RISCV Add FPU context save (#1250)

* port: riscv: Split the number of registers and the size of the context

* port: riscv: Create some macros for the FPU context

* port: riscv: Add a couple of macros that store fpu context

* port: riscv: Update the stack init function to include the fpu context size

* port: riscv: Add a chip_specific_extensions file that includes the F extension

* Update dictionary to include some risc-v instructions

* port: riscv: Fix a few typos

* port: riscv: Apply @aggarg's sugestions
diff --git a/.github/.cSpellWords.txt b/.github/.cSpellWords.txt
index 3582741..27fb0fc 100644
--- a/.github/.cSpellWords.txt
+++ b/.github/.cSpellWords.txt
@@ -23,6 +23,7 @@
 ALMIEN
 ALMV
 ANDC
+andi
 ANDCCR
 APIC
 APROCFREQ
@@ -47,6 +48,7 @@
 BCPC
 beevt
 BEEVT
+beqz
 BERR
 bfextu
 Biagioni
@@ -298,6 +300,7 @@
 FCMD
 fcolor
 FCSE
+fcsr
 fdiagnostics
 fdiv
 FDIV
diff --git a/portable/GCC/RISC-V/chip_specific_extensions/Pulpino_Vega_RV32M1RM/freertos_risc_v_chip_specific_extensions.h b/portable/GCC/RISC-V/chip_specific_extensions/Pulpino_Vega_RV32M1RM/freertos_risc_v_chip_specific_extensions.h
index 4d8a5fb..c10869e 100644
--- a/portable/GCC/RISC-V/chip_specific_extensions/Pulpino_Vega_RV32M1RM/freertos_risc_v_chip_specific_extensions.h
+++ b/portable/GCC/RISC-V/chip_specific_extensions/Pulpino_Vega_RV32M1RM/freertos_risc_v_chip_specific_extensions.h
@@ -80,22 +80,22 @@
 csrr t3, lpstart1
 csrr t4, lpend1
 csrr t5, lpcount1
-sw t0, 1 * portWORD_SIZE( sp )
-sw t1, 2 * portWORD_SIZE( sp )
-sw t2, 3 * portWORD_SIZE( sp )
-sw t3, 4 * portWORD_SIZE( sp )
-sw t4, 5 * portWORD_SIZE( sp )
-sw t5, 6 * portWORD_SIZE( sp )
+sw t0, 2 * portWORD_SIZE( sp )
+sw t1, 3 * portWORD_SIZE( sp )
+sw t2, 4 * portWORD_SIZE( sp )
+sw t3, 5 * portWORD_SIZE( sp )
+sw t4, 6 * portWORD_SIZE( sp )
+sw t5, 7 * portWORD_SIZE( sp )
    .endm
 
 /* Restore the additional registers found on the Pulpino. */
    .macro portasmRESTORE_ADDITIONAL_REGISTERS
-lw t0, 1 * portWORD_SIZE( sp ) /* Load additional registers into accessible temporary registers. */
-lw t1, 2 * portWORD_SIZE( sp )
-lw t2, 3 * portWORD_SIZE( sp )
-lw t3, 4 * portWORD_SIZE( sp )
-lw t4, 5 * portWORD_SIZE( sp )
-lw t5, 6 * portWORD_SIZE( sp )
+lw t0, 2 * portWORD_SIZE( sp ) /* Load additional registers into accessible temporary registers. */
+lw t1, 3 * portWORD_SIZE( sp )
+lw t2, 4 * portWORD_SIZE( sp )
+lw t3, 5 * portWORD_SIZE( sp )
+lw t4, 6 * portWORD_SIZE( sp )
+lw t5, 7 * portWORD_SIZE( sp )
 csrw lpstart0, t0
 csrw lpend0, t1
 csrw lpcount0, t2
diff --git a/portable/GCC/RISC-V/portASM.S b/portable/GCC/RISC-V/portASM.S
index 99613ad..1fe50f6 100644
--- a/portable/GCC/RISC-V/portASM.S
+++ b/portable/GCC/RISC-V/portASM.S
@@ -162,7 +162,6 @@
  * where the global and thread pointers are currently assumed to be constant so
  * are not saved:
  *
- * mstatus
  * xCriticalNesting
  * x31
  * x30
@@ -192,18 +191,12 @@
  * x6
  * x5
  * portTASK_RETURN_ADDRESS
+ * [FPU registers (when enabled/available) go here]
  * [chip specific registers go here]
+ * mstatus
  * pxCode
  */
 pxPortInitialiseStack:
-    csrr t0, mstatus                    /* Obtain current mstatus value. */
-    andi t0, t0, ~0x8                   /* Ensure interrupts are disabled when the stack is restored within an ISR.  Required when a task is created after the schedulre has been started, otherwise interrupts would be disabled anyway. */
-    addi t1, x0, 0x188                  /* Generate the value 0x1880, which are the MPIE and MPP bits to set in mstatus. */
-    slli t1, t1, 4
-    or t0, t0, t1                       /* Set MPIE and MPP bits in mstatus value. */
-
-    addi a0, a0, -portWORD_SIZE
-    store_x t0, 0(a0)                   /* mstatus onto the stack. */
     addi a0, a0, -portWORD_SIZE         /* Space for critical nesting count. */
     store_x x0, 0(a0)                   /* Critical nesting count starts at 0 for every task. */
 
