[rtl] instruction prefetch buffer (IPB) improvements (#455)

CHANGELOG.md

@@ -32,6 +32,7 @@ mimpid = 0x01040312 => Version 01.04.03.12 => v1.4.3.12
 
 | Date (*dd.mm.yyyy*) | Version | Comment |
 |:-------------------:|:-------:|:--------|
+| 13.12.2022 | 1.7.8.5 | code cleanup of FIFO module; improved **instruction prefetch buffer (IPB)** - IPD depth can be as small as "1" and will be adjusted automatically when enabling the `C` ISA extension; update hardware implementation results; [#455](https://github.com/stnolting/neorv32/pull/455) |
 | 09.12.2022 | 1.7.8.4 | :sparkles: new option to add custom **R5-type** (4 source registers, 1 destination register) instructions to **Custom Functions Unit (CFU)**; [#452](https://github.com/stnolting/neorv32/pull/452) |
 | 08.12.2022 | 1.7.8.3 | :bug: fix interrupt behavior when in user-mode; minor core rtl fixes; do not check registers specifiers in CFU instructions (i.e. using registers above `x15` when `E` ISA extension is enabled); [#450](https://github.com/stnolting/neorv32/pull/450) |
 | 03.12.2022 | 1.7.8.2 | :sparkles: new option to add custom **R4-type** RISC-V instructions to **Custom Functions Unit (CFU)**; rework CFU hardware module, intrinsic library and example program; [#449](https://github.com/stnolting/neorv32/pull/449) |

README.md

@@ -200,10 +200,10 @@ for custom tightly-coupled co-processors, accelerators or interfaces
 Implementation results for **exemplary CPU configurations** generated for an Intel Cyclone IV `EP4CE22F17C6` FPGA
 using Intel Quartus Prime Lite 21.1 (no timing constrains, _balanced optimization_, f_max from _Slow 1200mV 0C Model_).
 
-| CPU Configuration (version [1.7.7.8](https://github.com/stnolting/neorv32/blob/main/CHANGELOG.md)) | LEs | FFs | Memory bits | DSPs | f_max |
+| CPU Configuration (version [1.7.8.5](https://github.com/stnolting/neorv32/blob/main/CHANGELOG.md)) | LEs | FFs | Memory bits | DSPs | f_max |
 |:-----------------------|:----:|:----:|:----:|:-:|:-------:|
-| `rv32i_Zicsr`          | 1328 |  678 | 1024 | 0 | 130 MHz |
-| `rv32i_Zicsr_Zicntr`   | 1614 |  808 | 1024 | 0 | 130 MHz |
+| `rv32i_Zicsr`          | 1223 |  607 | 1024 | 0 | 130 MHz |
+| `rv32i_Zicsr_Zicntr`   | 1578 |  773 | 1024 | 0 | 130 MHz |
 | `rv32imc_Zicsr_Zicntr` | 2338 |  992 | 1024 | 0 | 130 MHz |
 
 Implementation results for an **exemplary SoC/Processor configurations** generated for a Xilinx Artix-7 `xc7a35ticsg324-1L` FPGA

