FMUL_IDX(NAME, TYPE, H) \
+#define DO_FMUL_IDX(NAME, ADD, TYPE, H)                                    \
 void HELPER(NAME)(void *vd, void *vn, void *vm, void *stat, uint32_t desc) \
 {                                                                          \
     intptr_t i, j, oprsz = simd_oprsz(desc);                               \
@@ -1096,15 +1096,33 @@ void HELPER(NAME)(void *vd, void *vn, void *vm, void *stat, uint32_t desc) \
     for (i = 0; i < oprsz / sizeof(TYPE); i += segment) {                  \
         TYPE mm = m[H(i + idx)];                                           \
         for (j = 0; j < segment; j++) {                                    \
-            d[i + j] = TYPE##_mul(n[i + j], mm, stat);                     \
+            d[i + j] = TYPE##_##ADD(d[i + j],                              \
+                                    TYPE##_mul(n[i + j], mm, stat), stat); \
         }                                                                  \
     }                                                                      \
     clear_tail(d, oprsz, simd_maxsz(desc));                                \
 }
 
-DO_FMUL_IDX(gvec_fmul_idx_h, float16, H2)
+#define float16_nop(N, M, S) (M)
-DO_FMUL_IDX(gvec_fmul_idx_s, float32, H4)
+#define float32_nop(N, M, S) (M)
-DO_FMUL_IDX(gvec_fmul_idx_d, float64, )
+#define float64_nop(N, M, S) (M)
+
+DO_FMUL_IDX(gvec_fmul_idx_h, nop, float16, H2)
+DO_FMUL_IDX(gvec_fmul_idx_s, nop, float32, H4)
+DO_FMUL_IDX(gvec_fmul_idx_d, nop, float64, )
+
+/*
+ * Non-fused multiply-accumulate operations, for Neon. NB that unlike
+ * the fused ops below they assume accumulate both from and into Vd.
+ */
+DO_FMUL_IDX(gvec_fmla_nf_idx_h, add, float16, H2)
+DO_FMUL_IDX(gvec_fmla_nf_idx_s, add, float32, H4)
+DO_FMUL_IDX(gvec_fmls_nf_idx_h, sub, float16, H2)
+DO_FMUL_IDX(gvec_fmls_nf_idx_s, sub, float32, H4)
+
+#undef float16_nop
+#undef float32_nop
+#undef float64_nop
 
 #undef DO_FMUL_IDX
 

qemu/x86_64.h

@@ -1852,6 +1852,10 @@
 #define helper_gvec_vrint_rm_s helper_gvec_vrint_rm_s_x86_64
 #define helper_gvec_vrintx_h helper_gvec_vrintx_h_x86_64
 #define helper_gvec_vrintx_s helper_gvec_vrintx_s_x86_64
+#define helper_gvec_fmla_nf_idx_h helper_gvec_fmla_nf_idx_h_x86_64
+#define helper_gvec_fmla_nf_idx_s helper_gvec_fmla_nf_idx_s_x86_64
+#define helper_gvec_fmls_nf_idx_h helper_gvec_fmls_nf_idx_h_x86_64
+#define helper_gvec_fmls_nf_idx_s helper_gvec_fmls_nf_idx_s_x86_64
 #define helper_power_down helper_power_down_x86_64
 #define helper_pre_hvc helper_pre_hvc_x86_64
 #define helper_pre_smc helper_pre_smc_x86_64