@@ -212,10 +205,12 @@
 #else
     addi a0, a0, -(22 * portWORD_SIZE)  /* Space for registers x10-x31. */
 #endif
-    store_x a2, 0(a0)                   /* Task parameters (pvParameters parameter) goes into register X10/a0 on the stack. */
-    addi a0, a0, -(6 * portWORD_SIZE)   /* Space for registers x5-x9 + taskReturnAddress. */
+    store_x a2, 0(a0)                   /* Task parameters (pvParameters parameter) goes into register x10/a0 on the stack. */
+
+    addi a0, a0, -(6 * portWORD_SIZE)   /* Space for registers x5-x9 + taskReturnAddress (register x1). */
     load_x t0, xTaskReturnAddress
     store_x t0, 0(a0)                   /* Return address onto the stack. */
+
     addi t0, x0, portasmADDITIONAL_CONTEXT_SIZE /* The number of chip specific additional registers. */
 chip_specific_stack_frame:              /* First add any chip specific registers to the stack frame being created. */
     beq t0, x0, 1f                      /* No more chip specific registers to save. */
@@ -224,6 +219,23 @@
     addi t0, t0, -1                     /* Decrement the count of chip specific registers remaining. */
     j chip_specific_stack_frame         /* Until no more chip specific registers. */
 1:
+    csrr t0, mstatus                    /* Obtain current mstatus value. */
+    andi t0, t0, ~0x8                   /* Ensure interrupts are disabled when the stack is restored within an ISR.  Required when a task is created after the scheduler has been started, otherwise interrupts would be disabled anyway. */
+    addi t1, x0, 0x188                  /* Generate the value 0x1880, which are the MPIE=1 and MPP=M_Mode in mstatus. */
+    slli t1, t1, 4
+    or t0, t0, t1                       /* Set MPIE and MPP bits in mstatus value. */
+
+#if( configENABLE_FPU == 1 )
+    /* Mark the FPU as clean in the mstatus value. */
+    li t1, ~MSTATUS_FS_MASK
+    and t0, t0, t1
+    li t1, MSTATUS_FS_CLEAN
+    or t0, t0, t1
+#endif
+
+    addi a0, a0, -portWORD_SIZE
+    store_x t0, 0(a0)                   /* mstatus onto the stack. */
+
     addi a0, a0, -portWORD_SIZE
     store_x a1, 0(a0)                   /* mret value (pxCode parameter) onto the stack. */
     ret
@@ -235,46 +247,46 @@
 
     load_x  x1, 0( sp ) /* Note for starting the scheduler the exception return address is used as the function return address. */
 