docs/datasheet/overview.adoc

@@ -250,7 +250,7 @@ just _exemplary_. If not otherwise mentioned all implementations use the default
 [cols="<2,<8"]
 [grid="topbot"]
 |=======================
-| HW version:  | `1.7.7.8`
+| HW version:  | `1.7.8.5`
 | Top entity:  | `rtl/core/neorv32_cpu.vhd`
 | FPGA:        | Intel Cyclone IV E `EP4CE22F17C6`
 | Toolchain:   | Quartus Prime Lite 21.1
@@ -261,10 +261,10 @@ just _exemplary_. If not otherwise mentioned all implementations use the default
 [options="header",grid="rows"]
 |=======================
 | CPU ISA Configuration                             | LEs  | FFs  | MEM bits | DSPs | _f~max~_
-| `rv32e`                                           |  830 |  400 |      512 |    0 | 130 MHz
-| `rv32i`                                           |  834 |  400 |     1024 |    0 | 130 MHz
-| `rv32i_Zicsr`                                     | 1328 |  678 |     1024 |    0 | 130 MHz
-| `rv32i_Zicsr_Zicntr`                              | 1614 |  808 |     1024 |    0 | 130 MHz
+| `rv32e`                                           |  720 |  360 |      512 |    0 | 130 MHz
+| `rv32i`                                           |  724 |  364 |     1024 |    0 | 130 MHz
+| `rv32i_Zicsr`                                     | 1223 |  607 |     1024 |    0 | 130 MHz
+| `rv32i_Zicsr_Zicntr`                              | 1578 |  773 |     1024 |    0 | 130 MHz
 | `rv32im_Zicsr_Zicntr`                             | 2087 |  983 |     1024 |    0 | 130 MHz
 | `rv32imc_Zicsr_Zicntr`                            | 2338 |  992 |     1024 |    0 | 130 MHz
 | `rv32imcb_Zicsr_Zicntr`                           | 3175 | 1247 |     1024 |    0 | 130 MHz

docs/datasheet/soc.adoc

@@ -454,12 +454,18 @@ The state of this generic can be retrieved by software via the <<_mxisa>> CSR.
 [cols="4,4,2"]
 [frame="all",grid="none"]
 |======
-| **CPU_IPB_ENTRIES** | _natural_ | 2
+| **CPU_IPB_ENTRIES** | _natural_ | 1
 3+| This generic configures the number of entries in the CPU's instruction prefetch buffer.
-The value has to be a power of two and has to be greater than or equal to two (>= 2).
-Long linear sequences of code can benefit from an increased IPB size.
+The value has to be a power of two and has to be greater than or equal to one (>= 1). The
+IPB can help improving memory access latency. Furthermore, long linear code sequences will
+benefit from an increased IPB size.
 |======
 
+[WARNING]
+If the compressed ISA extension `_CPU_EXTENSION_RISCV_C_` (<<_cpu_extension_riscv_c>>) is enabled and the IPB depth
+is set to 1, this configuration is internally overridden and the IPB will be implemented with **2** entries. This is required
+for handling unaligned 32-bit instructions.
+
 
 // ####################################################################################################################
 :sectnums:

docs/userguide/application_specific_configuration.adoc

@@ -24,8 +24,7 @@ multiplications, `FAST_SHIFT_EN => true` use a fast barrel shifter for shift ope
 * Use as many _internal_ memory as possible to reduce memory access latency: `MEM_INT_IMEM_EN => true` and
 `MEM_INT_DMEM_EN => true`, maximize `MEM_INT_IMEM_SIZE` and `MEM_INT_DMEM_SIZE`
 * Increase the CPU's instruction prefetch buffer size: if **no** instruction cache is implemented `CPU_IPB_ENTRIES` should be
-quite large (recommended value is >= 8); if the instruction cache is implemented `CPU_IPB_ENTRIES` values above 4 are
-rather inefficient
+quite large
 * _To be continued..._
 
 
@@ -55,7 +54,7 @@ also reduces program code size by approximately 30%.
 * If not explicitly used/required, exclude the CPU standard counters `[m]instret[h]`
 (number of instruction) and `[m]cycle[h]` (number of cycles) from synthesis by disabling the `Zicntr` ISA extension
 (note, this is not RISC-V compliant).
-* Reduce the CPU's prefetch buffer size (`CPU_IPB_ENTRIES`).
+* Reduce the CPU's prefetch buffer size (`CPU_IPB_ENTRIES`) to its minimum (=1).
 * Map CPU shift operations to a small and iterative shifter unit (`FAST_SHIFT_EN => false`).
 * If you have unused DSP block available, you can map multiplication operations to those slices instead of
 using LUTs to implement the multiplier (`FAST_MUL_EN => true`).