+    load_x  x5, 1 * portWORD_SIZE( sp ) /* Initial mstatus into x5 (t0). */
+    addi    x5, x5, 0x08                /* Set MIE bit so the first task starts with interrupts enabled - required as returns with ret not eret. */
+    csrw    mstatus, x5                 /* Interrupts enabled from here! */
+
     portasmRESTORE_ADDITIONAL_REGISTERS /* Defined in freertos_risc_v_chip_specific_extensions.h to restore any registers unique to the RISC-V implementation. */
 
-    load_x  x7, 4 * portWORD_SIZE( sp )     /* t2 */
-    load_x  x8, 5 * portWORD_SIZE( sp )     /* s0/fp */
-    load_x  x9, 6 * portWORD_SIZE( sp )     /* s1 */
-    load_x  x10, 7 * portWORD_SIZE( sp )    /* a0 */
-    load_x  x11, 8 * portWORD_SIZE( sp )    /* a1 */
-    load_x  x12, 9 * portWORD_SIZE( sp )    /* a2 */
-    load_x  x13, 10 * portWORD_SIZE( sp )   /* a3 */
-    load_x  x14, 11 * portWORD_SIZE( sp )   /* a4 */
-    load_x  x15, 12 * portWORD_SIZE( sp )   /* a5 */
+    load_x  x7,  5  * portWORD_SIZE( sp )   /* t2 */
+    load_x  x8,  6  * portWORD_SIZE( sp )   /* s0/fp */
+    load_x  x9,  7  * portWORD_SIZE( sp )   /* s1 */
+    load_x  x10, 8  * portWORD_SIZE( sp )   /* a0 */
+    load_x  x11, 9  * portWORD_SIZE( sp )   /* a1 */
+    load_x  x12, 10 * portWORD_SIZE( sp )   /* a2 */
+    load_x  x13, 11 * portWORD_SIZE( sp )   /* a3 */
+    load_x  x14, 12 * portWORD_SIZE( sp )   /* a4 */
+    load_x  x15, 13 * portWORD_SIZE( sp )   /* a5 */
 #ifndef __riscv_32e
-    load_x  x16, 13 * portWORD_SIZE( sp )   /* a6 */
-    load_x  x17, 14 * portWORD_SIZE( sp )   /* a7 */
-    load_x  x18, 15 * portWORD_SIZE( sp )   /* s2 */
-    load_x  x19, 16 * portWORD_SIZE( sp )   /* s3 */
-    load_x  x20, 17 * portWORD_SIZE( sp )   /* s4 */
-    load_x  x21, 18 * portWORD_SIZE( sp )   /* s5 */
-    load_x  x22, 19 * portWORD_SIZE( sp )   /* s6 */
-    load_x  x23, 20 * portWORD_SIZE( sp )   /* s7 */
-    load_x  x24, 21 * portWORD_SIZE( sp )   /* s8 */
-    load_x  x25, 22 * portWORD_SIZE( sp )   /* s9 */
-    load_x  x26, 23 * portWORD_SIZE( sp )   /* s10 */
-    load_x  x27, 24 * portWORD_SIZE( sp )   /* s11 */
-    load_x  x28, 25 * portWORD_SIZE( sp )   /* t3 */
-    load_x  x29, 26 * portWORD_SIZE( sp )   /* t4 */
-    load_x  x30, 27 * portWORD_SIZE( sp )   /* t5 */
-    load_x  x31, 28 * portWORD_SIZE( sp )   /* t6 */
+    load_x  x16, 14 * portWORD_SIZE( sp )   /* a6 */
+    load_x  x17, 15 * portWORD_SIZE( sp )   /* a7 */
+    load_x  x18, 16 * portWORD_SIZE( sp )   /* s2 */
+    load_x  x19, 17 * portWORD_SIZE( sp )   /* s3 */
+    load_x  x20, 18 * portWORD_SIZE( sp )   /* s4 */
+    load_x  x21, 19 * portWORD_SIZE( sp )   /* s5 */
+    load_x  x22, 20 * portWORD_SIZE( sp )   /* s6 */
+    load_x  x23, 21 * portWORD_SIZE( sp )   /* s7 */
+    load_x  x24, 22 * portWORD_SIZE( sp )   /* s8 */
+    load_x  x25, 23 * portWORD_SIZE( sp )   /* s9 */
+    load_x  x26, 24 * portWORD_SIZE( sp )   /* s10 */
+    load_x  x27, 25 * portWORD_SIZE( sp )   /* s11 */
+    load_x  x28, 26 * portWORD_SIZE( sp )   /* t3 */
+    load_x  x29, 27 * portWORD_SIZE( sp )   /* t4 */
+    load_x  x30, 28 * portWORD_SIZE( sp )   /* t5 */
+    load_x  x31, 29 * portWORD_SIZE( sp )   /* t6 */
 #endif
 
     load_x  x5, portCRITICAL_NESTING_OFFSET * portWORD_SIZE( sp )    /* Obtain xCriticalNesting value for this task from task's stack. */
     load_x  x6, pxCriticalNesting           /* Load the address of xCriticalNesting into x6. */
     store_x x5, 0( x6 )                     /* Restore the critical nesting value for this task. */
 
-    load_x  x5, portMSTATUS_OFFSET * portWORD_SIZE( sp )    /* Initial mstatus into x5 (t0). */
-    addi    x5, x5, 0x08                    /* Set MIE bit so the first task starts with interrupts enabled - required as returns with ret not eret. */
-    csrrw   x0, mstatus, x5                 /* Interrupts enabled from here! */
-
-    load_x  x5, 2 * portWORD_SIZE( sp )     /* Initial x5 (t0) value. */
-    load_x  x6, 3 * portWORD_SIZE( sp )     /* Initial x6 (t1) value. */
+    load_x  x5, 3 * portWORD_SIZE( sp )     /* Initial x5 (t0) value. */
+    load_x  x6, 4 * portWORD_SIZE( sp )     /* Initial x6 (t1) value. */
 
     addi    sp, sp, portCONTEXT_SIZE
     ret
diff --git a/portable/GCC/RISC-V/portContext.h b/portable/GCC/RISC-V/portContext.h
index 6baae75..9afe6ee 100644
--- a/portable/GCC/RISC-V/portContext.h
+++ b/portable/GCC/RISC-V/portContext.h
@@ -29,6 +29,10 @@
 #ifndef PORTCONTEXT_H
 #define PORTCONTEXT_H
 
+#ifndef configENABLE_FPU
+    #define configENABLE_FPU 0
+#endif
+
 #if __riscv_xlen == 64
     #define portWORD_SIZE    8
     #define store_x          sd
@@ -50,65 +54,199 @@
  * notes at the top of portASM.S file. */
 #ifdef __riscv_32e
     #define portCONTEXT_SIZE               ( 15 * portWORD_SIZE )
-    #define portCRITICAL_NESTING_OFFSET    13
-    #define portMSTATUS_OFFSET             14
+    #define portCRITICAL_NESTING_OFFSET    14
 #else
     #define portCONTEXT_SIZE               ( 31 * portWORD_SIZE )
-    #define portCRITICAL_NESTING_OFFSET    29
-    #define portMSTATUS_OFFSET             30
+    #define portCRITICAL_NESTING_OFFSET    30
 #endif
 
+#if ( configENABLE_FPU == 1 )
+    /* Bit [14:13] in the mstatus encode the status of FPU state which is one of
+     * the following values:
+     * 1. Value: 0, Meaning: Off.
+     * 2. Value: 1, Meaning: Initial.
+     * 3. Value: 2, Meaning: Clean.
+     * 4. Value: 3, Meaning: Dirty.
+     */
+    #define MSTATUS_FS_MASK                 0x6000
+    #define MSTATUS_FS_INITIAL              0x2000
+    #define MSTATUS_FS_CLEAN                0x4000
+    #define MSTATUS_FS_DIRTY                0x6000
+    #define MSTATUS_FS_OFFSET               13
+
+    #ifdef __riscv_fdiv
+        #if __riscv_flen == 32
+            #define load_f                  flw
+            #define store_f                 fsw
+        #elif __riscv_flen == 64
+            #define load_f                  fld
+            #define store_f                 fsd
+        #else
+            #error Assembler did not define __riscv_flen
+        #endif
+
+        #define portFPU_REG_SIZE                ( __riscv_flen / 8 )
+        #define portFPU_REG_COUNT               33 /* 32 Floating point registers plus one CSR. */
+        #define portFPU_CONTEXT_SIZE            ( portFPU_REG_SIZE * portFPU_REG_COUNT )
+    #else
+        #error configENABLE_FPU must not be set to 1 if the hardwar does not have FPU
+    #endif
+#endif
 /*-----------------------------------------------------------*/
 