rtl/core/neorv32_cpu.vhd

@@ -67,7 +67,7 @@ entity neorv32_cpu is
     -- Extension Options --
     FAST_MUL_EN                  : boolean; -- use DSPs for M extension's multiplier
     FAST_SHIFT_EN                : boolean; -- use barrel shifter for shift operations
-    CPU_IPB_ENTRIES              : natural; -- entries in instruction prefetch buffer, has to be a power of 2, min 2
+    CPU_IPB_ENTRIES              : natural; -- entries in instruction prefetch buffer, has to be a power of 2, min 1
     -- Physical Memory Protection (PMP) --
     PMP_NUM_REGIONS              : natural; -- number of regions (0..16)
     PMP_MIN_GRANULARITY          : natural; -- minimal region granularity in bytes, has to be a power of 2, min 4 bytes
@@ -120,10 +120,14 @@ architecture neorv32_cpu_rtl of neorv32_cpu is
   constant XLEN : natural := 32; -- data path width
   -- ----------------------------------------------------------------------------------------------
 
-  -- local constants --
+  -- local constants: additional register file read ports --
   constant regfile_rs3_en_c : boolean := CPU_EXTENSION_RISCV_Zxcfu or CPU_EXTENSION_RISCV_Zfinx; -- 3rd register file read port (rs3)
   constant regfile_rs4_en_c : boolean := CPU_EXTENSION_RISCV_Zxcfu; -- 4th register file read port (rs4)
 
+  -- local constant: instruction prefetch buffer depth --
+  constant ipb_override_c : boolean := (CPU_EXTENSION_RISCV_C = true) and (CPU_IPB_ENTRIES < 2); -- override IPB size: set to 2?
+  constant ipb_depth_c    : natural := cond_sel_natural_f(ipb_override_c, 2, CPU_IPB_ENTRIES);
+
   -- local signals --
   signal ctrl        : std_ulogic_vector(ctrl_width_c-1 downto 0); -- main control bus
   signal imm         : std_ulogic_vector(XLEN-1 downto 0); -- immediate
@@ -206,8 +210,8 @@ begin
   -- Instruction prefetch buffer --
   assert not (is_power_of_two_f(CPU_IPB_ENTRIES) = false) report
     "NEORV32 CPU CONFIG ERROR! Number of entries in instruction prefetch buffer <CPU_IPB_ENTRIES> has to be a power of two." severity error;
-  assert not (CPU_IPB_ENTRIES < 2) report
-    "NEORV32 CPU CONFIG ERROR! Number of entries in instruction prefetch buffer <CPU_IPB_ENTRIES> has to be >= 2." severity error;
+  assert not (ipb_override_c = true) report
+    "NEORV32 CPU CONFIG WARNING! Overriding <CPU_IPB_ENTRIES> configuration (setting =2) because C ISA extension is enabled." severity warning;
 
   -- PMP --
   assert not (PMP_NUM_REGIONS > 0) report
@@ -276,7 +280,7 @@ begin
     -- Tuning Options --
     FAST_MUL_EN                  => FAST_MUL_EN,                  -- use DSPs for M extension's multiplier
     FAST_SHIFT_EN                => FAST_SHIFT_EN,                -- use barrel shifter for shift operations
-    CPU_IPB_ENTRIES              => CPU_IPB_ENTRIES,              -- entries is instruction prefetch buffer, has to be a power of 2, min 2
+    CPU_IPB_ENTRIES              => ipb_depth_c,                  -- entries is instruction prefetch buffer, has to be a power of 2, min 1
     -- Physical memory protection (PMP) --
     PMP_NUM_REGIONS              => PMP_NUM_REGIONS,              -- number of regions (0..16)
     PMP_MIN_GRANULARITY          => PMP_MIN_GRANULARITY,          -- minimal region granularity in bytes, has to be a power of 2, min 4 bytes

rtl/core/neorv32_cpu_control.vhd

@@ -73,7 +73,7 @@ entity neorv32_cpu_control is
     -- Tuning Options --
     FAST_MUL_EN                  : boolean; -- use DSPs for M extension's multiplier
     FAST_SHIFT_EN                : boolean; -- use barrel shifter for shift operations
-    CPU_IPB_ENTRIES              : natural; -- entries in instruction prefetch buffer, has to be a power of 2, min 2
+    CPU_IPB_ENTRIES              : natural; -- entries in instruction prefetch buffer, has to be a power of 2, min 1
     -- Physical memory protection (PMP) --
     PMP_NUM_REGIONS              : natural; -- number of regions (0..16)
     PMP_MIN_GRANULARITY          : natural; -- minimal region granularity in bytes, has to be a power of 2, min 4 bytes

rtl/core/neorv32_fifo.vhd

@@ -140,30 +140,48 @@ begin
   fifo_half_level_simple:
   if (FIFO_DEPTH = 1) generate
     half_o <= fifo.full;
-  end generate;
+  end generate; -- /fifo_half_level_simple
 
   fifo_half_level_complex:
   if (FIFO_DEPTH > 1) generate
     level_diff <= std_ulogic_vector(unsigned(fifo.w_pnt) - unsigned(fifo.r_pnt));
     half_o     <= level_diff(level_diff'left-1) or fifo.full;
-  end generate;
+  end generate; -- /fifo_half_level_complex
 
 
-  -- FIFO Memory ----------------------------------------------------------------------------
+  -- FIFO Memory - Write --------------------------------------------------------------------
   -- -------------------------------------------------------------------------------------------
-  fifo_write: process(clk_i)
-  begin
-    if rising_edge(clk_i) then
-      if (fifo.we = '1') then
-        if (FIFO_DEPTH = 1) then
-          fifo.buf <= wdata_i;
-        else
+  -- "real" FIFO memory (several entries) --
+  fifo_memory:
+  if (FIFO_DEPTH > 1) generate
+    fifo_write: process(clk_i)
+    begin
+      if rising_edge(clk_i) then
+        if (fifo.we = '1') then
           fifo.data(to_integer(unsigned(fifo.w_pnt(fifo.w_pnt'left-1 downto 0)))) <= wdata_i;
         end if;
       end if;
-    end if;
-  end process fifo_write;
+    end process fifo_write;
+    fifo.buf <= (others => '0'); -- unused
+  end generate; -- /fifo_memory
 
+  -- simple register/buffer (single entry) --
+  fifo_buffer:
+  if (FIFO_DEPTH = 1) generate
+    fifo_write: process(clk_i)
+    begin
+      if rising_edge(clk_i) then
+        if (fifo.we = '1') then
+          fifo.buf <= wdata_i;
+        end if;
+      end if;
+    end process fifo_write;
+    fifo.data <= (others => (others => '0')); -- unused
+  end generate; -- /fifo_buffer
+
+
+  -- FIFO Memory - Read ---------------------------------------------------------------------
+  -- -------------------------------------------------------------------------------------------
   -- "asynchronous" read --
   fifo_read_async:
   if (FIFO_RSYNC = false) generate
@@ -175,7 +193,7 @@ begin
         rdata <= fifo.data(to_integer(unsigned(fifo.r_pnt(fifo.r_pnt'left-1 downto 0))));
       end if;
     end process fifo_read;
-  end generate;
+  end generate; -- /fifo_read_async
 
   -- synchronous read --
   fifo_read_sync:
@@ -190,7 +208,7 @@ begin
         end if;
       end if;
     end process fifo_read;
-  end generate;
+  end generate; -- /fifo_read_sync
 
 
   -- Output Gate ----------------------------------------------------------------------------

rtl/core/neorv32_package.vhd

@@ -62,7 +62,7 @@ package neorv32_package is
 
   -- Architecture Constants (do not modify!) ------------------------------------------------
   -- -------------------------------------------------------------------------------------------
-  constant hw_version_c : std_ulogic_vector(31 downto 0) := x"01070804"; -- NEORV32 version - no touchy!
+  constant hw_version_c : std_ulogic_vector(31 downto 0) := x"01070805"; -- NEORV32 version - no touchy!
   constant archid_c     : natural := 19; -- official RISC-V architecture ID - hands off!
 