 .extern pxCurrentTCB
-   .extern xISRStackTop
-   .extern xCriticalNesting
-   .extern pxCriticalNesting
+.extern xISRStackTop
+.extern xCriticalNesting
+.extern pxCriticalNesting
+/*-----------------------------------------------------------*/
+
+    .macro portcontexSAVE_FPU_CONTEXT
+addi sp, sp, -( portFPU_CONTEXT_SIZE )
+/* Store the FPU registers. */
+store_f f0,  2  * portWORD_SIZE( sp )
+store_f f1,  3  * portWORD_SIZE( sp )
+store_f f2,  4  * portWORD_SIZE( sp )
+store_f f3,  5  * portWORD_SIZE( sp )
+store_f f4,  6  * portWORD_SIZE( sp )
+store_f f5,  7  * portWORD_SIZE( sp )
+store_f f6,  8  * portWORD_SIZE( sp )
+store_f f7,  9  * portWORD_SIZE( sp )
+store_f f8,  10 * portWORD_SIZE( sp )
+store_f f9,  11 * portWORD_SIZE( sp )
+store_f f10, 12 * portWORD_SIZE( sp )
+store_f f11, 13 * portWORD_SIZE( sp )
+store_f f12, 14 * portWORD_SIZE( sp )
+store_f f13, 15 * portWORD_SIZE( sp )
+store_f f14, 16 * portWORD_SIZE( sp )
+store_f f15, 17 * portWORD_SIZE( sp )
+store_f f16, 18 * portWORD_SIZE( sp )
+store_f f17, 19 * portWORD_SIZE( sp )
+store_f f18, 20 * portWORD_SIZE( sp )
+store_f f19, 21 * portWORD_SIZE( sp )
+store_f f20, 22 * portWORD_SIZE( sp )
+store_f f21, 23 * portWORD_SIZE( sp )
+store_f f22, 24 * portWORD_SIZE( sp )
+store_f f23, 25 * portWORD_SIZE( sp )
+store_f f24, 26 * portWORD_SIZE( sp )
+store_f f25, 27 * portWORD_SIZE( sp )
+store_f f26, 28 * portWORD_SIZE( sp )
+store_f f27, 29 * portWORD_SIZE( sp )
+store_f f28, 30 * portWORD_SIZE( sp )
+store_f f29, 31 * portWORD_SIZE( sp )
+store_f f30, 32 * portWORD_SIZE( sp )
+store_f f31, 33 * portWORD_SIZE( sp )
+csrr t0, fcsr
+store_x t0,  34 * portWORD_SIZE( sp )
+    .endm
+/*-----------------------------------------------------------*/
+
+    .macro portcontextRESTORE_FPU_CONTEXT
+/* Restore the FPU registers. */
+load_f f0,  2  * portWORD_SIZE( sp )
+load_f f1,  3  * portWORD_SIZE( sp )
+load_f f2,  4  * portWORD_SIZE( sp )
+load_f f3,  5  * portWORD_SIZE( sp )
+load_f f4,  6  * portWORD_SIZE( sp )
+load_f f5,  7  * portWORD_SIZE( sp )
+load_f f6,  8  * portWORD_SIZE( sp )
+load_f f7,  9  * portWORD_SIZE( sp )
+load_f f8,  10 * portWORD_SIZE( sp )
+load_f f9,  11 * portWORD_SIZE( sp )
+load_f f10, 12 * portWORD_SIZE( sp )
+load_f f11, 13 * portWORD_SIZE( sp )
+load_f f12, 14 * portWORD_SIZE( sp )
+load_f f13, 15 * portWORD_SIZE( sp )
+load_f f14, 16 * portWORD_SIZE( sp )
+load_f f15, 17 * portWORD_SIZE( sp )
+load_f f16, 18 * portWORD_SIZE( sp )
+load_f f17, 19 * portWORD_SIZE( sp )
+load_f f18, 20 * portWORD_SIZE( sp )
+load_f f19, 21 * portWORD_SIZE( sp )
+load_f f20, 22 * portWORD_SIZE( sp )
+load_f f21, 23 * portWORD_SIZE( sp )
+load_f f22, 24 * portWORD_SIZE( sp )
+load_f f23, 25 * portWORD_SIZE( sp )
+load_f f24, 26 * portWORD_SIZE( sp )
+load_f f25, 27 * portWORD_SIZE( sp )
+load_f f26, 28 * portWORD_SIZE( sp )
+load_f f27, 29 * portWORD_SIZE( sp )
+load_f f28, 30 * portWORD_SIZE( sp )
+load_f f29, 31 * portWORD_SIZE( sp )
+load_f f30, 32 * portWORD_SIZE( sp )
+load_f f31, 33 * portWORD_SIZE( sp )
+load_x t0,  34 * portWORD_SIZE( sp )
+csrw fcsr, t0
+addi sp, sp, ( portFPU_CONTEXT_SIZE )
+    .endm
 /*-----------------------------------------------------------*/
 
    .macro portcontextSAVE_CONTEXT_INTERNAL
 addi sp, sp, -portCONTEXT_SIZE
-store_x x1, 1 * portWORD_SIZE( sp )
-store_x x5, 2 * portWORD_SIZE( sp )
-store_x x6, 3 * portWORD_SIZE( sp )
-store_x x7, 4 * portWORD_SIZE( sp )
-store_x x8, 5 * portWORD_SIZE( sp )
-store_x x9, 6 * portWORD_SIZE( sp )
-store_x x10, 7 * portWORD_SIZE( sp )
-store_x x11, 8 * portWORD_SIZE( sp )
-store_x x12, 9 * portWORD_SIZE( sp )
-store_x x13, 10 * portWORD_SIZE( sp )
-store_x x14, 11 * portWORD_SIZE( sp )
-store_x x15, 12 * portWORD_SIZE( sp )
+store_x x1,  2  * portWORD_SIZE( sp )
+store_x x5,  3  * portWORD_SIZE( sp )
+store_x x6,  4  * portWORD_SIZE( sp )
+store_x x7,  5  * portWORD_SIZE( sp )
+store_x x8,  6  * portWORD_SIZE( sp )
+store_x x9,  7  * portWORD_SIZE( sp )
+store_x x10, 8  * portWORD_SIZE( sp )
+store_x x11, 9  * portWORD_SIZE( sp )
+store_x x12, 10 * portWORD_SIZE( sp )
+store_x x13, 11 * portWORD_SIZE( sp )
+store_x x14, 12 * portWORD_SIZE( sp )
+store_x x15, 13 * portWORD_SIZE( sp )
 #ifndef __riscv_32e
-    store_x x16, 13 * portWORD_SIZE( sp )
-    store_x x17, 14 * portWORD_SIZE( sp )
-    store_x x18, 15 * portWORD_SIZE( sp )
-    store_x x19, 16 * portWORD_SIZE( sp )
-    store_x x20, 17 * portWORD_SIZE( sp )
-    store_x x21, 18 * portWORD_SIZE( sp )
-    store_x x22, 19 * portWORD_SIZE( sp )
-    store_x x23, 20 * portWORD_SIZE( sp )
-    store_x x24, 21 * portWORD_SIZE( sp )
-    store_x x25, 22 * portWORD_SIZE( sp )
-    store_x x26, 23 * portWORD_SIZE( sp )
-    store_x x27, 24 * portWORD_SIZE( sp )
-    store_x x28, 25 * portWORD_SIZE( sp )
-    store_x x29, 26 * portWORD_SIZE( sp )
-    store_x x30, 27 * portWORD_SIZE( sp )
-    store_x x31, 28 * portWORD_SIZE( sp )
+    store_x x16, 14 * portWORD_SIZE( sp )
+    store_x x17, 15 * portWORD_SIZE( sp )
+    store_x x18, 16 * portWORD_SIZE( sp )
+    store_x x19, 17 * portWORD_SIZE( sp )
+    store_x x20, 18 * portWORD_SIZE( sp )
+    store_x x21, 19 * portWORD_SIZE( sp )
+    store_x x22, 20 * portWORD_SIZE( sp )
+    store_x x23, 21 * portWORD_SIZE( sp )
+    store_x x24, 22 * portWORD_SIZE( sp )
+    store_x x25, 23 * portWORD_SIZE( sp )
+    store_x x26, 24 * portWORD_SIZE( sp )
+    store_x x27, 25 * portWORD_SIZE( sp )
+    store_x x28, 26 * portWORD_SIZE( sp )
+    store_x x29, 27 * portWORD_SIZE( sp )
+    store_x x30, 28 * portWORD_SIZE( sp )
+    store_x x31, 29 * portWORD_SIZE( sp )
 #endif /* ifndef __riscv_32e */
 