   -- Check if we're inside the Matrix -------------------------------------------------------
@@ -1007,7 +1007,7 @@ package neorv32_package is
       -- Tuning Options --
       FAST_MUL_EN                  : boolean := false;  -- use DSPs for M extension's multiplier
       FAST_SHIFT_EN                : boolean := false;  -- use barrel shifter for shift operations
-      CPU_IPB_ENTRIES              : natural := 2;      -- entries in instruction prefetch buffer, has to be a power of 2, min 2
+      CPU_IPB_ENTRIES              : natural := 1;      -- entries in instruction prefetch buffer, has to be a power of 2, min 1
       -- Physical Memory Protection (PMP) --
       PMP_NUM_REGIONS              : natural := 0;      -- number of regions (0..16)
       PMP_MIN_GRANULARITY          : natural := 4;      -- minimal region granularity in bytes, has to be a power of 2, min 4 bytes
@@ -1173,7 +1173,7 @@ package neorv32_package is
       -- Tuning Options --
       FAST_MUL_EN                  : boolean; -- use DSPs for M extension's multiplier
       FAST_SHIFT_EN                : boolean; -- use barrel shifter for shift operations
-      CPU_IPB_ENTRIES              : natural; -- entries in instruction prefetch buffer, has to be a power of 2, min 2
+      CPU_IPB_ENTRIES              : natural; -- entries in instruction prefetch buffer, has to be a power of 2, min 1
       -- Physical Memory Protection (PMP) --
       PMP_NUM_REGIONS              : natural; -- number of regions (0..16)
       PMP_MIN_GRANULARITY          : natural; -- minimal region granularity in bytes, has to be a power of 2, min 4 bytes
@@ -1245,7 +1245,7 @@ package neorv32_package is
       -- Extension Options --
       FAST_MUL_EN                  : boolean; -- use DSPs for M extension's multiplier
       FAST_SHIFT_EN                : boolean; -- use barrel shifter for shift operations
-      CPU_IPB_ENTRIES              : natural; -- entries is instruction prefetch buffer, has to be a power of 2, min 2
+      CPU_IPB_ENTRIES              : natural; -- entries is instruction prefetch buffer, has to be a power of 2, min 1
       -- Physical memory protection (PMP) --
       PMP_NUM_REGIONS              : natural; -- number of regions (0..16)
       PMP_MIN_GRANULARITY          : natural; -- minimal region granularity in bytes, has to be a power of 2, min 4 bytes

rtl/core/neorv32_top.vhd

@@ -72,7 +72,7 @@ entity neorv32_top is
     -- Tuning Options --
     FAST_MUL_EN                  : boolean := false;  -- use DSPs for M extension's multiplier
     FAST_SHIFT_EN                : boolean := false;  -- use barrel shifter for shift operations
-    CPU_IPB_ENTRIES              : natural := 2;      -- entries in instruction prefetch buffer, has to be a power of 2, min 2
+    CPU_IPB_ENTRIES              : natural := 1;      -- entries in instruction prefetch buffer, has to be a power of 2, min 1
 