 load_x t0, xCriticalNesting                                   /* Load the value of xCriticalNesting into t0. */
 store_x t0, portCRITICAL_NESTING_OFFSET * portWORD_SIZE( sp ) /* Store the critical nesting value to the stack. */
 
+#if( configENABLE_FPU == 1 )
+    csrr t0, mstatus
+    srl t1, t0, MSTATUS_FS_OFFSET
+    andi t1, t1, 3
+    addi t2, x0, 3
+    bne t1, t2, 1f /* If FPU status is not dirty, do not save FPU registers. */
 
-csrr t0, mstatus /* Required for MPIE bit. */
-store_x t0, portMSTATUS_OFFSET * portWORD_SIZE( sp )
-
+    portcontexSAVE_FPU_CONTEXT
+1:
+#endif
 
 portasmSAVE_ADDITIONAL_REGISTERS /* Defined in freertos_risc_v_chip_specific_extensions.h to save any registers unique to the RISC-V implementation. */
 
+csrr t0, mstatus
+store_x t0, 1 * portWORD_SIZE( sp )
+
+#if( configENABLE_FPU == 1 )
+    /* Mark the FPU as clean, if it was dirty and we saved FPU registers. */
+    srl t1, t0, MSTATUS_FS_OFFSET
+    andi t1, t1, 3
+    addi t2, x0, 3
+    bne t1, t2, 2f
+
+    li t1, ~MSTATUS_FS_MASK
+    and t0, t0, t1
+    li t1, MSTATUS_FS_CLEAN
+    or t0, t0, t1
+    csrw mstatus, t0
+2:
+#endif
+
 load_x t0, pxCurrentTCB          /* Load pxCurrentTCB. */
 store_x sp, 0 ( t0 )             /* Write sp to first TCB member. */
 
@@ -142,46 +280,57 @@
 load_x t0, 0 ( sp )
 csrw mepc, t0
 
+/* Restore mstatus register. */
+load_x t0, 1 * portWORD_SIZE( sp )
+csrw mstatus, t0
+
 /* Defined in freertos_risc_v_chip_specific_extensions.h to restore any registers unique to the RISC-V implementation. */
 portasmRESTORE_ADDITIONAL_REGISTERS
 
-/* Load mstatus with the interrupt enable bits used by the task. */
-load_x t0, portMSTATUS_OFFSET * portWORD_SIZE( sp )
-csrw mstatus, t0                                             /* Required for MPIE bit. */
+#if( configENABLE_FPU == 1 )
+    csrr t0, mstatus
+    srl t1, t0, MSTATUS_FS_OFFSET
+    andi t1, t1, 3
+    addi t2, x0, 3
+    bne t1, t2, 3f /* If FPU status is not dirty, do not restore FPU registers. */
+
+    portcontextRESTORE_FPU_CONTEXT
+3:
+#endif /* ifdef portasmSTORE_FPU_CONTEXT */
 
 load_x t0, portCRITICAL_NESTING_OFFSET * portWORD_SIZE( sp ) /* Obtain xCriticalNesting value for this task from task's stack. */
 load_x t1, pxCriticalNesting                                 /* Load the address of xCriticalNesting into t1. */
 store_x t0, 0 ( t1 )                                         /* Restore the critical nesting value for this task. */
 