     -- Physical Memory Protection (PMP) --
     PMP_NUM_REGIONS              : natural := 0;      -- number of regions (0..16)
@@ -564,7 +564,7 @@ begin
     -- Extension Options --
     FAST_MUL_EN                  => FAST_MUL_EN,                  -- use DSPs for M extension's multiplier
     FAST_SHIFT_EN                => FAST_SHIFT_EN,                -- use barrel shifter for shift operations
-    CPU_IPB_ENTRIES              => CPU_IPB_ENTRIES,              -- entries is instruction prefetch buffer, has to be a power of 2
+    CPU_IPB_ENTRIES              => CPU_IPB_ENTRIES,              -- entries is instruction prefetch buffer, has to be a power of 1
     -- Physical Memory Protection (PMP) --
     PMP_NUM_REGIONS              => PMP_NUM_REGIONS,              -- number of regions (0..16)
     PMP_MIN_GRANULARITY          => PMP_MIN_GRANULARITY,          -- minimal region granularity in bytes, has to be a power of 2, min 4 bytes

rtl/system_integration/neorv32_ProcessorTop_stdlogic.vhd

@@ -64,6 +64,7 @@ entity neorv32_ProcessorTop_stdlogic is
     -- Extension Options --
     FAST_MUL_EN                  : boolean := false;  -- use DSPs for M extension's multiplier
     FAST_SHIFT_EN                : boolean := false;  -- use barrel shifter for shift operations
+    CPU_IPB_ENTRIES              : natural := 1;      -- entries in instruction prefetch buffer, has to be a power of 2, min 1
     -- Physical Memory Protection (PMP) --
     PMP_NUM_REGIONS              : natural := 0;      -- number of regions (0..16)
     PMP_MIN_GRANULARITY          : natural := 4;      -- minimal region granularity in bytes, has to be a power of 2, min 4 bytes
@@ -311,6 +312,7 @@ begin
     -- Extension Options --
     FAST_MUL_EN                  => FAST_MUL_EN,        -- use DSPs for M extension's multiplier
     FAST_SHIFT_EN                => FAST_SHIFT_EN,      -- use barrel shifter for shift operations
+    CPU_IPB_ENTRIES              => CPU_IPB_ENTRIES,    -- entries in instruction prefetch buffer, has to be a power of 2, min 1
     -- Physical Memory Protection (PMP) --
     PMP_NUM_REGIONS              => PMP_NUM_REGIONS,    -- number of regions (0..16)
     PMP_MIN_GRANULARITY          => PMP_MIN_GRANULARITY, -- minimal region granularity in bytes, has to be a power of 2, min 4 bytes

rtl/system_integration/neorv32_SystemTop_AvalonMM.vhd

@@ -70,7 +70,7 @@ entity neorv32_top_avalonmm is
     -- Extension Options --
     FAST_MUL_EN                  : boolean := false;  -- use DSPs for M extension's multiplier
     FAST_SHIFT_EN                : boolean := false;  -- use barrel shifter for shift operations
-    CPU_IPB_ENTRIES              : natural := 2;      -- entries is instruction prefetch buffer, has to be a power of 2
+    CPU_IPB_ENTRIES              : natural := 1;      -- entries is instruction prefetch buffer, has to be a power of 1, min 1
 
     -- Physical Memory Protection (PMP) --
     PMP_NUM_REGIONS              : natural := 0;      -- number of regions (0..16)

rtl/system_integration/neorv32_SystemTop_axi4lite.vhd

@@ -309,6 +309,7 @@ begin
     -- Extension Options --
     FAST_MUL_EN                  => FAST_MUL_EN,        -- use DSPs for M extension's multiplier
     FAST_SHIFT_EN                => FAST_SHIFT_EN,      -- use barrel shifter for shift operations
+    CPU_IPB_ENTRIES              => 2,                  -- entries is instruction prefetch buffer, has to be a power of 2, min 1
     -- Physical Memory Protection (PMP) --
     PMP_NUM_REGIONS              => PMP_NUM_REGIONS,    -- number of regions (0..16)
     PMP_MIN_GRANULARITY          => PMP_MIN_GRANULARITY, -- minimal region granularity in bytes, has to be a power of 2, min 4 bytes