-load_x x1, 1 * portWORD_SIZE( sp )
-load_x x5, 2 * portWORD_SIZE( sp )
-load_x x6, 3 * portWORD_SIZE( sp )
-load_x x7, 4 * portWORD_SIZE( sp )
-load_x x8, 5 * portWORD_SIZE( sp )
-load_x x9, 6 * portWORD_SIZE( sp )
-load_x x10, 7 * portWORD_SIZE( sp )
-load_x x11, 8 * portWORD_SIZE( sp )
-load_x x12, 9 * portWORD_SIZE( sp )
-load_x x13, 10 * portWORD_SIZE( sp )
-load_x x14, 11 * portWORD_SIZE( sp )
-load_x x15, 12 * portWORD_SIZE( sp )
+load_x x1,  2  * portWORD_SIZE( sp )
+load_x x5,  3  * portWORD_SIZE( sp )
+load_x x6,  4  * portWORD_SIZE( sp )
+load_x x7,  5  * portWORD_SIZE( sp )
+load_x x8,  6  * portWORD_SIZE( sp )
+load_x x9,  7  * portWORD_SIZE( sp )
+load_x x10, 8  * portWORD_SIZE( sp )
+load_x x11, 9  * portWORD_SIZE( sp )
+load_x x12, 10 * portWORD_SIZE( sp )
+load_x x13, 11 * portWORD_SIZE( sp )
+load_x x14, 12 * portWORD_SIZE( sp )
+load_x x15, 13 * portWORD_SIZE( sp )
 #ifndef __riscv_32e
-    load_x x16, 13 * portWORD_SIZE( sp )
-    load_x x17, 14 * portWORD_SIZE( sp )
-    load_x x18, 15 * portWORD_SIZE( sp )
-    load_x x19, 16 * portWORD_SIZE( sp )
-    load_x x20, 17 * portWORD_SIZE( sp )
-    load_x x21, 18 * portWORD_SIZE( sp )
-    load_x x22, 19 * portWORD_SIZE( sp )
-    load_x x23, 20 * portWORD_SIZE( sp )
-    load_x x24, 21 * portWORD_SIZE( sp )
-    load_x x25, 22 * portWORD_SIZE( sp )
-    load_x x26, 23 * portWORD_SIZE( sp )
-    load_x x27, 24 * portWORD_SIZE( sp )
-    load_x x28, 25 * portWORD_SIZE( sp )
-    load_x x29, 26 * portWORD_SIZE( sp )
-    load_x x30, 27 * portWORD_SIZE( sp )
-    load_x x31, 28 * portWORD_SIZE( sp )
+    load_x x16, 14 * portWORD_SIZE( sp )
+    load_x x17, 15 * portWORD_SIZE( sp )
+    load_x x18, 16 * portWORD_SIZE( sp )
+    load_x x19, 17 * portWORD_SIZE( sp )
+    load_x x20, 18 * portWORD_SIZE( sp )
+    load_x x21, 19 * portWORD_SIZE( sp )
+    load_x x22, 20 * portWORD_SIZE( sp )
+    load_x x23, 21 * portWORD_SIZE( sp )
+    load_x x24, 22 * portWORD_SIZE( sp )
+    load_x x25, 23 * portWORD_SIZE( sp )
+    load_x x26, 24 * portWORD_SIZE( sp )
+    load_x x27, 25 * portWORD_SIZE( sp )
+    load_x x28, 26 * portWORD_SIZE( sp )
+    load_x x29, 27 * portWORD_SIZE( sp )
+    load_x x30, 28 * portWORD_SIZE( sp )
+    load_x x31, 29 * portWORD_SIZE( sp )
 #endif /* ifndef __riscv_32e */
 addi sp, sp, portCONTEXT_SIZE